Interaction processing method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of machine learning, in particular to an interaction processing method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining target interaction information of a target object; performing service analysis processing on the target interaction information based on a target interaction model to obtain target service calling information; the target service calling information is used for calling a target service required for processing the target interaction information; the target interaction model is obtained by performing model fine adjustment on a pre-trained language model based on sample interaction information and label service calling information corresponding to the sample interaction information; calling the target service based on the target service calling information to obtain a service calling result; and returning target reply information generated based on the service calling result to the target object. According to the invention, the calling efficiency of the extended service in the interaction process can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine learning, and in particular to an interaction processing method and device, electronic equipment and a storage medium. BACKGROUND

[0002] With the rapid development of the Internet, social networks attract a large number of interactive objects, and various large amounts of interactive objects form large amounts of interactive object relationships in these social networks. For social networks, the relationship chain established by social interaction is one of the core elements of social products, and interaction refers to the process of mutual dependence behavior between individuals and groups on social networks through text or other means to spread information. In order to increase the interaction of social networks and establish more contacts, some network virtual dialogue robots are usually built on social networks now, which participate in various interactions in the community through these virtual dialogue robots to activate the community atmosphere.

[0003] The social network needs the virtual dialogue robot to have more social related play and interaction mechanism, and in the prior art, a large amount of additional development work is required to understand and call the extended services required by the interactive object in the interaction process, which will affect the calling efficiency of the extended services, and thus the calling efficiency of the extended services is low. SUMMARY

[0004] The technical problem to be solved by the present application is to provide an interaction processing method, device, electronic equipment and storage medium, which can improve the calling efficiency of the extended services in the interaction process.

[0005] To solve the above technical problems, on the one hand, the present application provides an interaction processing method, comprising:

[0006] obtaining target interaction information of a target object;

[0007] performing service analysis and processing on the target interaction information based on a target interaction model to obtain target service calling information; the target service calling information is used to call a target service required for processing the target interaction information; the target interaction model is obtained by model fine-tuning of a pre-trained language model based on sample interaction information and label service calling information corresponding to the sample interaction information;

[0008] calling the target service based on the target service calling information to obtain a service calling result;

[0009] returning target reply information generated based on the service calling result to the target object.

[0010] On the other hand, the present application provides an interaction processing device, comprising:

[0011] An interaction information acquisition module is configured to acquire target interaction information of a target object;

[0012] A call analysis module is configured to perform service analysis processing on the target interaction information based on a target interaction model to obtain target service call information, the target service call information being used to call a target service required for processing the target interaction information, and the target interaction model being obtained by performing model fine-tuning on a pre-trained language model based on sample interaction information and label service call information corresponding to the sample interaction information;

[0013] A service call module is configured to call the target service based on the target service call information to obtain a service call result.

[0014] An information return module is configured to return target reply information generated based on the service call result to the target object.

[0015] In another aspect, the present application provides an electronic device including a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the interaction processing method as described above.

[0016] In another aspect, the present application provides a computer storage medium, the storage medium storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by a processor to implement the interaction processing method as described above.

[0017] Implementing the embodiments of the present application has the following beneficial effects:

[0018] The target interaction information of the target object can be analyzed and processed based on the target interaction model in the application to obtain target service calling information, wherein the target interaction model is obtained by fine-tuning a pre-trained language model based on sample interaction information and label service calling information, so that the target interaction model can be analyzed and processed to obtain target service calling information corresponding to the target interaction information when receiving input target interaction information; on the one hand, by inputting the target interaction information into the target interaction model, the corresponding target service calling information can be directly obtained without the need for additional means, thereby improving the generation efficiency and convenience of the service calling information; on the other hand, the target interaction model is obtained by fine-tuning the pre-trained language model, so that the target interaction model can be obtained through light-weight model training, thereby reducing the data processing amount in the target interaction model training process while making the target interaction model applicable to the interaction scene. Then, the corresponding target service is called by the obtained target service calling information to obtain a service calling result, the target reply information is generated based on the service calling result and returned to the target object, so that the interaction process from the target interaction information input by the target object to the target reply information returned to the target object is realized, and the related services required to be called by the target object are determined according to the target interaction information of the target object, so that the social platform has more interactive capabilities, the interactive interest of the target object is stimulated, and the interactive enthusiasm is improved. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the drawings needed in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0020] Figure 1 is an implementation environment schematic diagram provided by the embodiments of the present application;

[0021] Figure 2 is a flowchart of an interaction processing method provided by the embodiments of the present application;

[0022] Figure 3 is another flowchart of an interaction processing method provided by the embodiments of the present application;

[0023] Figure 4 is another flowchart of an interaction processing method provided by the embodiments of the present application;

[0024] Figure 5 is a network structure schematic diagram of a virtual community social interaction method and system based on a large language model provided by the embodiments of the present application;

[0025] Figure 6 is a schematic diagram of an interaction processing device provided by an embodiment of the present application;

[0026] Figure 7 is a schematic diagram of an electronic device structure provided by an embodiment of the present application;

[0027] Figure 8 is another schematic diagram of an electronic device structure provided by an embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to make the objectives, technical solutions and advantages of the present application clearer, the following will further describe the present application with reference to the accompanying drawings. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0029] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product or device.

[0030] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program with a predetermined service, and works together with other related parts to achieve a predetermined target, and can be implemented entirely or partially by using software, hardware (such as a processing circuit or a memory) or a combination thereof. Similarly, one processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of the overall module or unit that includes the service of the module or unit.

[0031] It can be understood that in the specific embodiments of the present application, data related to interactive object information and the like is obtained with the permission or consent of the interactive object when the above embodiments of the present application are applied to specific products or technologies, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0032] First, the relevant terms involved in the embodiments of the present specification are explained as follows:

[0033] Machine Learning: Machine Learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithmic complexity theory, etc. It is a field of study that focuses on how computers can simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge structures to continuously improve their performance.

[0034] Deep Learning: The concept of deep learning originated from the study of artificial neural networks. A multi-layer perceptron with multiple hidden layers is a deep learning structure. Deep learning forms more abstract high-level representation attributes or features by combining low-level features to discover distributed feature representations of data.

[0035] Virtual Community: A group of people who communicate with each other mainly through computer networks, share some degree of knowledge and information, and care for each other to a great extent as if they were friends, thus forming a community. In addition, there is a virtual community composed of social relationships formed by virtual characters such as bionic people in the current artificial intelligence era.

[0036] LLM: Large Language Model (LLM) refers to a computer model that can process and generate natural language; it represents a major breakthrough in the field of artificial intelligence and has the potential to change the field through acquired knowledge. LLM can predict the next word or sentence by learning the statistical rules and semantic information of language data. As the input data set and parameter space continue to expand, the ability of LLM will also improve accordingly. It is used in various application fields such as robotics, machine learning, machine translation, speech recognition, image processing, etc., so it is called multi-modal large language model (MLLM).

[0037] Instruction Tuning: Instruction tuning refers to generating instructions for each task separately, fine-tuning on several full-shot tasks, and then evaluating the generalization ability (zeroshot) on specific tasks, where the pre-training model parameters are unfreezed. It is usually performed on a large set of public NLP task data to stimulate the language model's understanding ability and make the model understand and provide correct feedback by giving more explicit instructions.

[0038] Please refer to Figure 1Fig. 1 shows an implementation environment provided by an embodiment of the present application, which can include at least one electronic terminal 110 and an interaction server 120. The electronic terminal 110 and the interaction server 120 can communicate data through a network.

