Human-computer interaction data processing methods and server

Abstract

Provided in the present disclosure are human-computer interaction data processing methods and a server. In the method of the present disclosure, on the basis of pre-labeled information in multiple rounds of interaction in any session, labeled preceding information of input information in each round of interaction is determined, and according to the input information in each round of interaction and the labeled preceding information, a response result of the input information in each round of interaction is outputted by means of a human-computer interaction model, such that it can be ensured that preceding information of the same input information of each human-computer interaction model in multiple rounds of interaction is consistent; and according to the input information in each round of interaction in at least one session, the preceding information of the input information in each round of interaction and the response result of the input information in each round of interaction, multi-round interaction capability evaluation information of the human-computer interaction model is determined. In this way, uniform and pre-labeled information in multiple rounds of interaction can be used to fairly evaluate the multi-round interaction capability of a human-computer interaction model, and therefore evaluation information of the multi-round interaction capability of the human-computer interaction model is obtained, and launch determination of the human-computer interaction model can be guided or an optimized version of the human-computer interaction model can be updated, thereby improving the human-computer interaction quality of the human-computer interaction model.

Description

[0001] Cross-reference to this disclosure claims priority to Chinese patent application No. 202310721009.0, filed with the Patent Office of China on June 19, 2023, entitled "Data Processing Method and Server for Human-Computer Interaction," the entire contents of which are incorporated herein by reference. Technical Field This disclosure relates to computer technology, and more particularly to a data processing method and server for human-computer interaction. Background Natural language is an important vehicle for human logic and thinking, and is of great significance in human-computer interaction and even in the field of general artificial intelligence. However, due to the complexity and ambiguity of natural language, machine facilities that directly address unconstrained natural language have long been lacking. With the development of artificial intelligence, large models have been widely used in human-computer interaction in the field of natural language processing. Large models refer to large-scale deep learning models, such as large-scale language models and multimodal models, with large model parameters, typically containing hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than ten trillion model parameters. Multi-round interaction capability is a valuable capability of current human-computer interaction models, and objectively evaluating these models' multi-round interaction capabilities is crucial. Different human-computer interaction models produce different results for the same input command, making it difficult to use a unified set of pre-defined commands when evaluating the multi-round interaction capabilities of human-computer interaction models and humans. Consequently, it is difficult to fairly evaluate the multi-round interaction capabilities of different human-computer interaction models. This hinders the selection of high-quality models with excellent multi-round interaction capabilities during model iteration and the quality control of multi-round interactions for online models, resulting in poor human-computer interaction quality. This disclosure provides a human-computer interaction data processing method and server to address the issue of being unable to fairly evaluate the multi-round interaction capabilities of different human-computer interaction models, hindering the selection of high-quality models with excellent multi-round interaction capabilities during model iteration and the quality control of multi-round interactions for online models, leading to poor human-computer interaction quality. In a first aspect, the present disclosure provides a human-computer interaction data processing method, comprising: obtaining pre-annotated multi-round input information from at least one conversation, and pre-annotated response information corresponding to each round of input information; determining context information for each round of input information in the conversation based on the pre-annotated multi-round input information from any one conversation and the pre-annotated response information corresponding to each round of input information; inputting each round of input information and the context information of each round of input information in the conversation into a human-computer interaction model; and outputting, via the human-computer interaction model, response results for each round of input information; determining evaluation information of the multi-round interaction capability of the human-computer interaction model based on each round of input information, the context information of each round of input information, and the response results of each round of input information in the at least one conversation; and outputting the evaluation information.In a second aspect, the present disclosure provides a human-computer interaction data processing method, applied to a server, comprising: receiving a multi-round interaction capability evaluation request for multiple language models sent by a client-side device; obtaining pre-annotated multi-round input information of at least one session, as well as pre-annotated response information for each round of input information; determining context information for each round of input information in the session based on the pre-annotated multi-round input information of any one session and the pre-annotated response information for each round of input information; inputting the input information for each round of the session and the pre-annotated response information for each round of input information into each language model, and outputting a response result for each round of input information through each language model; determining evaluation information for multiple language models in various multi-round interaction capability dimensions based on the input information for each round of the at least one session, the context information for each round of input information, and the response results for each round of input information output by each language model, as well as the multi-round interaction capabilities covered by each session; and outputting the evaluation information to the client-side device. In a third aspect, the present disclosure provides a human-computer interaction data processing method, applied to a server, comprising: obtaining a model to be evaluated provided by a terminal-side device; obtaining, in response to a multi-round interaction capability evaluation request for the model to be evaluated sent by the terminal-side device, multi-round input information of at least one pre-annotated session, and pre-annotated response information corresponding to each round of input information; determining context information of each round of input information in the session based on the pre-annotated multi-round input information of any one session and the pre-annotated response information corresponding to each round of input information, inputting the input information of each round of the session and the context information of each round of input information into the model to be evaluated, and outputting a response result of each round of input information through the model to be evaluated; determining evaluation information of the multi-round interaction capability of the model to be evaluated based on the input information of each round of the at least one session, the context information of each round of input information, and the response result of each round of input information; and outputting the evaluation information to the terminal-side device. In a fourth aspect, the present disclosure provides a server comprising: a processor and a memory communicatively connected to the processor; the memory storing computer-executable instructions; and the processor executing the computer-executable instructions stored in the memory to implement the method described in any of the preceding aspects. In a fifth aspect, the present disclosure provides an electronic device comprising: a memory and a processor; the memory being configured to store computer-executable instructions, and the processor being configured to execute the computer-executable instructions, wherein the computer-executable instructions, when executed by the processor, implement the method described in any of the preceding aspects. In a sixth aspect, the present disclosure provides a computer program product comprising a computer program, wherein the computer program, when executed by the processor, implements the method described in any of the preceding aspects.In a seventh aspect, the present disclosure provides a computer program product, comprising a non-volatile computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method described in any of the preceding aspects. In an eighth aspect, the present disclosure provides a computer program. When executed by a processor, the computer program implements the method described in any of the preceding aspects. In a ninth aspect, the present disclosure provides a computer-readable storage medium storing computer-executable instructions. When executed by a processor, the instructions implement the method described in any of the preceding aspects. The present disclosure provides a human-computer interaction data processing method and server. The method obtains pre-annotated multi-round interaction information from at least one conversation, wherein the multi-round interaction information from one conversation includes multiple rounds of input information and pre-annotated response information corresponding to each round of input information. The method then determines context information for each round of input information in the conversation based on the pre-annotated multi-round interaction information from any conversation. The method then inputs the input information and the context information for each round of input information into a human-computer interaction model. The human-computer interaction model then outputs a response result for each round of input information. This ensures that, when the pre-annotated multi-round interaction information from any conversation is used to evaluate multiple different human-computer interaction models, the historical context information for the same input information in the multi-round interactions of the conversation is completely consistent across all human-computer interaction models. Furthermore, the method determines evaluation information for the multi-round interaction capability of the human-computer interaction model based on the input information, the context information for each round of input information, and the response result from the at least one conversation. This allows for a fair and objective evaluation of the multi-round interaction capabilities of different human-computer interaction models using the unified pre-annotated multi-round interaction information. The system also outputs evaluation information on the multi-round interaction capabilities of the human-computer interaction model. This evaluation information is used to guide the launch decision of the human-computer interaction model or to update the optimized version of the human-computer interaction model. This allows for accurate selection of high-quality models during human-computer interaction model iteration, improving the multi-round interaction quality of the iteratively updated human-computer interaction model, and improving the multi-round interaction quality of the launched model, thereby improving the multi-round interaction quality of human-computer interaction. BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings herein are incorporated into and constitute a part of this specification, illustrate embodiments consistent with the present disclosure, and together with the specification, serve to explain the principles of the present disclosure.Figure 1 is a schematic diagram of an exemplary system architecture applicable to the present disclosure; Figure 2 is a flowchart of a data processing method for human-computer interaction provided by an exemplary embodiment of the present disclosure; Figure 3 is an exemplary diagram of a front-end interactive interface provided by an exemplary embodiment of the present disclosure; Figure 4 is a flowchart of a data processing method for human-computer interaction provided by another exemplary embodiment of the present disclosure; Figure 5 is an interaction flowchart of a data processing method for human-computer interaction provided by an exemplary embodiment of the present disclosure; Figure 6 is a schematic diagram of the structure of a server provided by an exemplary embodiment of the present disclosure; and Figure 7 is a schematic diagram of the structure of a processor provided by an exemplary embodiment of the present disclosure. The above figures illustrate specific embodiments of the present disclosure, which will be described in more detail below. These figures and the accompanying description are not intended to limit the scope of the present disclosure in any way, but rather to illustrate the concepts of the present disclosure for those skilled in the art by reference to specific embodiments. The exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying figures. In the following description, when referring to the figures, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Rather, they are merely examples of apparatuses and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims. First, let's explain the terms used in this disclosure: "Conversation": A computer term that refers to the process of communication between an end user and a human-computer interaction system. For example, a conversation occurs from the moment a user enters the system and begins using the interaction functions to the moment the user exits the system and ends the interaction. During a conversation, the user enters a command, and the system responds to the command. This constitutes a round of dialogue, and a conversation can include one or more rounds of dialogue between the user and the system. Visual Question Answering Task: Given an input image and a question, the user determines the answer to a question from the visual information of the input image. Image Captioning Task: Generates a description of the input image. Visual Entailment Task: Predicts the semantic relevance between the input image and text, i.e., entailment, neutrality, or contradiction. Referential Expression and Comprehension Task: Based on the input text, the user locates the image region corresponding to the input text. Image Generation Task: Generates an image based on the input description text. Text-Based Sentiment Classification Task: Predicts the sentiment classification information of the input text. Text Summarization Task: Generates a summary of the input text. Multimodal tasks: These are downstream tasks whose input and output data involve multiple modal data such as images and text, such as visual question answering tasks, image description tasks, visual implication tasks, referential expression and understanding tasks, and image generation tasks.Multimodal pre-trained models are pre-trained models whose input and output data involve multiple modalities, such as images and text. After fine-tuning and training, they can be applied to multimodal task processing. Large models are deep learning models with large model parameters, typically containing hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than 10 trillion. Large models, also known as foundation models, are pre-trained on large-scale unlabeled corpora, producing pre-trained models with parameters exceeding 100 million. Such models are adaptable to a wide range of downstream tasks and exhibit good generalization capabilities. Examples include large language models (LLMs) and multi-modal pre-training models. In practical applications, large models require only a small number of samples to fine-tune the pre-trained model and can be applied to various tasks. Large models are widely applicable in fields such as natural language processing (NLP) and computer vision. Specifically, they can be applied to computer vision tasks such as visual question answering (VQA), image captioning (IC), and image generation, as well as natural language processing tasks such as text-based sentiment classification, text summarization, and machine translation. Key application scenarios for large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, and intelligent design. When applied to human-computer interaction scenarios (such as intelligent robots), large models generate responses based on user input commands and are also called human-computer interaction models. During the iteration process of human-computer interaction models, the pros and cons of different versions of the models need to be evaluated to ensure that they can be iteratively updated. Before launching a human-computer interaction model, it's necessary to evaluate its performance to ensure it meets launch requirements. Models with excellent performance should be selected for launch, while models with poor performance should be avoided. Multi-round interaction is a key value proposition for current human-computer interaction models, and objectively evaluating these models' multi-round interaction capabilities is crucial. Given the varying results produced by different human-computer interaction models for the same input command, it's difficult to use a unified set of pre-defined commands when evaluating the multi-round interaction capabilities of human-computer interaction models and humans. Consequently, it's difficult to fairly and objectively evaluate the multi-round interaction capabilities of different models. This hinders the selection of high-quality models with strong multi-round interaction capabilities during model iteration and hinders the quality of multi-round interactions for the launched models, resulting in poor human-computer interaction quality.The present disclosure provides a human-computer interaction data processing method. The method obtains pre-annotated multi-round interaction information from at least one conversation, the multi-round interaction information including multi-round input information and pre-annotated response information for each round of input information. Based on the pre-annotated multi-round input information and pre-annotated response information for each round of input information, the method determines contextual information for each round of input information in the conversation. The pre-annotated multi-round input information and the pre-annotated response information for each round of input information are input into a human-computer interaction model. The human-computer interaction model outputs a response result for each round of input information. This ensures that, when the pre-annotated multi-round interaction information from any conversation is used to evaluate multiple different human-computer interaction models, the historical contextual information for the same input information in the multi-round interactions of the conversation is completely consistent for each human-computer interaction model. Furthermore, based on the pre-annotated multi-round interaction information, the pre-annotated response information, and the response result for each round of input information in the at least one conversation, evaluation information for the multi-round interaction capability of the human-computer interaction model is determined. The method uses the unified pre-annotated multi-round interaction information to fairly and objectively evaluate the multi-round interaction capabilities of different human-computer interaction models. The system also outputs evaluation information on the multi-round interaction capability of the human-computer interaction model. This evaluation information is used to guide the launch decision of the human-computer interaction model or to update the optimized version of the human-computer interaction model. This helps accurately select high-quality models during iteration of the human-computer interaction model, improve the multi-round interaction quality of the iteratively updated human-computer interaction model, and improve the multi-round interaction quality of the launched model, thereby improving the quality of multi-round interactions in human-computer interaction. The human-computer interaction model in this disclosure can be a large-scale pre-trained language model, a multimodal pre-trained model, or the like, without specific limitation. Figure 1 is a schematic diagram of an example system architecture applicable to this disclosure. As shown in Figure 1, the system architecture includes a first server responsible for evaluating the multi-round interaction capability of the human-computer interaction model, a second server running the human-computer interaction model, and an end-side device. A communication link is established between the first server and the second server, enabling communication between the first server and the second server. A communication link is established between the first server and the end-side device, enabling communication between the first server and the end-side device. The second server can be a server cluster deployed in the cloud or a local device with computing capabilities. The second server is responsible for running the human-computer interaction model to be evaluated and outputting a response based on the given input information. One or more human-computer interaction models can be deployed on a single second server. Multiple human-computer interaction models