[0039] In this embodiment, each interaction object can interact information with other interaction objects or the auxiliary virtual object corresponding to other interaction objects based on the corresponding electronic terminal 110; the interaction server 120 can receive the information sent by each interaction object and forward it to the corresponding interaction object; and the auxiliary virtual object corresponding to the interaction object is a virtual object created by the interaction object in the social platform.

[0040] Further, each interaction object in this embodiment can interact information with the auxiliary virtual object corresponding to the interaction object based on the corresponding electronic terminal 110; specifically, the interaction object can input corresponding interaction information based on the electronic terminal 110, the auxiliary virtual object can perform service analysis and processing on the target interaction information, obtain target service calling information, and call corresponding target services based on the target service calling information to obtain service calling results; and the target object returns target reply information generated based on the service calling results to the interaction object to realize information interaction with the interaction object.

[0041] The auxiliary virtual object in this embodiment can include a virtual dialogue robot, an AI intelligent robot, a virtual pet, an electronic pet, and other objects capable of interacting information with the interaction object.

[0042] The electronic terminal 110 can communicate with the interaction server 120 based on a browser / server mode (B / S) or a client / server mode (C / S). The electronic terminal 110 can include a smart phone, a tablet computer, a notebook computer, a digital assistant, a smart wearable device, a vehicle-mounted terminal, and other types of physical devices, and can also include software running in a physical device, such as an application program. The operating system running on the electronic terminal 110 in this embodiment can include, but is not limited to, an Android system, an IOS system, linux, windows, and the like.

[0043] The interaction server 120 and the electronic terminal 110 can establish a communication connection through a wire or wireless. The interaction server 120 can include a server running independently, a distributed server, or a server cluster composed of multiple servers, wherein the server can be a cloud server.

[0044] In order to solve the problem of low efficiency of extended service call in the prior art social network, the embodiment of the present application provides an interaction processing method, the execution subject of the method can be the electronic terminal or the interaction server. Figure 2 The method can specifically include the following steps.

[0045] S210. Obtain target interaction information of a target object.

[0046] The target object in the embodiment can be a real interaction object in a social network, and the target interaction information of the target object can include information of the type of text, image, voice and the like input by the target object; so that the target object can input information of at least one type of the type of text, image, voice and the like in the interaction process, and the target interaction information input by the target object can reflect the interaction demand or the interaction intention of the target object.

[0047] In the case where the execution subject of the interaction processing method in the embodiment is the electronic terminal, the corresponding electronic terminal can directly obtain the target interaction information input by the target object in response to the input operation of the target object; in the case where the execution subject of the interaction processing method in the embodiment is the interaction server, the corresponding electronic terminal can directly obtain the target interaction information input by the target object in response to the input operation of the target object, and the electronic terminal can send the obtained target interaction information to the interaction server, so that the interaction server can obtain the target interaction information of the target object.

[0048] S220. Perform service analysis processing on the target interaction information based on a target interaction model, to obtain target service call information; the target service call information is used to call a target service required for processing the target interaction information; the target interaction model is obtained by model fine-tuning of a pre-trained language model based on sample interaction information and label service call information corresponding to the sample interaction information.

[0049] In a case where the target interaction information of the target object is acquired, the target interaction information can be input into the target interaction model to realize service analysis and processing of the target interaction information based on the target interaction model, and target service calling information is obtained; that is, by directly calling the target interaction model to perform service analysis and processing on the target interaction information, the target service calling information can be obtained. The target interaction model in this embodiment is obtained by model fine-tuning of a pre-trained language model based on sample interaction information and label service calling information corresponding to the sample interaction information. The sample interaction information and the label service calling information corresponding to the sample interaction information are determined in an interaction process in an interaction scenario, that is, the pre-trained language model is fine-tuned in a self-supervised manner, and then the target interaction model applicable to the interaction scenario is obtained. The sample interaction information can include actual interaction information input by an interaction object in a social network, and the corresponding label service calling information can include calling information of a service required to be called by the interaction object after understanding the actual interaction information input by the interaction object; the sample interaction information can also include interaction information constructed for the interaction scenario, and the corresponding label service calling information can include calling information of a service required to be called after understanding the constructed interaction information.

[0050] The target service calling information can be used to call a target service required to process the target interaction information. It can be understood that, generally, when the target object inputs the target interaction information, a reply needs to be made to the target interaction information, which can specifically be a reply to a demand or a question raised in the target interaction information. Here, the target service required to process the target interaction information can be understood as a target service required to be called to make a reply to the target interaction information, that is, a target service required to be called to generate a reply result corresponding to the target interaction information. The target service in this embodiment can be an extended service, that is, a service that is not provided by the social platform and needs to be called from a third party. The target service can include, but is not limited to, extended services such as drawing, singing, writing poems, weather forecasting, voice changing, and the like.

[0051] S230. Based on the target service calling information, the target service is called to obtain a service calling result.

[0052] In a case where the target service calling information is determined, the corresponding target service can be directly called based on the target service calling information, and then a corresponding service calling result is obtained. The target service can be called based on the middle control service, the target service calling information is sent to the middle control service, the middle control service can analyze the target service calling information to obtain a target service that needs to be called and related calling parameters, and then the corresponding target service is directly called, which can improve the directivity of service calling and then improve the accuracy of service calling.

[0053] S240. Return the target reply information generated based on the service call result to the target object.

[0054] In a case where the service call result returned by directly calling the target service can be directly returned as the reply information to the target object, the service call result can be directly determined as the target reply information and returned to the target object; in a case where the service call result cannot be directly returned as the reply information to the target object, the service call result can be further data processed to obtain the target reply information. By returning the target reply information to the target object, the target reply information can be displayed on the electronic terminal corresponding to the target object.

[0055] In one example, the target interaction information of the target object can be "want to listen to song A", and the corresponding target reply information can be the audio information of song A; in another example, the target interaction information of the target object can be "weather in city B", and the corresponding target reply information can be the recent weather forecast of city B.

[0056] In the present application, the target interaction information of the target object can be service analyzed and processed based on the target interaction model to obtain the target service call information, wherein the target interaction model is obtained by model fine-tuning a pre-trained language model based on sample interaction information and label service call information, so that the target interaction model can perform service analysis and processing when receiving the input target interaction information, to obtain the target service call information corresponding to the target interaction information; on the one hand, by inputting the target interaction information into the target interaction model, the corresponding target service call information can be directly obtained without the need for other additional means, improving the generation efficiency and convenience of the service call information; on the other hand, the target interaction model is obtained by model fine-tuning a pre-trained language model, so that the target interaction model can be obtained through light-weight model training, thereby reducing the data processing amount in the training process of the target interaction model while making the target interaction model applicable to the interaction scenario. Then, by calling the corresponding target service through the obtained target service call information, the service call result is obtained, the target reply information is generated based on the service call result and returned to the target object, thereby realizing the interaction process from the target interaction information input by the target object to the target reply information returned to the target object, and determining the related service to be called by the target object according to the target interaction information of the target object, which can enable the social platform to have more interactive capabilities, stimulate the interactive interest of the target object, and improve the interactive enthusiasm.

[0057] The interaction process of inputting the target interaction information from the target object to the target reply information returned to the target object can be a process of dialogue interaction between the target object and the auxiliary virtual object corresponding to the target object, that is, the target object sends target interaction information to the auxiliary virtual object, and the auxiliary virtual object corresponding to the target object returns corresponding target reply information based on the target interaction information.

[0058] Further, the target interaction model in the embodiment can specifically include an intention recognition submodule and a service analysis submodule. Correspondingly, please refer to Figure 3 which shows another interaction processing method. The execution subject of the method can be the electronic terminal or the interaction server. The method can specifically include:

[0059] S210. Obtain target interaction information of a target object.