to be evaluated can be deployed on one or more second servers. A terminal device is an electronic device used by a user. Specifically, it can be a hardware device with network communication, computing, and information display capabilities. These devices include, but are not limited to, smartphones, tablets, desktop computers, and servers.A user sends a human-computer interaction model evaluation request to a first server via a client-side device. This evaluation request includes information about one or more human-computer interaction models to be evaluated. The first server can be a server cluster deployed in the cloud or a local device with computing power. The first server is responsible for evaluating the multi-round interaction capabilities of the human-computer interaction models, generating evaluation information on the multi-round interaction capabilities of the human-computer interaction models, and guiding the decision to launch the human-computer interaction models, updating optimized versions of the human-computer interaction models, or selecting high-quality human-computer interaction models with strong multi-round interaction capabilities. In one example scenario, before launching a human-computer interaction model, a user sends a multi-round interaction capability evaluation request to the first server via a client-side device. This evaluation request includes relevant information about the human-computer interaction models to be evaluated, such as the application programming interface for invoking the human-computer interaction model and the access address of the human-computer interaction model. In response to the evaluation request, the first server obtains pre-annotated multi-round interaction information from at least one session. The multi-round interaction information includes multi-round input information and pre-annotated response information for each round of input information. Based on the pre-annotated multi-round input information and pre-annotated response information for each round of input information from any session, the first server determines context information for each round of input information in the current session. The first server then inputs the input information and the context information for each round of input information in the current session into a human-computer interaction model. The human-computer interaction model then outputs a response result for each round of input information. Based on the input information, context information, and response result for each round of input information in the at least one session, the first server determines evaluation information for the multi-round interaction capability of the human-computer interaction model, thereby accurately evaluating the multi-round interaction capability of the human-computer interaction model. Furthermore, the evaluation information for the multi-round interaction capability of the human-computer interaction model can be used to guide the decision on whether to launch the human-computer interaction model. Optionally, the first server determines whether the human-computer interaction model meets the online conditions based on the evaluation information of the multi-round interaction capability of the human-computer interaction model, and outputs online prompt information indicating whether the human-computer interaction model meets the online conditions. Optionally, the first server sends the evaluation information of the multi-round interaction capability of the human-computer interaction model to the end-side device. The end-side device outputs the evaluation information of the multi-round interaction capability of the human-computer interaction model to guide the user in determining whether the human-computer interaction model meets the online conditions. Alternatively, the end-side device determines whether the human-computer interaction model meets the online conditions based on the evaluation information of the multi-round interaction capability of the human-computer interaction model, and outputs online prompt information indicating whether the human-computer interaction model meets the online conditions. Optionally, the end-side device may further output the evaluation information of the multi-round interaction capability of the human-computer interaction model. In another example scenario, during the iterative optimization process of the human-computer interaction model, the new version obtained is evaluated.A user can send a request to the first server via a client device to evaluate the multi-round interaction capabilities of a new version of the human-computer interaction model. The request includes relevant information about the new version of the human-computer interaction model, such as the application program interface (API) for invoking the human-computer interaction model and the access address of the human-computer interaction model. In response to the request, the first server obtains pre-annotated multi-round interaction information from at least one session, including multiple rounds of input information and pre-annotated responses for each round of input information. Based on the relevant information about the new version of the human-computer interaction model and the pre-annotated multi-round interaction information from any session, the first server determines the context of each round of input information in the current session. The server then inputs the input information and the context of each round of input information into the new version of the human-computer interaction model, and outputs responses for each round of input information via the new version of the human-computer interaction model. The first server then determines evaluation information for the multi-round interaction capabilities of the new version of the human-computer interaction model based on the input information, the context of each round of input information, and the responses from the at least one session, thereby accurately evaluating the multi-round interaction capabilities of the new version of the human-computer interaction model. Furthermore, the evaluation information of the multi-round interaction capability of the new version of the human-computer interaction model can be used to guide the update of the optimized version of the human-computer interaction model. Optionally, the first server compares the evaluation information of the multi-round interaction capability of the new version of the human-computer interaction model with the evaluation information of the multi-round interaction capability of the previous version, obtaining a comparison result. The comparison result is used to guide the update of the optimized version of the human-computer interaction model. Specifically, the first server can send the comparison result to the end-side device. The end-side device outputs the comparison result of the evaluation information of the multi-round interaction capability of different versions of the human-computer interaction model to guide the user to select the optimized version with stronger multi-round interaction capability for iterative update of the human-computer interaction model. In another example scenario, based on the evaluation information of the multi-round interaction capability of multiple selected human-computer interaction models, the user can select a human-computer interaction model with stronger multi-round interaction capability for human-computer interaction, thereby improving the quality of human-computer interaction. The user can send a request for evaluating the multi-round interaction capabilities of multiple human-computer interaction models to the first server through the next-side device. The evaluation request includes relevant information of multiple human-computer interaction models, such as the application interface for calling the human-computer interaction model and the access address of the human-computer interaction model.In response to the evaluation request, the first server obtains pre-annotated multi-round interaction information from at least one session, the multi-round interaction information including multi-round input information and pre-annotated response information for each round of input information. Based on the pre-annotated multi-round interaction information from any session, the first server determines context information for each round of input information in the session. The first server then inputs the input information and context information for each round of input information into each human-computer interaction model, and each human-computer interaction model outputs a response result for each round of input information. Based on the input information for each round of input information, the context information for each round of input information, and the response result for each round of input information output by each human-computer interaction model in the at least one session, the first server determines evaluation information for the multi-round interaction capability of each human-computer interaction model, thereby fairly and accurately evaluating the multi-round interaction capability of each human-computer interaction model. Furthermore, the first server compares the evaluation information for the multi-round interaction capability of each human-computer interaction model to obtain a comparison result for the multi-round interaction capability evaluation information of each human-computer interaction model. The first server then sends the comparison result for the multi-round interaction capability evaluation information of each human-computer interaction model to the end-side device. The end-side device outputs the comparison results to guide the user in selecting a human-computer interaction model with stronger multi-round interaction capabilities as the human-computer interaction model of their choice. Optionally, based on the comparison results of the multi-round interaction capability evaluation information of each human-computer interaction model, the end-side device may select a human-computer interaction model with stronger multi-round interaction capabilities and, based on the relevant information of the selected human-computer interaction model, download and obtain the human-computer interaction model, or use the human-computer interaction model to implement human-computer interaction. The following detailed description of the technical solution of the present disclosure and how it solves the aforementioned technical problems is provided with specific embodiments. The following specific embodiments may be combined with each other, and identical or similar concepts or processes may not be described in detail in certain embodiments. The embodiments of the present disclosure are described below with reference to the accompanying drawings. Figure 2 is a flow chart of a human-computer interaction data processing method provided by an exemplary embodiment of the present disclosure. This embodiment is executed by the first server in the aforementioned system architecture. As shown in Figure 2, the method comprises the following steps: Step S201: Obtain pre-annotated multi-round input information for at least one conversation, as well as pre-annotated response information corresponding to each round of input information. The multi-round interaction information of a conversation includes multiple rounds of input information (user commands) and response information corresponding to each round of input information. In this embodiment, the multi-round interaction information of at least one conversation of human-computer interaction can be collected or manually constructed. By annotating the response information of the input information in the multi-round interaction information of at least one conversation, the response information corresponding to the input information in the annotated multi-round interaction information is the correct / standard response information of the input information.For example, an example of pre-annotated multi-turn interaction information from a conversation (referred to as Example 1) is as follows: Input: What is the largest animal on Earth? Response: The blue whale, a marine mammal of the family Mysticetiidae and genus Mysticeti, can reach up to 33 meters in length and weigh over 180 tons. Input: What does it eat? Response: Blue whales primarily feed on small crustaceans and small fish, sometimes also ichthys. Input: Where is their habitat? Response: Blue whales are found worldwide, with the largest population in Antarctic waters. Input: Excluding this question, how many questions did I ask in total? Response: Three questions: "What is the largest animal on Earth?", "What does it eat?", and "Where is its habitat?" The multi-turn interaction information in this example includes four rounds of dialogue, i.e., four rounds of input information and pre-annotated responses for each round. In subsequent steps, the multi-round interaction capabilities of one or more human-computer interaction models are evaluated using pre-annotated multi-round interaction information from at least one conversation to obtain evaluation information on the multi-round interaction capabilities of the human-computer interaction models. Step S202: Based on the pre-annotated multi-round input information from any conversation and the pre-annotated response information corresponding to each round of input information, the context of each round of input information in the conversation is determined. Each round of input information and its context are then input into the human-computer interaction model. The human-computer interaction model then outputs a response result for each round of input information. In this step, the pre-annotated multi-round interaction information from each conversation is used to determine the context of any round of input information in the conversation based on the multi-round interaction information. The input information and its context are then input into the human-computer interaction model to be evaluated. The human-computer interaction model then outputs a response result for the input information. In this embodiment, since the response information in the multi-round interaction information of each conversation is pre-annotated, the context of any round of input information can be determined based on the pre-annotated multi-round interaction information of any conversation. Furthermore, the response information included in the context is a pre-annotated, accurate / standard response. This context can be understood as pre-annotated context information. When applied to the evaluation of multiple different human-computer interaction models, the context of the same input information is fixed, pre-annotated context information. This ensures that the historical context of the same input information across multiple rounds of interaction in the conversation is completely consistent across all human-computer interaction models. This allows for the use of unified, pre-annotated multi-round interaction information to fairly and objectively evaluate the multi-round interaction capabilities of different human-computer interaction models.In an optional embodiment, in this step, the first server may, based on the multi-round interaction information of any session, use the input information prior to the current round and the pre-annotated response information corresponding to the input information prior to the current round as the context information for the current round of input information for any round of input information in the current session. For the multi-round interaction information of each session, if no context information exists for the first round of input information, the context information table TF for the first round of input information may be empty. For example, taking Example 1 in step S201 above, the first round of input information "What is the largest animal on Earth?" in Example 1 does not have any context information, i.e., the context information for the first round of input information is empty. The context information for the second round of input information "What does it eat?" in Example 1 includes the first round of input information and the response information for the first round of input information. The context information for the third round of input information "Where is the habitat?" in Example 1 includes the first round of input information, the response information for the first round of input information, the second round of input information, and the response information for the second round of input information. In Example 1, the context of the fourth round of input information, "Excluding this question, how many questions did I ask in total?", includes the first round of input information, the response information to the first round of input information, the second round of input information, the response information to the second round of input information, the third round of input information, and the response information to the third round of input information. In another optional embodiment, a multi-round interaction process can be simulated based on pre-annotated multi-round interaction information of any conversation, and the following interaction processing can be performed on each round of input information in the conversation: the context information of the conversation is used as the context information of the current round of input information, where the context information of the conversation includes historical interaction information of the conversation; the current round of input information and the context information of the conversation are input into a human-computer interaction model, and the human-computer interaction model outputs a response result for the current round of input information; and the context information of the conversation is updated, replacing the response result of the current round of input information in the updated context information of the conversation with the pre-annotated response information corresponding to the current round of input information. In this implementation, the human-computer interaction model executes multiple rounds of interaction in each conversation, and the responses to the input information in the context information used in each round are forcibly corrected to the pre-annotated response information for that input information. This ensures that the contextual information of the input information provided by each human-computer interaction model in each round of the conversation is consistent. This allows for the use of unified, pre-annotated multi-round interaction information to fairly and objectively evaluate the multi-round interaction capabilities of different human-computer interaction models. In this embodiment, the human-computer interaction model is used to generate and output responses based on user-provided input information, thereby implementing human-computer interaction.The human-computer interaction model, as the object to be evaluated, can specifically be a large-scale pre-trained language model or a multimodal pre-trained model. It is specifically applied to fields such as natural language processing (NLP) and computer vision. Specifically, it can be applied to tasks at the intersection of NLP and computer vision, such as visual question answering (VQA), image captioning (IC), visual entailment (VE), and referential expression and understanding (REC), as well as tasks in the field of natural language processing, such as text-based sentiment classification and text summarization. It can be applied to various application scenarios, such as digital assistants, intelligent robots, search, online education, office software, e-commerce, and intelligent design. Optionally, the first server can obtain an application programming interface (API) or a service interface for the human-computer interaction model, and by calling the API or service interface of the human-computer interaction model, input the input information and the context information into the human-computer interaction model, and receive a response output by the human-computer interaction model. Optionally, the first server may obtain access address information for the human-computer interaction model. The first server sends a reply generation request to a second server where the human-computer interaction model resides. The request includes the input information and context information to be replied to. In response to the reply generation request, the second server inputs the input information and context information into the human-computer interaction model, obtains a response result output by the human-computer interaction model for the input information, and sends the response result to the first server. Step S203: Determine evaluation information for the multi-round interaction capability of the human-computer interaction model based on the input information, context information, and response result of each round of input information in at least one session. In this embodiment, based on the input information, annotated context information, and response result output by the human-computer interaction model, the response quality of the response results during the multi-round interaction process of the human-computer interaction model is evaluated, thereby evaluating the multi-round interaction capability of the human-computer interaction model and obtaining evaluation information for the multi-round interaction capability of the human-computer interaction model. Step S204: Output the evaluation information.In this embodiment, after obtaining evaluation information on the multi-round interaction capability of a human-computer interaction model, the first server visually outputs the evaluation information to provide the user with the evaluation results. This information can guide the user in determining whether the human-computer interaction model should be put online. Alternatively, by comparing the evaluation information on the multi-round interaction capability of multiple human-computer interaction model versions, a high-quality human-computer interaction model version can be identified and iteratively optimized. Alternatively, by comparing the evaluation information on the