[0060] S220. Perform service analysis processing on the target interaction information based on a target interaction model to obtain target service calling information. The target service calling information is used to call a target service required for processing the target interaction information. The target interaction model is obtained by model fine-tuning a pre-trained language model based on sample interaction information and label service calling information corresponding to the sample interaction information.

[0061] Specifically, step S220. Perform service analysis processing on the target interaction information based on a target interaction model to obtain target service calling information can include:

[0062] S310. Perform intention recognition on the target interaction information based on the intention recognition submodule to obtain a target intention corresponding to the target interaction information.

[0063] The target object can express its intention through the target interaction information, so that by inputting the target interaction information into the intention recognition submodule, the intention recognition submodule can recognize the intention of the target interaction information to obtain the target intention contained in the target interaction information. In the embodiment, the intention recognition submodule can be obtained by model fine-tuning a pre-trained language model based on sample interaction information and label intention corresponding to the sample interaction information.

[0064] S320. Perform service analysis processing on the target intention based on the service analysis submodule to obtain the target service calling information.

[0065] The target intention can include a requirement of calling a related extended service. The target service calling information can be obtained by performing service analysis processing on the target intention by using the service analysis sub-model, and the target service calling information can be used to call a target service required for processing the target interactive information. The target intention can not include a requirement of calling a related extended service. The target service calling information can be obtained by performing service analysis processing on the target intention by using the service analysis sub-model, and the target service calling information can be empty, indicating that no extended service needs to be called to reply to the target interactive information.

[0066] For example, the target interactive information can be "good morning", the target intention output by the corresponding intention recognition sub-model can be "greeting", the target intention does not need to call an extended service, and the target service calling information output by the corresponding service analysis sub-model can be empty. For another example, the target interactive information of the target object can be "want to listen to song A", the target intention output by the corresponding intention recognition sub-model can be "listen to song A", the target intention needs to call an extended service related to the song, and the target service calling information output by the corresponding service analysis sub-model can indicate calling the extended service related to the song and specifically returning the audio of song A.

[0067] S230. Calling the target service based on the target service calling information to obtain a service calling result.

[0068] S240. Returning target reply information generated based on the service calling result to the target object.

[0069] The specific implementation details of steps S210, S230 and S240 can be referred to the above description of the embodiment, and will not be described here.

[0070] In this embodiment, the target interactive model can specifically include an intention recognition sub-model and a service analysis sub-model. The intention recognition sub-model can be used to perform intention recognition on the target interactive information to obtain a corresponding target intention. The service analysis sub-model can be used to perform service analysis processing on the target intention to obtain corresponding target service calling information. Therefore, service calling analysis can be performed on the basis of recognizing the target intention. Since the target intention can accurately, briefly and clearly reflect the interactive requirement of the target object, service calling analysis can be performed based on the target intention, without involving the original target interactive information of the target object and processing more data, thereby improving the accuracy and efficiency of service calling analysis.

[0071] In this embodiment, for the target interactive information of the target object, target service calling information corresponding to one extended service or target service calling information corresponding to multiple extended services can be obtained by performing service analysis processing on the target interactive information by using the target interactive model. Correspondingly, please refer to Figure 4The method shown in the embodiment 2 illustrates another interaction processing method, and an execution subject of the method can be the electronic terminal or the interaction server; and the method can specifically include the following steps.

[0072] S210. Obtain target interaction information of a target object.

[0073] S220. Perform service analysis processing on the target interaction information based on a target interaction model to obtain target service calling information; the target service calling information is used to call a target service required for processing the target interaction information; and the target interaction model is obtained by performing model fine-tuning on a pre-trained language model based on sample interaction information and label service calling information corresponding to the sample interaction information.

[0074] Specifically, the step S220 of performing service analysis processing on the target interaction information based on the target interaction model to obtain the target service calling information can include the following steps.

[0075] S410. Perform service analysis processing on the target interaction information based on the target interaction model to obtain calling information corresponding to each of a plurality of services.

[0076] In the embodiment, in order to achieve a corresponding interaction result in an interaction process, a plurality of services can be called, and accordingly, the service analysis processing on the target interaction information can obtain calling information corresponding to each of the plurality of services; the calling information corresponding to each of the plurality of services is scattered, and the calling information corresponding to each of the plurality of services can be combined to obtain target service calling information corresponding to the interaction process. It should be noted that the plurality of services corresponding to the interaction process can have a dependency relationship or can not have a dependency relationship, which can be determined according to the target interaction information input by the target object.

[0077] In a case where the target interaction information of the target object is acquired, the target interaction information can be input into the target interaction model to realize service analysis and processing of the target interaction information based on the target interaction model, and target service calling information is obtained; that is, by directly calling the target interaction model to perform service analysis and processing on the target interaction information, the target service calling information can be obtained. The target interaction model in this embodiment is obtained by model fine-tuning of a pre-trained language model based on sample interaction information and label service calling information corresponding to the sample interaction information. The sample interaction information and the label service calling information corresponding to the sample interaction information are determined in an interaction process in an interaction scenario; that is, the pre-trained language model is fine-tuned in a self-supervised manner, and then the target interaction model applicable to the interaction scenario is obtained. Correspondingly, in a case where the sample interaction information represents that multiple services need to be called in the current interaction process and there is no dependency relationship between the multiple services, the respective calling information of each service included in the label service calling information corresponding to the sample interaction information has no dependency relationship; in a case where the sample interaction information represents that multiple services need to be called in the current interaction process and there is a dependency relationship between the multiple services, the respective calling information of each service included in the label service calling information corresponding to the sample interaction information has a dependency relationship.

[0078] S420. Information combination is performed on the respective calling information of the multiple services to obtain the target service calling information.

[0079] It should be noted that, in one example, the step of combining the respective calling information of the multiple services to obtain the target service calling information can be realized based on the target interaction model; that is, by inputting the target interaction information of the target object into the target interaction model, the target interaction model can output the target service calling information corresponding to the target interaction information, that is, the target interaction model has the function of performing information combination on the respective calling information of the multiple services; in another example, the step of combining the respective calling information of the multiple services to obtain the target service calling information can be realized based on an execution subject of the interaction processing method in this embodiment; that is, based on an electronic terminal or an interaction server; that is, by inputting the target interaction information of the target object into the target interaction model, the target interaction model can output the respective calling information of the multiple services, and the electronic terminal or the interaction server can combine the respective calling information of the multiple services to obtain the target service calling information.

[0080] S230. The target service is called based on the target service calling information, and a service calling result is obtained.

[0081] S240. Target reply information generated based on the service calling result is returned to the target object.

[0082] The specific implementation details of steps S210, S230, and S240 can be referred to the foregoing content of the embodiment, and will not be described here again.

[0083] In the embodiment, in a process of one interaction, a plurality of extended services can be called to achieve corresponding interaction results, that is, a plurality of extended services are called to achieve corresponding interaction results, so that the social platform has more extended capabilities, obtains rich and diverse interaction results, enhances the attraction of the social platform to the interaction object, and improves the interaction enthusiasm of the interaction object.

[0084] In the case where there is a dependency relationship between the plurality of services, the information combination of the calling information corresponding to each of the plurality of services obtains the target service calling information, including:

[0085] Based on the calling sequence of the plurality of services, the calling information corresponding to each of the plurality of services is arranged to obtain a calling sequence; the calling sequence is determined by dependency relationship analysis on the plurality of services;

[0086] The target service calling information is determined based on the calling sequence.