multi-round interaction capability of multiple human-computer interaction models, a target human-computer interaction model with strong multi-round interaction capability can be selected as the human-computer interaction model for implementing human-computer interaction. For example, the first server can select one of the multiple human-computer interaction models as the target human-computer interaction model based on the evaluation information on the multi-round interaction capability of the multiple human-computer interaction models, and output information about the target human-computer interaction model to the end-side device. For example, the first server may select one of the multiple versions as an optimized version based on evaluation information of the multi-round interaction capabilities of multiple different versions of the human-computer interaction model, and update the optimized version of the human-computer interaction model. For example, after obtaining the evaluation information of the multi-round interaction capabilities of the human-computer interaction model, the first server may determine whether the human-computer interaction model meets online conditions based on the evaluation information of the multi-round interaction capabilities of the human-computer interaction model, and output online prompt information of the human-computer interaction model and / or the evaluation information of the multi-round interaction capabilities of the human-computer interaction model. The online prompt information indicates whether the human-computer interaction model meets the online conditions. The online conditions include a first threshold value of the evaluation information of the multi-round interaction capabilities of the human-computer interaction model. If the evaluation information of the multi-round interaction capabilities of the human-computer interaction model is greater than or equal to the first threshold value, the human-computer interaction model meets the online conditions; otherwise, the human-computer interaction model does not meet the online conditions. The first threshold value of the online conditions can be customized by the user based on the needs of a specific application scenario.The solution of this embodiment obtains pre-annotated multi-round interaction information from at least one conversation, where the multi-round interaction information includes multi-round input information and pre-annotated response information for each round of input information. Based on the pre-annotated multi-round input information and pre-annotated response information for each round of input information from any conversation, contextual information for each round of input information in the current conversation is determined. The input information and contextual information for each round of input information in the current conversation are then input into a human-computer interaction model, which then outputs a response result for each round of input information. When the pre-annotated multi-round interaction information from any conversation is used to evaluate multiple different human-computer interaction models, the historical contextual information for the same input information in the multi-round interactions of the current conversation is ensured to be completely consistent for each human-computer interaction model. Furthermore, based on the input information, contextual information, and response results for each round of input information in the at least one conversation, evaluation information for the multi-round interaction capability of the human-computer interaction model is determined. Using the unified pre-annotated multi-round interaction information, the multi-round interaction capabilities of different human-computer interaction models can be fairly and objectively evaluated, and the evaluation information for the multi-round interaction capability of the human-computer interaction model can be output. The multi-round interaction capability evaluation information of the human-computer interaction model is used to guide the launch decision of the human-computer interaction model, update the optimized version of the human-computer interaction model, or select the target human-computer interaction model for human-computer interaction. This information can accurately select high-quality models during human-computer interaction model iteration, improve the multi-round interaction quality of the iteratively updated human-computer interaction model, and thus improve the quality of multi-round interactions in human-computer interaction. In an optional embodiment, the human-computer interaction model to be evaluated can be provided by a third party to a first server via a device-side device. The first server obtains the model to be evaluated provided by the device-side device. For example, the third party can upload the model to be evaluated to the first server via the device-side device, and the first server can deploy the model to be evaluated to a second server or the first server. For example, the third party can send a download link of the model to be evaluated to the first server via the device-side device. The first server downloads the model to be evaluated via the download link and deploys the model to be evaluated to the second server or the first server. For example, the third party can send an application programming interface (API) or access address of the model to be evaluated to the first server via the device-side device. The first server inputs the input information and the above information into the model to be evaluated through an application programming interface (API) or an access address, and receives a response output by the model to be evaluated. Furthermore, the first server evaluates the multi-round interaction capability of the model to be evaluated through steps S201-S204, obtaining evaluation information on the multi-round interaction capability of the model to be evaluated. The first server can output the evaluation information on the multi-round interaction capability of the model to be evaluated to the client-side device.Furthermore, based on the evaluation information of the multi-round interaction capability of the model to be evaluated, the user can be guided to determine whether the human-computer interaction model should be put online. Alternatively, by comparing the multi-round interaction capability evaluation information of multiple human-computer interaction model versions, a high-quality human-computer interaction model version can be identified and iterative optimization of the human-computer interaction model can be performed. Alternatively, by comparing the multi-round interaction capability evaluation information of multiple human-computer interaction models, a target human-computer interaction model with stronger multi-round interaction capability can be selected as the human-computer interaction model for implementing human-computer interaction. For details, see the relevant content of step S204, which will not be repeated here. For example, FIG3 is an example diagram of a front-end interactive interface provided in an exemplary embodiment of the present disclosure. As shown in FIG3, the front-end interactive interface provides an input area for the model to be evaluated. Through this input area, the model to be evaluated can be provided to the first server by uploading the model, inputting a model call interface, or inputting a model download address. After completing the evaluation of the multi-round interaction capability of the model to be evaluated, the first server can output the evaluation information of the multi-round interaction capability of the model to be evaluated through the interactive interface. When determining whether a model is ready for launch, a launch prompt for the model to be evaluated can also be output. When comparing multiple models, the system can also output comparison results of the multi-round interaction capabilities of the multiple models, or recommend high-quality target models. Figure 4 is a flow chart of a data processing method for human-computer interaction provided by another exemplary embodiment of the present disclosure. Building upon any of the aforementioned embodiments, this embodiment provides a detailed description of the data processing method for human-computer interaction. As shown in Figure 4, the method of this embodiment comprises the following steps: Step S401: Obtain pre-annotated multi-round interaction information for at least one conversation, where the multi-round interaction information includes multiple rounds of input information and pre-annotated response information corresponding to each round of input information. This step is implemented in the same manner as Step S201 above. For details, please refer to the relevant section in the aforementioned embodiment and will not be further described in this embodiment.After obtaining pre-annotated multi-round interaction information of at least one conversation, steps S402-S406 sequentially perform the following interaction processing on each round of input information in the conversation based on the pre-annotated multi-round input information of any conversation and the pre-annotated response information of each round of input information: using the context information of the conversation as the context information of the current round of input information, where the context information of the conversation includes historical interaction information of the conversation; inputting the current round of input information and the context information of the conversation into a human-computer interaction model, and outputting a response result of the current round of input information through the human-computer interaction model; updating the context information of the conversation, and replacing the response result of the current round of input information in the updated context information of the conversation with the pre-annotated response information of the current round of input information, thereby achieving the aforementioned steps of determining the context information of each round of input information in the current conversation based on the pre-annotated multi-round interaction information of any conversation in step 202; inputting the input information of each round of the current conversation and the context information of the input information into the human-computer interaction model, and outputting a response result of the input information of each round of the current conversation through the human-computer interaction model. Step S402: Sequentially use the pre-annotated multi-round interaction information of each conversation as the multi-round interaction information of the current conversation. In this embodiment, for the pre-annotated multi-round interaction information of at least one conversation, the multi-round interaction information of each conversation is sequentially used as the multi-round interaction information of the current conversation. Based on the multi-round interaction information of the current conversation, steps S403-S406 are performed. The multi-round interaction of the current conversation is performed using the human-computer interaction model to be evaluated. Based on the input information of each round and the pre-annotated context information of each round of input information, a response result of each round of input information is output. Step S403: Sequentially use the input information of each round in the current conversation as the input information of the current round. For the multi-round dialogue information (including input information and annotated response information) in the multi-round interaction information of the current conversation, the input information of each round in the current conversation is sequentially used as the input information of the current round. Based on the human-computer interaction model to be evaluated, the interaction processing of steps S404-S406 is performed based on the input information of the current round to generate a response result of the input information of the current round based on the input information of the current round and the pre-annotated context information. Step S404: Use the context information of the current session as the context of the current round of input information. The context information of the current session includes historical interaction information of the current session. Typically, in a multi-round interactive session, after outputting a response to any round of input information, the context information of the current session is updated and stored. The context information of the current session records the historical interaction information of the current session, specifically including the input information processed in one or more rounds of the current session, as well as the response results of each round of input information. For any session, the context information of the current session is empty at the beginning.Step S405: Input the current round of input information and the context information of the current session into the human-computer interaction model, and the human-computer interaction model outputs a response result for the current round of input information. In this embodiment, for the current round of input information (the next input information to be processed), the stored current context information of the current session is used as the context information for the current round of input information. The current round of input information and the context information of the current session are then input into the human-computer interaction model to be evaluated. The human-computer interaction model generates and outputs a response result for the current round of input information based on the input information and context information. The human-computer interaction model is used to generate and output a response result based on the input information provided by the user, thereby implementing human-computer interaction. The human-computer interaction model to be evaluated can specifically be various language models, multimodal pre-trained models, etc., and is not specifically limited here. Optionally, the first server can obtain an application programming interface / human-computer interaction model service interface of the human-computer interaction model, and by calling the application programming interface / human-computer interaction model service interface of the human-computer interaction model, input the input information and the context information of the current session into the human-computer interaction model, and receive a response result output by the human-computer interaction model. Optionally, the first server may obtain access address information for the human-computer interaction model. The first server may then send a reply generation request to a second server hosting the human-computer interaction model. The request includes the input information to be replied to and the context information for the current session. In response to the reply generation request, the second server inputs the input information and the context information for the current session into the human-computer interaction model, obtains a response result for the input information output by the human-computer interaction model, and sends the response result to the first server. Furthermore, based on the response result for the current round of input information output by the human-computer interaction model, the context information for the current session is updated, thereby adding the current round of input information and the response result to the context information for the current session to obtain updated context information. Step S406: The context information for the current session is updated, and the response result for the current round of input information in the updated context information for the session is replaced with the pre-annotated response information corresponding to the current round of input information. In this embodiment, the pre-annotated response information corresponding to the current round of input information is used to replace the response result of the current round of input information in the updated context information of the current session. This forcibly corrects the response result of the current round of input information in the context information of the current session to the pre-annotated response information of the input information. This ensures that the context information used by each human-computer interaction model in the next round of dialogue is consistent, thereby ensuring that the context information of the input information used by each human-computer interaction model when processing each round of input information is consistent. This allows the use of unified pre-annotated multi-round interaction information to fairly evaluate the multi-round interaction capabilities of different human-computer interaction models.Through steps S402-S406, based on pre-annotated multi-round interaction information from any conversation, the context of each round of input information in this conversation is determined. The input information and context of each round of input information in this conversation are then input into the human-computer interaction model, which then outputs a response result for each round of input information. For example, based on the multi-round interaction information from Example 1 in step S201, and using a human-computer interaction model as an example, the processing of steps S403-S406 yields the response results shown in Table 1 below. The response results for any round listed in Table 1, along with the input data in the same row, constitute three sets of samples to be evaluated. Based on the example in Table 1, by replacing the response results of the current round of input information in the updated context information of the current conversation with the pre-annotated response information of the current round of input information, and introducing a teacher forcing mechanism to forcibly correct the response results of each round of input information in the context information of the current conversation to the pre-annotated response information, we can ensure that the context information used by each human-computer interaction model in each round of dialogue is consistent. Therefore, the above-mentioned information used by each human-computer interaction model when processing the input information in each round is consistent. This allows the use of unified pre-annotated multi-round interaction information to fairly evaluate the multi-round interaction capabilities of different human-computer interaction models. Table 1. In this embodiment, steps S401-S406 described above generate multiple groups of samples to be evaluated. Each group of samples includes input information and context information provided to the human-computer interaction model, as well as the response output by the human-computer interaction model. By breaking down the evaluation of the multi-round interaction capability of the human-computer interaction model into an evaluation of the response generated by a given input information and context information in a single-round dialogue, the unified, pre-labeled multi-round interaction information can be used to fairly evaluate the multi-round interaction capabilities of different human-computer interaction models. Step S407: Determine the interaction quality information of the human-computer interaction model in each session based on the input information of each round, the context of each round of input information, and the response results of each round of input information in at least one session. In this step, evaluation information of the response results of each round of input information can be determined based on the input information of each round of input information, the context of each round of input information, and the response results of each round of input information in at least one session. The evaluation information of the response results of each round of input information in a session is combined to determine the interaction quality information of the human-computer interaction model in the current session. For example, the three rounds of response results shown in Table 1 are three response results to be evaluated. For the input information of the current round shown in any row, different human-computer interaction models may output different response results. The different response results output by each human-computer interaction model for the same input information are evaluated separately to determine the evaluation information of the response results output by each human-computer interaction model for the same input information. Exemplarily, when determining evaluation information for the response results of any round of input information in any conversation based on the input information, contextual information of the input information, and the response results, the first server provides a first interactive interface. The first interactive interface displays the input information, contextual information of the input information, and the response results of the input information output by the human-computer interaction model. The first interactive interface does not display the correspondence between the response results and the human-computer interaction models. The first server receives a quality evaluation value of the response results for the input information entered in the first interactive interface. Exemplarily, the first server displays the first interactive interface via a client device running on a client-side device. A user can rate the response quality of the response results output by each human-computer interaction model for the current input information in the first interactive interface. The score is submitted to the client-side device as response result evaluation information. The client-side device then sends the response result evaluation information of each human-computer interaction model to the first server. In this way, users can manually annotate the response quality of the response results output by each human-computer interaction model