[0087] In one example, the step of combining the calling information corresponding to each of the plurality of services to obtain the target service calling information can be implemented based on a target interaction model. In the case where the sample interaction information represents that a plurality of services need to be called in the process of this interaction and there is a dependency relationship between the plurality of services, the calling information corresponding to each of the services contained in the label service calling information corresponding to the corresponding sample interaction information has a dependency relationship, that is, the label service calling information is obtained by arranging the calling information corresponding to each of the plurality of services based on the calling sequence of the plurality of services. Therefore, in the case where the target interaction information represents that a plurality of services need to be called in the process of interaction and there is a dependency relationship between the plurality of services, the target service calling information containing the dependency relationship between the services can be directly obtained by directly calling the target interaction model for service analysis and processing.

[0088] In another example, the step of combining the calling information corresponding to each of the plurality of services to obtain the target service calling information can be implemented based on the execution subject of the interaction processing method in the embodiment, that is, implemented based on the electronic terminal or the interaction server, that is, by inputting the target interaction information of the target object into the target interaction model, the target interaction model can output the calling information corresponding to each of the plurality of services, and the electronic terminal or the interaction server can combine the calling information corresponding to each of the plurality of services to obtain the target service calling information; specifically, the electronic terminal or the interaction server can analyze the dependency relationship of the plurality of services to obtain the dependency relationship between the plurality of services; and then determine the calling order of the plurality of services based on the dependency relationship between the plurality of services, the calling order of the service being depended on is prior to the calling order of the service depending on other services; for example, service 1 depends on the calling result of service 2, and accordingly service 2 is called first and then service 1, and the calling order is service 1→service 2.

[0089] For example, the target interaction information can be "displaying the weather of B city in a table form", and accordingly the weather forecast service and the table display service can be called. The general calling order can be to call the weather forecast service first to obtain the weather forecast information of B city, and then call the table display service to display the weather forecast information of B city in a table form; in this example, the table display service depends on the weather forecast service, and accordingly the calling order is weather forecast service→table display service, that is, the result of the previous step can be used as the input of the next step.

[0090] In the embodiment, when the target interaction information represents a case that a plurality of services need to be called and there is a dependency relationship between the plurality of services, the plurality of services can be called in turn based on the calling order determined based on the dependency relationship between the services, the calling order determined based on the dependency relationship between the services can ensure the integrity and rationality of the service calling chain, and accordingly the target service calling information can be obtained by arranging the calling information corresponding to each of the plurality of services based on the determined calling order, which can ensure the efficiency of the plurality of extended service calls and the accuracy of the service calling result in the interaction process.

[0091] When the service is called, the service parameter can be used; accordingly, when the calling information corresponding to the service is generated, the corresponding service parameter can be included in the calling information; specifically, the service analysis and processing of the target interaction information based on the target interaction model to obtain the calling information corresponding to each of the plurality of services comprises:

[0092] The service analysis and processing of the target interaction information based on the target interaction model to obtain the service identifier of each service and the calling parameter of each service; the calling parameter of each service is used to indicate the calling object when each service is called.

[0093] Based on the service identification of each service and the calling parameter of each service, calling information corresponding to each of the plurality of services is generated.

[0094] At least one calling object can correspond to each service, and the calling party can call according to actual conditions. The calling parameter of each service is used to indicate the calling object. The target interaction model can perform service analysis and processing on the target interaction information to obtain the service identification of each service and the calling parameter of each service. Specifically, the target interaction information can carry information representing the service identification and the calling parameter. For example, the target interaction information can be "want to listen to song A", the information representing the service identification is "song", the corresponding service identification can be "sing", the information representing the calling parameter is the song name "A", and the corresponding calling parameter is "A". For another example, the target interaction information can be "weather in city B", the information representing the service identification is "weather", the corresponding service identification can be "weather", and the information representing the calling parameter is the city name "B", and the corresponding calling parameter is "B".

[0095] Further, after the target interaction model performs intent recognition on the target interaction information, the target intent corresponding to the target interaction information can be obtained, and the target intent can also be directly analyzed and processed to obtain the service identification of each service and the calling parameter of each service. Further, the target interaction model can also analyze and process based on the target interaction information and the target intent to obtain the service identification of each service and the calling parameter of each service, so as to avoid information omission.

[0096] For the service identification of each service and the calling parameter of each service, corresponding calling information can be formed, which can be in the form of a calling tuple, for example, calling tuple c=(ac,ic), where ac is the service identification, ic is the calling parameter, and c is the calling result.

[0097] In this embodiment, for the calling of each service, the service identification and the calling parameter can be included in the corresponding calling information to clearly indicate the service target to be called, so as to realize accurate expression of the calling information, and further improve the accuracy and efficiency of service calling.

[0098] According to the above content of the embodiment, based on the sample interaction information and the label service call information corresponding to the sample interaction information, the pre-trained language model is fine-tuned to obtain a target interaction model with a service analysis function. Further, on this basis, a target interaction model with a target interaction style can also be generated, that is, the target interaction model has both a service analysis function and a function of interacting in a target interaction style. Specifically, the embodiment also provides a fine-tuning method of a target interaction model with a target interaction style, which can include:

[0099] Obtaining sample corpus information of a target interaction style; the sample corpus information includes object corpus information of a sample object and label corpus information; the label corpus information is reply information with the target interaction style and corresponding to the object corpus information;

[0100] Fine-tuning the pre-trained language model based on the object corpus information and the label corpus information to obtain a target interaction model with the target interaction style.

[0101] The target interaction style in the embodiment can include styles such as seriousness, liveliness, loveliness, and randomness. The interaction style can also be understood as an object personality or a conversation style embodied through interaction. In the embodiment, the interaction style can be customized. When the target interaction style is determined, the pre-trained language model can be fine-tuned based on the sample corpus information of the target interaction style to obtain a target interaction model with the target interaction style.

[0102] The sample corpus information can include object corpus information of a sample object and corresponding label corpus information. The object corpus information of the sample object can be information generated by an interaction object in an actual interaction process or constructed corpus information. The label corpus information is reply information for the object corpus information. For different interaction styles, the object corpus information of the sample object is the same, and the label corpus information of different interaction styles is different. For example, for a target interaction style of seriousness, the corresponding label corpus information is serious reply information. For a target interaction style of loveliness, the corresponding label corpus information is lovable reply information.

[0103] Through the above fine-tuning method, a target interaction model with both a service analysis function and a function of interacting in a target interaction style can be obtained, the applicability of the target interaction model in multiple scenarios is improved, and the universality of the target interaction model is improved. Moreover, an interaction object can select a target interaction model with a corresponding interaction style according to its own interaction requirements, the flexibility of interaction is improved, and the interest and enthusiasm of the interaction object are improved.

[0104] Further, the target interaction model can also be concretized, i.e., carried by a virtual object; specifically, the interaction processing method of the embodiment can include:

[0105] In response to a virtual object creation request, an initial virtual object is created in the social platform;

[0106] A target interaction model with the target interaction style is embedded into the initial virtual object, obtaining an auxiliary virtual object corresponding to the target object; the auxiliary virtual object is used to receive messages for the target object and send messages with the target interaction style in the social platform under authorization of the target object.

[0107] The embodiment can also provide external materials for maintaining a human exhibition, including 2D / 3D virtual object image materials, personalized signatures, pet tags, dynamic settings, etc. The initial virtual object in the embodiment can be created from scratch by external materials to meet the requirements, or can be selected from a plurality of preset virtual objects to meet the requirements. In the case of creating an initial virtual object, the target interaction model with the target interaction style can be embedded into the initial virtual object to obtain an auxiliary virtual object corresponding to the target object. The auxiliary virtual object in the embodiment can be regarded as an object of the target object, which can have the same or similar interaction style as the target object, or can have the interaction style set by the target object, which is not limited here.