without knowing which human-computer interaction model the response results originated from, thereby objectively evaluating the response results of each human-computer interaction model in a double-blind manner. The first server obtains evaluation information for the manually annotated response results. Optionally, when outputting the response results, the response results are desensitized to remove information related to the human-computer interaction model that generated the response results (e.g., information containing the human-computer interaction model's own identifier in the response results generated by the human-computer interaction model). This prevents annotators from knowing the correspondence between the response results and the human-computer interaction model. This ensures that annotators are unaware of the correspondence between the response results and the human-computer interaction model, achieving double-blind evaluation and improving the objectivity of the multi-round interaction capability evaluation of the human-computer interaction models.Furthermore, based on the weight coefficients corresponding to each round in a session, the evaluation information of the response results of each round of input information in the same session can be weighted averaged to obtain the interaction quality information of the human-computer interaction model in the current session. Optionally, the evaluation information of the response results of each round of input information in the same session can be averaged to obtain the interaction quality information of the human-computer interaction model in the current session. The weight coefficients of different rounds in a session can be configured based on empirical values ​​or customized by the user. Exemplarily, the first server provides a configuration interface for weight coefficients of different rounds in a session. This configuration interface is used to configure the weight coefficients of different rounds in a session. The weight coefficients of the same round in different sessions are consistent. The configuration interface can provide an input area for specifying the weight coefficient of a specific round. The weight coefficient corresponding to each round can be entered in the input area corresponding to each round. Rounds without configured weight coefficients are configured with default weight coefficients. The default weight coefficient can be configured based on empirical values. For example, the default weight coefficient can be configured to 1, 0.5, or 0, etc., which is not specifically limited here. The first server obtains the weight coefficients for each round in a session configured on the configuration interface. Step S408: Determine evaluation information for the multi-round interaction capability of the human-computer interaction model based on the interaction quality information of the human-computer interaction model in each session. In an optional embodiment, the first server may average the interaction quality information of the human-computer interaction model in each session and use the average interaction quality information of the human-computer interaction model in each session as the evaluation information for the multi-round interaction capability of the human-computer interaction model. In another optional embodiment, after obtaining pre-annotated multi-round input information from at least one session and pre-annotated response information corresponding to each round of input information, the multi-round interaction capability covered by each session may be annotated. The multi-round interaction capability includes at least one of the following: context-relevant, context-independent, and context-consistent. Context-relevant means that the previous information in a multi-round interaction may be helpful (positively impact) the response to the current input information. Context-independent means that the previous information in a multi-round interaction has a negative impact on the response to the current input information. Context-consistent means that the response to the current input information and the previous response in a multi-round interaction should remain consistent. The multi-round interaction capabilities covered by each conversation can be obtained by manual annotation by annotators. For example, Table 2 provides examples of multi-round interaction information covering different types of multi-round interaction capabilities: Table 2.

[0002] In this step, the first server may average the interaction quality information of the human-computer interaction model in sessions covering each type of multi-round interaction capability based on the multi-round interaction capabilities covered by at least one session and the interaction quality information of the human-computer interaction model in each session, to obtain evaluation information of the human-computer interaction model in each multi-round interaction capability dimension. Optionally, the first server may sum the interaction quality information of the human-computer interaction model in sessions covering each type of multi-round interaction capability based on the multi-round interaction capabilities covered by at least one session and the interaction quality information of the human-computer interaction model in each session, to obtain evaluation information of the human-computer interaction model in each multi-round interaction capability dimension. Furthermore, after obtaining the evaluation information of the human-computer interaction model in each multi-round interaction capability dimension, the first server may also weightedly combine the evaluation information of the human-computer interaction model in each multi-round interaction capability dimension based on the weight coefficient corresponding to each multi-round interaction capability, to obtain comprehensive evaluation information of the multi-round interaction capability of the human-computer interaction model. Optionally, when implementing weighted integration of evaluation information of the human-computer interaction model in various multi-round interaction capability dimensions based on weight coefficients corresponding to each multi-round interaction capability, the first server may perform a weighted summation of the evaluation information of the human-computer interaction model in various multi-round interaction capability dimensions based on the weight coefficients corresponding to each multi-round interaction capability to obtain comprehensive evaluation information of the multi-round interaction capability of the human-computer interaction model. Optionally, when implementing weighted integration of evaluation information of the human-computer interaction model in various multi-round interaction capability dimensions based on weight coefficients corresponding to each multi-round interaction capability, the first server may perform a weighted average of the evaluation information of the human-computer interaction model in various multi-round interaction capability dimensions based on the weight coefficients corresponding to each multi-round interaction capability to obtain comprehensive evaluation information of the multi-round interaction capability of the human-computer interaction model. Optionally, a weight configuration interface for each multi-round interaction capability is displayed; and the weight coefficients corresponding to each multi-round interaction capability configured on the weight configuration interface are obtained. The weight coefficients of each multi-round interaction capability can be configured based on empirical values ​​or customized by the user. Exemplarily, the first server provides a configuration interface for weight coefficients of various multi-round interaction capabilities. This configuration interface is used to configure the weight coefficients of various multi-round interaction capabilities. The configuration interface for the weight coefficients of various multi-round interaction capabilities is displayed on the end-side device. The configuration interface provides input areas for the weight coefficients of various multi-round interaction capabilities. The weight coefficients corresponding to the various multi-round interaction capabilities can be entered in the input areas corresponding to the various multi-round interaction capabilities. The end-side device obtains the weight coefficients of the various multi-round interaction capabilities configured on the configuration interface and sends them to the first server. The first server receives the weight coefficients of the various multi-round interaction capabilities configured on the configuration interface from the end-side device.In an optional embodiment, after obtaining comprehensive evaluation information on the multi-round interaction capability of the human-computer interaction model, the first server may determine whether the human-computer interaction model meets online requirements based on the comprehensive evaluation information on the multi-round interaction capability of the human-computer interaction model, and output online prompt information for the human-computer interaction model, indicating whether the human-computer interaction model meets the online requirements. Optionally, while outputting the online prompt information for the human-computer interaction model, the first server may also output evaluation information for the human-computer interaction model, such as comprehensive evaluation information on the multi-round interaction capability of the human-computer interaction model and evaluation information on various multi-round interaction capability dimensions of the human-computer interaction model. In an optional embodiment, there may be multiple human-computer interaction models to be evaluated. In step S203, when determining the evaluation information of the multi-round interaction capability of the human-computer interaction model based on each round of input information, contextual information of each round of input information, and response results in at least one conversation, the response quality of the response results output by each human-computer interaction model for the same input information is ranked. Based on the ranking results, relative evaluation information of the response quality of each human-computer interaction model can be determined as the evaluation information of the multi-round interaction capability of the human-computer interaction model. Specifically, for any round of input information in any conversation, the first server can provide a second interactive interface, which displays the input information of the current round, the contextual information of the current round of input information, and the response results of the current round of input information output by each human-computer interaction model. The second interactive interface does not display the correspondence between the response results and the human-computer interaction models. Annotators can rank the response results of the same input information displayed on the second interactive interface. The