[0108] In the case of creating an auxiliary virtual object, the auxiliary virtual object has both service analysis function and interaction function with the target interaction style; the auxiliary virtual object of the target object can interact with the corresponding target object, or other interaction objects on the social platform, or the auxiliary virtual objects corresponding to other interaction objects.

[0109] Thus, in the social platform, the auxiliary virtual object has both service analysis function and interaction function with the target interaction style, thereby improving the interest and attractiveness of interaction; through the assistance of the auxiliary virtual object, the information interaction between interaction objects and between interaction objects and auxiliary virtual objects can be increased. In the case that the target object is offline, the auxiliary virtual object of the target object can replace the target object to receive and reply information, as a split of the target object, increase the online time of the target object, and improve the activity of the interaction objects on the social platform.

[0110] In this embodiment, each interactive object has a corresponding auxiliary virtual object, which can be displayed in the form of a cartoon image in the virtual community. Here, the image of a tiger is used for display. The target object corresponding to the auxiliary virtual object can be regarded as the master of the auxiliary virtual object, i.e., the master of the tiger. The auxiliary virtual object can be a virtual pet of the target object. The tiger can communicate and transmit information with the master and the auxiliary virtual objects of other interactive objects, help to activate the interactive atmosphere of the entire virtual community, and can be regarded as a virtual avatar and assistant of the master. The auxiliary virtual object can also have its own personality and make unexpected things (such as the joy of raising pets), call third-party capabilities, and derive various play methods (such as singing, etc.). Its messages are displayed on the top of the image, and the auxiliary virtual object can chat with the master in the message list, and the auxiliary virtual object can be displayed in the message list as an ordinary friend, and the like. Correspondingly, the above-mentioned interaction method can further include:

[0111] displaying the auxiliary virtual objects corresponding to the plurality of interactive objects, respectively;

[0112] in response to a selection operation based on any auxiliary virtual object, establishing an interaction link between the target object and the any auxiliary virtual object, or establishing an interaction link between the auxiliary virtual object corresponding to the target object and the any auxiliary virtual object; the any auxiliary virtual object is any auxiliary virtual object of the auxiliary virtual objects corresponding to the plurality of interactive objects, respectively.

[0113] The auxiliary virtual objects of different interactive objects can communicate and interact in a community, thereby forming a community square. The social relationship chain can be further strengthened by selecting and communicating with the auxiliary virtual object according to the speech of the different auxiliary virtual objects, establishing a social relationship between the auxiliary virtual object and the master behind the auxiliary virtual object, and the like.

[0114] Figure 5The network structure diagram of the virtual community social interaction method and system based on a large language model is shown, the whole structure includes a lightweight social terminal and a community virtual image cloud system, and the core is a virtual image backend processing service group system. By increasing virtual pets as interactive objects to assist the main body of social interaction, pets can talk to each other, build a pet community, and increase the active atmosphere of the community. In the case that the owner is not online, the owner authorizes the auxiliary virtual object pet to reply and replace the main body to collect information, as a split image of the interactive object, increases the online time of the interactive object, creates more opportunities for dialogue, and at the same time, in order to let the pet have more expansion capabilities consistent with its own identity; in a self-supervised manner, the calls of the pet to various third-party external capabilities are uniformly processed, the LLM is fine-tuned through the language model to learn how to use different third-party tools through API calls, supports more interesting play calls and parallel expansion, and finally the LLM completes multiple functions through one call, greatly improves the interaction effect, and the main components of the whole virtual community social interaction system and the functions of each part are introduced in detail as follows:

[0115] (1) Appearance customization model of auxiliary virtual object: the external material maintained by the main interactive object (establishing a good social image) to show, including auxiliary virtual object pet 2D / 3D image material, personalized signature, pet label, dynamic setting, etc. Here, a batch of auxiliary virtual object information with different appearances can be preset, and the interactive object can claim it, which includes material request download and update processing module, and pet material image storage library;

[0116] (2) Personality customization: mainly through the multi-person preset template library (the typical implementation is to design a batch of personality or label words such as serious and lively according to the platform needs in advance) and the person setup maintenance module to build a question and answer model and service based on the LLM large language model, and then through the large language model to identify the intention of the interactive object dialogue, and reply the content corresponding to the virtual image personality. The content of personality customization here is mainly used to provide the corresponding auxiliary virtual object person setup feature input corpus for the language model, which will affect the dialogue style of the auxiliary virtual object and the interactive object. Personality definition also includes the following information: for example, a social design active auxiliary virtual object will go out to talk to other auxiliary virtual objects every day, increasing the frequency of initiating dialogue. Through the person setup information, a plurality of groups of multi-round dialogue corpus (saved in the virtual pet person setup and multi-round dialogue fine-tuning corpus library in the above figure) are constructed, and then the LLM language model is fine-tuned to obtain the final model. The specific debugging method adopts the LoRA method (LoRA, which stands for Low-Rank Adaptation of Large Language Models, which means low-order adaptation of large language models. Specifically: LoRA freezes the original parameters of the large language model, and adds a Dropout+Linear+Conv convolutional layer as an additional parameter in each layer of the Decoder to obtain the LLM core dialogue model. The whole process is lightweight, and the dialogue results of the virtual pet that meet the virtual pet personality style and the correct question and answer are generated.

[0117] (3) The social relationship module is the core module. The ultimate goal of building a virtual pet is to increase social interaction between the interactive object and the auxiliary virtual object and between the interactive object and the auxiliary virtual object in the virtual community with the help of the virtual pet, to handle social relationships, achieve relationship matching, relationship management, relationship chain sedimentation, and relationship iteration module capabilities. The specific functions of the module are as follows:

[0118] (a) Relationship matching (goal - faster and more accurate matching to the corresponding interactive object or auxiliary virtual object): here, the interactive object's interest points, tags, interactive object target person setup, and selected interactive object region are used to recall the "social distance" candidates, provide more accurate display to help the interactive object filter, help the interactive object match high-quality friends, and greatly improve the interactive object's social range, etc.; person connection community (goal two is to match the corresponding network group): combined with the interactive object's interests and social relationship chain, through intelligent recommendation + intelligent search, using the interactive object's interests and network group classification and tag data, quickly matching groups and channels, and providing unique social emotional value;

[0119] (b) Relationship management and relationship precipitation: The typical functions of the relationship include efficient relationship management: corresponding to the first matching and the establishment of various social relationships, providing intelligent grouping (such as friends, classmates, colleagues, neighbors, and interest groups), providing closeness scores, relationship activity trend reports, etc.; At the same time, it provides core relationship maintenance: such as important event reminders (birthdays or special anniversaries of close friends, etc.), improving the social quality of the core relationship of the interaction object; Relationship degree: the core purpose is to establish long-term memory between the interaction objects, including helping the interaction objects to establish long-term memory, automatically organizing important messages / events / labels / impressions of the two people, and assisting the interaction objects in social development relationship upgrade, and continuously improving the relationship of the power tool, such as dialogue assistance / personalized event prompts, etc.

[0120] (c) Relationship iteration: Specifically, the above interaction method can further include:

[0121] In the case of authorization of the target object, interaction information is sent to a friend object of the target object based on an auxiliary virtual object corresponding to the target object.

[0122] Here, it mainly includes pan-relationship maintenance: such as maintaining pan-relationships at low cost through auxiliary virtual objects or authorization of the interaction object, auxiliary virtual objects initiating greetings to the friends of the owner in the name of the owner, and maintaining social activity; It also includes invalid relationship removal: periodically batch statistics to give invalid relationship reminders, whether to optimize contact or eliminate deletion, etc.

[0123] (4) Social play: Typical examples of light interesting social functions and social play are as follows, and the ability of the auxiliary virtual object is expanded through a large language model to identify the intention of the interaction object, and a third-party service is called to implement the auxiliary virtual object, such as painting, singing, and writing poems, etc. More skills, increase the attraction to the interaction object, and create more social interactions. Other examples of light social play:

[0124] Scenario one: Auxiliary virtual object teasing:

[0125] (a) With a clear goal: Specifically, the above interaction method can further include:

[0126] In response to the object specifying operation, in the case of authorization of the target object, interaction information is sent to a specified object or an auxiliary virtual object corresponding to the specified object based on an auxiliary virtual object corresponding to the target object.

[0127] In the case of specifying an object, the target object's auxiliary virtual object can initiate a greeting with the specified object or the specified object's auxiliary virtual object: generate some greetings and opening information through the LLM language model, such as "My mother asked me to pass you a message, XXX"; or start with the latest hot topic and consultation, such as "Have you seen the TV series XXX? Do you know about the recent hot topic? This game is said to be very fun, etc.", which can be displayed in the space where the specified object's auxiliary virtual object is located.

[0128] (b) No specific target: Specifically, the above-mentioned interaction method can further include:

[0129] Generating an interaction request based on the object information of the target object; the interaction request indicates the object information of the to-be-interacted object;

[0130] In response to the interaction request, determining a matching object that meets the interaction request, and generating to-be-sent interaction information;

[0131] Sending the to-be-sent interaction information to the matching object.

[0132] For example, the language model generates an interaction request based on the object information of the target object: "I want to make friends with an 18-year-old Scorpio", at this time, search can be called to find the corresponding tagged interaction object, and then the corresponding greeting dialogue is generated through the language model to increase the interaction with the interaction object;

[0133] Scenario two: content information brought back by the auxiliary virtual object: Specifically, the above-mentioned interaction method can further include:

[0134] In the case of displaying interaction information between the auxiliary virtual objects corresponding to the plurality of interaction objects, in response to a comment operation triggered by at least one interaction object, display the comment information of the at least one interaction object; the at least one interaction object is at least one of the plurality of interaction objects;

[0135] And / or,

[0136] In the case of displaying interaction information between the auxiliary virtual objects corresponding to the plurality of interaction objects, in response to a selection operation of a first interaction object on a second interaction object, establishing an interaction link between the first interaction object and the second interaction object; the first interaction object and the second interaction object are different interaction objects in the plurality of interaction objects.

[0137] For example, a message showing an auxiliary virtual object and the interaction of the auxiliary virtual object, both owners can leave comments below, at this time if the owner is interested in the owner of the other auxiliary virtual object, can initiate a friend or start chatting. This is very similar to the way two pet owners in real life socialize through their pets as a medium, but the large language model simulates the corresponding social light interaction scene in the community, promotes activity, and develops more social relationships.

[0138] The interaction processing method of the present application can use the dialogue interaction system of the auxiliary virtual object pet to recognize the owner's instructions and make corresponding feedback and replies, and various function extensions can be completed and implemented through external input and output, so as to enable the user and the auxiliary virtual object to realize more advanced emotional communication. The large language model gives the auxiliary virtual object the ability of emotional dialogue and communication, completes the specific character and the ability of mutual communication in the specific community, becomes the owner's assistant, can be used as a topic and social medium for communication between users, increases the interaction and socialization between users, thereby improving the user experience and the social value of the platform, promoting user communication and socialization, which has very important significance; It can increase the user stickiness of the community, users can make new friends and keep in touch with old friends on social platforms through virtual community images, increase social fun, and auxiliary virtual objects can help users establish emotional links through interaction, increase user dependence and stickiness on social networking platforms, thereby improving user retention rate; Introducing social networks through auxiliary virtual objects increases more social fun and new social channels. Virtual robots have many specific carriers and forms, such as virtual pet images, whose personality and characteristics are set by the owner, which can stimulate user interest and participation, increase user online time on social networking platforms, and thereby increase platform activity.

[0139] The main functions of each service module of the virtual community social interaction method and system based on the large language model are described as follows:

[0140] I. End

[0141] (1) Through communication with the message and content business access server, complete the uplink and downlink of the message function, in addition, the content producer of PGC or UGC, MCN or PUGC, through the mobile end or back-end interface API system, provides local or shot video, which is the main source of distributed content, and can also be considered as a broad sense of end;

[0142] (2) The carrier of the function of various scenes in the content and social business ecology, such as friend search or group search on the mobile end, adding strangers, friend point-to-point chat, friend group chat, live streaming and anchor exchange, channel post release, space release, etc.

[0143] (3) When publishing content, the address of the upload server interface is usually obtained first, and then the local file is uploaded. During the shooting process, the local text and image content can be selected to match the music, filter template, and filter beautification function, etc.;

[0144] (4) Communicate with the reporting and analysis interface server, collect detailed object data and feedback data in each sub-business scenario under the social network scenario, save the collected data in the statistical analysis database as the basic data source of the analysis platform, and use it as an important basis for measuring the service effect of the community interaction robot;

[0145] II. Message and content business access server

[0146] (1) Synchronize with the terminal to complete the uplink and downlink communication of social network messages and synchronization;

[0147] (2) Connect the message content through the message queue system and the message content database storage and indexing system to complete the message storage processing logic;

[0148] (3) Directly communicate with the content production end, and the content submitted from the front end, usually the title of the content, the publisher, the abstract, the cover picture, the release time, or the video shot directly through the server into the server, and store the file in the message and content database;

[0149] (4) Write the meta information of the video content, such as file size, cover picture link, code rate, file format, title, release time, author, etc. into the message and content database;

[0150] III. Message and content database

[0151] (1) Temporarily save the messages of user conversations to realize message roaming and multi-terminal message synchronization, such as point-to-point messages and group messages;

[0152] (2) As the core module of the message system, it optimizes the storage and indexing of messages with high efficiency, and is the information source of message multi-terminal synchronization;

[0153] (3) The core database of the content, the meta information of all content published by the producer is saved in this business database, the focus is on the meta information of the content itself, such as file size, cover picture link, code rate, file format, title, release time, author, whether original or first published, and also includes the classification of the content in the manual review process;

[0154] (4) When the uplink and downlink content interface service receives the video file, it performs standard transcoding operation on the content. After transcoding, the meta information is returned asynchronously, mainly the file size, code rate, specification, and these information will be saved in the content database;

[0155] IV. Message system

[0156] (1) Responsible for the whole flow of social message synchronization and communication scheduling distribution, including point-to-point messages and group messages, and various interactions and exchanges with community robots here and community robots. The dialogue process is realized through messages as a medium;

[0157] (2) Responsible for communication with the message content database, complete message distribution and processing;

[0158] V. Reporting and analysis interface service

[0159] (1) Communicate with the terminal, receive reported message consumption metadata and various feedback during distribution, such as reports and feedback on content distribution quality, satisfaction scores for community service robot dialogue results, and various third-party service call success rates, satisfaction rates for different interactive services, etc.

[0160] (2) The terminal reports data in different business scenarios, stores it in different storage engines after real-time data cleaning, and constructs the data and feedback information required for upper-level basic model training in combination with different business scenario content flows;

[0161] VI. Statistical analysis database

[0162] (1) Report and analysis interface service, save message content after desensitization processing and preliminary processing of data cleaning and verification for different business scenarios;

[0163] (2) Statistics of various third-party service call success rates, etc. For third-party services with successful call rates, more self-supervised fine-tuning sample data for calling services can be constructed to learn how to call large language models, increase the success rate of adaptation, and provide more social interaction function module expansion;

[0164] VII. User information

[0165] (1) This mainly saves the user's basic information and results, with many dimensions, usually generated by a dedicated algorithm team;

[0166] (2) As a source of basic information data for building virtual community interactive dialogue processing models, it can better meet the personalized needs of users;

[0167] VIII. Virtual community social interaction dialogue processing model and service

[0168] (1) According to the detailed method described above, the natural language processing ability and rich knowledge of the large language model are used to build the core brain of the virtual community pet, fully activate the potential natural language processing ability and user intention understanding ability of the large language model, and build virtual pets that meet the user's identity and role setting at the same time. The dialogue between these pet roles has consistency and stability, and in-depth intention understanding is called in the virtual pet dialogue sequence, such as various social play and multi-round dialogue games;

[0169] (2) The above model is completed as a service;

[0170] Nine. Platform system business services

[0171] (1) Generally refers to the operation system of the platform, the content recommendation system, the Push push system, the friend recommendation, the group business, the expression, the search, the channel business, etc.;

[0172] Ten. Community role setting database and fine-tuning corpus

[0173] (1) There are two sources of data, one is to build related tasks and corresponding task reply expected result data as fine-tuning sample library based on the current dialogue business scene, and the other is to filter the top user expected result pairs as further fine-tuning and alignment sample data based on the actual online user feedback and conversion click results. The former is the main part;

[0174] Eleven. Statistical reporting interface and analysis service

[0175] (1) Collect various content quality problems actively fed back and reported by the consumer end users, and also include the feedback and reporting of various interactive behaviors of the virtual community social interaction dialogue robot system generated results;

[0176] (2) The reported results are saved in the statistical analysis database and the instruction fine-tuning sample database after cleaning, which is used to evaluate and measure the performance of the service dialogue robot itself and whether it achieves the expected effect, and guide the subsequent improvement direction;

[0177] Twelve. Large language pre-training model

[0178] (1) Here it is not limited to a fixed large language model, as long as the model using a large amount of rich Internet corpus to construct a large-scale generative Transform architecture can be classified into this category.

[0179] It should be noted that any method described above in this embodiment can be combined based on actual implementation, and has the corresponding beneficial effects.

[0180] The embodiment also provides an interaction processing apparatus, which refers to Figure 6 The apparatus can include:

[0181] An interaction information acquisition module 610, configured to acquire target interaction information of a target object;

[0182] A call analysis module 620, configured to perform service analysis processing on the target interaction information based on a target interaction model to obtain target service call information; the target service call information is used to call a target service required for processing the target interaction information; the target interaction model is obtained by performing model fine-tuning on a pre-trained language model based on sample interaction information and label service call information corresponding to the sample interaction information;

[0183] A service call module 630, configured to call the target service based on the target service call information to obtain a service call result;

[0184] An information return module 640, configured to return target reply information generated based on the service call result to the target object.

[0185] Further, the target interaction model includes an intent recognition submodel and a service analysis submodel.

[0186] The call analysis module includes:

[0187] An intent recognition module, configured to perform intent recognition on the target interaction information based on the intent recognition submodel to obtain a target intent corresponding to the target interaction information;

[0188] A first analysis module, configured to perform service analysis processing on the target intent based on the service analysis submodel to obtain the target service call information.

[0189] Further, the call analysis module includes:

[0190] A second analysis module, configured to perform service analysis processing on the target interaction information based on the target interaction model to obtain calling information corresponding to each of a plurality of services;

[0191] An information combination module, configured to combine the calling information corresponding to each of the plurality of services to obtain the target service call information.

[0192] Further, the information combination module includes:

[0193] A call sequence determination module, configured to arrange the calling information corresponding to each of the plurality of services based on a calling order of the plurality of services to obtain a calling sequence; the calling order is determined by performing dependency relationship analysis on the plurality of services.

[0194] The calling information determination module is configured to determine the target service calling information based on the calling sequence.

[0195] Further, the second analysis module comprises:

[0196] The third analysis module is configured to perform service analysis processing on the target interaction information based on the target interaction model, to obtain service identification of each service and calling parameters of the each service; the calling parameters of the each service are used to indicate calling objects in the case of calling the each service.

[0197] The calling information generation module is configured to generate calling information corresponding to the plurality of services respectively based on the service identification of the each service and the calling parameters of the each service.

[0198] Further, the apparatus further comprises:

[0199] The corpus information acquisition module is configured to acquire sample corpus information of a target interaction style; the sample corpus information comprises object corpus information of a sample object and label corpus information; the label corpus information is reply information having the target interaction style and corresponding to the object corpus information.

[0200] The model fine-tuning module is configured to fine-tune the pre-trained language model based on the object corpus information and the label corpus information, to obtain a target interaction model having the target interaction style.

[0201] Further, the apparatus further comprises:

[0202] The virtual object creation module is configured to create an initial virtual object in a social platform in response to a virtual object creation request.

[0203] The embedding module is configured to embed the target interaction model having the target interaction style into the initial virtual object, to obtain an auxiliary virtual object corresponding to the target object; the auxiliary virtual object is used to receive messages and send messages of the target interaction style for the target object in the social platform in the case of authorization of the target object.

[0204] Further, the apparatus further comprises:

[0205] The first display module is configured to display auxiliary virtual objects corresponding to a plurality of interaction objects respectively.

[0206] The first interaction module is configured to, in response to a selection operation based on any of the auxiliary virtual objects, establish an interaction link between the target object and the any of the auxiliary virtual objects, or establish an interaction link between the auxiliary virtual object corresponding to the target object and the any of the auxiliary virtual objects; the any of the auxiliary virtual objects is any of the auxiliary virtual objects corresponding to the plurality of interaction objects.

[0207] Further, the apparatus further comprises:

[0208] The second interaction module is configured to, in the case that the target object authorizes, send interaction information to a friend object of the target object based on the auxiliary virtual object corresponding to the target object.

[0209] Further, the apparatus further comprises:

[0210] The third interaction module is configured to, in the case that the target object authorizes, send interaction information to a specified object or the auxiliary virtual object corresponding to the specified object in response to an object designation operation based on the auxiliary virtual object corresponding to the target object.

[0211] Further, the apparatus further comprises:

[0212] The interaction request generation module is configured to generate an interaction request based on object information of the target object; the interaction request indicates object information of an object to be interacted with;

[0213] The interaction request response module is configured to, in response to the interaction request, determine a matching object that meets the interaction request, and generate interaction information to be sent;

[0214] The third interaction module is configured to send the interaction information to be sent to the matching object.

[0215] Further, the apparatus further comprises:

[0216] The comment module is configured to, in the case that interaction information between the auxiliary virtual objects corresponding to the plurality of interaction objects is displayed, display comment information of at least one interaction object in response to a comment operation triggered by the at least one interaction object; the at least one interaction object is at least one of the plurality of interaction objects.

[0217] And / or,

[0218] The fourth interaction module is configured to, in the case that interaction information between the auxiliary virtual objects corresponding to the plurality of interaction objects is displayed, establish an interaction link between a first interaction object and a second interaction object in response to a selection operation of the first interaction object on the second interaction object; the first interaction object and the second interaction object are different interaction objects of the plurality of interaction objects.

[0219] The apparatus provided in the above embodiments can execute the method provided in any of the embodiments of the present application, and has the corresponding service module and advantages of executing the method. Technical details not described in the above embodiments can be referred to the method provided in any of the embodiments of the present application.

[0220] The embodiment further provides a computer readable storage medium, wherein at least one instruction or at least one program is stored in the storage medium, and the at least one instruction or the at least one program is loaded and executed by a processor to perform any of the above methods.

[0221] According to an aspect of the present application, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. A processor of an electronic device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to make the computer device perform any of the above methods.

[0222] Figure 7 is a block diagram of an electronic device for interaction processing according to an exemplary embodiment. The electronic device can be a terminal, and its internal structure diagram can be as shown in Figure 7 The electronic device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the electronic device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement an interaction processing method. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen. The input device of the electronic device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the electronic device, or an external keyboard, touchpad or mouse, etc.

[0223] Figure 8 is a block diagram of an electronic device for interaction processing according to an exemplary embodiment. The electronic device can be a terminal, and its internal structure diagram can be as shown in Figure 8As shown. The electronic device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium, an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the electronic device is used to communicate with the external terminal through the network connection. The computer program is executed by the processor to implement an interactive processing method.

[0224] Those skilled in the art can understand that, Figure 7 and Figure 8 The structure shown in the embodiment is only a block diagram of part of the structure related to the scheme of the present disclosure, and does not constitute a limitation on the electronic device to which the scheme of the present disclosure is applied. The specific electronic device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0225] The present specification provides method operation steps as described in the embodiments or flowcharts, but can include more or fewer operation steps based on conventional or non-inventive labor. The steps and order listed in the embodiments are only one of the many execution orders of the steps, and do not represent the only execution order. When the system or interrupt product is executed in practice, it can be executed in sequence or in parallel (such as parallel processor or multi-threaded processing environment) according to the method order shown in the embodiments or drawings.

[0226] The structure shown in the embodiment is only part of the structure related to the scheme of the present disclosure, and does not constitute a limitation on the device to which the scheme of the present disclosure is applied. The specific device can include more or fewer components than those shown, or combine certain components, or have a different arrangement of components. It should be understood that the methods, apparatuses, etc. disclosed in the embodiments can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the modules is only a logical service division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the coupling or direct coupling or communication connection between the displayed or discussed each other can be through some interface, device or unit module indirect coupling or communication connection.

[0227] Based on such understanding, the technical solutions of the present application, essentially or in part, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and various other media that can store program codes.

[0228] Those skilled in the art can further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in the specification can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the foregoing description in a general manner. Whether the described services are executed in hardware or software depends on the specific application and design constraints of the technical solutions. Those skilled in the art can use different methods to implement the described services for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0229] The above-described embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An interactive processing method, characterized in that: include: Obtain target interaction information of the target object; Performing service analysis on the target interaction information based on the target interaction model to obtain target service call information; The target service calling information is used to call the target service required to process the target interaction information; The target interaction model is obtained by fine-tuning a pre-trained language model based on sample interaction information and tag service call information corresponding to the sample interaction information; Calling the target service based on the target service calling information to obtain a service calling result; Return target reply information generated based on the service call result to the target object.

2. The method according to claim 1, characterized in that The target interaction model includes an intention recognition sub-model and a service analysis sub-model; The performing service analysis processing on the target interaction information based on the target interaction model to obtain target service call information includes: Performing intent recognition on the target interaction information based on the intent recognition sub-model to obtain a target intent corresponding to the target interaction information; The target intention is subjected to service analysis processing based on the service analysis sub-model to obtain the target service call information.

3. The method according to claim 1 or 2, characterized in that The performing service analysis processing on the target interaction information based on the target interaction model to obtain target service call information includes: Performing service analysis on the target interaction information based on the target interaction model to obtain call information corresponding to each of the multiple services; The call information corresponding to each of the multiple services is combined to obtain the target service call information.

4. The method according to claim 3, characterized in that The combining of the call information corresponding to each of the plurality of services to obtain the target service call information includes: Based on the calling order of the multiple services, the calling information corresponding to each of the multiple services is arranged to obtain a calling sequence; the calling order is determined by performing dependency analysis on the multiple services; The target service calling information is determined based on the calling sequence.

5. The method according to claim 3, characterized in that The performing service analysis processing on the target interaction information based on the target interaction model to obtain call information corresponding to each of the multiple services includes: Performing service analysis on the target interaction information based on the target interaction model to obtain a service identifier of each service and a call parameter of each service; the call parameter of each service is used to indicate a call object when calling each service; Based on the service identifier of each service and the calling parameters of each service, calling information corresponding to each of the plurality of services is generated.

6. The method according to claim 1, characterized in that The method further comprises: Acquire sample corpus information of a target interaction style; the sample corpus information includes object corpus information of a sample object and label corpus information; the label corpus information is reply information having the target interaction style and corresponding to the object corpus information; The pre-trained language model is fine-tuned based on the object corpus information and the label corpus information to obtain a target interaction model with the target interaction style.

7. The method according to claim 6, characterized in that The method further comprises: In response to the virtual object creation request, creating an initial virtual object in the social platform; A target interaction model having the target interaction style is embedded into the initial virtual object to obtain an auxiliary virtual object corresponding to the target object; the auxiliary virtual object is used to receive messages for the target object and send messages in the target interaction style on the social platform with the authorization of the target object.

8. The method according to claim 7, characterized in that The method further comprises: Display auxiliary virtual objects corresponding to multiple interactive objects; In response to a selection operation based on any auxiliary virtual object, an interaction link is established between the target object and the any auxiliary virtual object, or an interaction link is established between the auxiliary virtual object corresponding to the target object and the any auxiliary virtual object; the any auxiliary virtual object is any auxiliary virtual object among the auxiliary virtual objects corresponding to each of the multiple interactive objects.

9. The method according to claim 7, characterized in that The method further comprises: In the case of authorization by the target object, interaction information is sent to a friend object of the target object based on the auxiliary virtual object corresponding to the target object.

10. The method according to claim 7, characterized in that The method further comprises: In response to the object designation operation, if the target object authorizes, the interaction information is sent to the designated object or the auxiliary virtual object corresponding to the designated object based on the auxiliary virtual object corresponding to the target object.

11. The method according to claim 7, characterized in that The method further comprises: generating an interaction request based on the object information of the target object; wherein the interaction request indicates the object information of the object to be interacted with; In response to the interaction request, determining a matching object that meets the interaction request, and generating interaction information to be sent; Send the interaction information to be sent to the matching object.

12. The method according to claim 7, characterized in that The method further comprises: In the case of displaying interaction information between auxiliary virtual objects corresponding to respective multiple interactive objects, in response to a comment operation triggered by at least one interactive object, displaying comment information of the at least one interactive object; the at least one interactive object is at least one interactive object among the multiple interactive objects; and / or, When displaying interaction information between auxiliary virtual objects corresponding to multiple interactive objects, an interaction link is established between the first interactive object and the second interactive object in response to the first interactive object's selection operation on the second interactive object; the first interactive object and the second interactive object are different interactive objects among the multiple interactive objects.

13. An interactive processing device, characterized in that: include: An interaction information acquisition module is used to acquire target interaction information of a target object; A call analysis module is used to perform service analysis on the target interaction information based on a target interaction model to obtain target service call information; The target service calling information is used to call the target service required to process the target interaction information; The target interaction model is obtained by fine-tuning a pre-trained language model based on sample interaction information and tag service call information corresponding to the sample interaction information; A service calling module is used to call the target service based on the target service calling information to obtain a service calling result; The information return module is used to return target reply information generated based on the service call result to the target object.

14. An electronic device, characterized in that: The device includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the interactive processing method according to any one of claims 1 to 12.

15. A computer storage medium, characterized in that The storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded by a processor and executes the interactive processing method according to any one of claims 1 to 12.

Images