first server receives the ranking results of the responses to the same input information output by each human-computer interaction model specified in the second interaction interface; and calculates relative evaluation information of the response quality of each human-computer interaction model based on the ranking results of the responses to the same input information output by each human-computer interaction model. For example, the second interaction interface can output the input information of any round, the context of the input information of the current round, and the response results of each human-computer interaction model for the input information of the current round. The response results output by different human-computer interaction models correspond to the display areas of the rejected results. In the second interaction interface, the annotator can change the ranking of each response result by dragging the display area of ​​each response result. Alternatively, the ranking of each response result can be changed by inputting a sequence value for the response result. The position of the display area of ​​each response result in the second interaction interface automatically adjusts as the sequence value of the response result changes. Optionally, a ranking algorithm can be used to automatically sort the responses to the same input information output by each human-computer interaction model, which can significantly improve data processing efficiency.Specifically, the first server uses a pre-trained response result ranking algorithm to rank the responses to the same input information output by each human-computer interaction model. Furthermore, based on the ranking results of the responses to the same input information output by each human-computer interaction model, the first server calculates relative evaluation information for the response results output by each human-computer interaction model. The ranking algorithm can be implemented using any existing method for ranking text based on text quality, and is not specifically limited herein. For example, when calculating the relative evaluation information for the response quality of each human-computer interaction model based on the ranking results of the response results output by each human-computer interaction model, the first server can calculate the distribution of the number of times the response results output by each human-computer interaction model are ranked in each ranking, using this distribution as the relative evaluation information for each human-computer interaction model. A higher number of top rankings indicates better response quality and stronger multi-round interaction capabilities for the human-computer interaction model. By outputting the ranking distribution of the response results output by each human-computer interaction model, the relative quality of any human-computer interaction model among the multiple human-computer interaction models to be compared can be intuitively demonstrated. Optionally, based on the ranking results of the response results output by each human-computer interaction model for the same input information, the win, loss, and tie between any two human-computer interaction models can be calculated as relative evaluation information for each human-computer interaction model. For the response results of two human-computer interaction models A and B for the same input information, if the response result of human-computer interaction model A ranks ahead of the response result of human-computer interaction model B in the ranking results, human-computer interaction model A wins. If the response result of human-computer interaction model A ranks behind the response result of human-computer interaction model B in the ranking results, human-computer interaction model A loses. If the response result of human-computer interaction model A and human-computer interaction model B ties in the ranking results, the two human-computer interaction models are tied. Based on multiple rounds of input information and the ranking of the responses of two human-computer interaction models to the same input information, the number of wins, losses, and ties for any human-computer interaction model can be calculated, or the win rate of any human-computer interaction model can be calculated. By outputting the number of wins, losses, and ties, or the win rate of any human-computer interaction model, the relative quality of the human-computer interaction model among the multiple human-computer interaction models to be compared can be intuitively demonstrated.Optionally, an Elo rating is calculated for each human-computer interaction model based on the ranking of responses to the same input information. Based on multiple rounds of input information, the Elo ratings for each human-computer interaction model are calculated multiple times and averaged to provide relative evaluation information on the response quality of each human-computer interaction model. By outputting the average Elo rating of each human-computer interaction model, the relative quality of each human-computer interaction model among the multiple human-computer interaction models to be compared can be intuitively displayed. Step S409: Output evaluation information on the multi-round interaction capabilities of the human-computer interaction model. In this embodiment, after obtaining evaluation information on the multi-round interaction capability of a human-computer interaction model, the first server visually outputs the evaluation information to provide the user with the evaluation results. This information can guide the user in determining whether the human-computer interaction model should be put online. Alternatively, by comparing the evaluation information on the multi-round interaction capability of multiple human-computer interaction model versions, a high-quality human-computer interaction model version can be identified and iteratively optimized. Alternatively, by comparing the evaluation information on the multi-round interaction capability of multiple human-computer interaction models, a target human-computer interaction model with strong multi-round interaction capability can be selected as the human-computer interaction model for implementing human-computer interaction. For example, the first server can select one of the multiple human-computer interaction models as the target human-computer interaction model based on the evaluation information on the multi-round interaction capability of the multiple human-computer interaction models, and output information about the target human-computer interaction model to the end-side device. For example, the first server may select one of the multiple versions as an optimized version based on evaluation information of the multi-round interaction capabilities of multiple different versions of the human-computer interaction model, and update the optimized version of the human-computer interaction model. For example, after obtaining the evaluation information of the multi-round interaction capabilities of the human-computer interaction model, the first server may determine whether the human-computer interaction model meets online conditions based on the evaluation information of the multi-round interaction capabilities of the human-computer interaction model; and output online prompt information of the human-computer interaction model and / or the evaluation information of the multi-round interaction capabilities of the human-computer interaction model. The online prompt information indicates whether the human-computer interaction model meets the online conditions. The online conditions include a first threshold value of the evaluation information of the multi-round interaction capabilities of the human-computer interaction model. If the evaluation information of the multi-round interaction capabilities of the human-computer interaction model is greater than or equal to the first threshold value, the human-computer interaction model meets the online conditions; otherwise, the human-computer interaction model does not meet the online conditions. The first threshold value of the online conditions can be customized by the user based on the needs of a specific application scenario.The solution of this embodiment obtains pre-annotated multi-round interaction information from at least one conversation. Based on the pre-annotated multi-round interaction information from any conversation, the following interaction processing is performed on each round of input information in the conversation: the context information of the conversation is used as the context information of the current round of input information, where the context information of the conversation includes historical interaction information of the conversation; the current round of input information and the context information of the conversation are input into a human-computer interaction model, and the human-computer interaction model outputs a response result for the current round of input information; the context information of the current conversation is updated and the pre-annotated response information of the current round of input information in the context information of the updated conversation is used to replace the response result of the current round of input information, thereby obtaining the response result of each round of input information output by the human-computer interaction model. When the pre-annotated multi-round interaction information from any conversation is used to evaluate multiple different human-computer interaction models, the historical context information of the same input information in the multi-round interactions of the conversation is ensured to be completely consistent for each human-computer interaction model. The unified pre-annotated multi-round interaction information can be used to fairly and objectively evaluate the multi-round interaction capabilities of different human-computer interaction models, thereby improving the quality of multi-round interactions in human-computer interaction based on human-computer interaction models. Figure 5 is an interaction flow chart of a data processing method for human-computer interaction provided by an exemplary embodiment of the present disclosure. In this embodiment, taking a language model as an example, the process for evaluating the multi-round interaction capabilities of multiple language models is exemplified. As shown in Figure 5, the interaction process between the first server and the end-side device is as follows: Step S501: The end-side device sends a request to the first server for evaluating the multi-round interaction capabilities of multiple language models. The multiple language models can be pre-trained language models, specifically applied to fields such as natural language processing (NLP) and computer vision. Specifically, they can be applied to tasks at the intersection of NLP and computer vision, such as visual question answering (VQA), image captioning (IC), visual entailment (VE), and reference, expression, and understanding (REC), as well as tasks in the field of natural language processing, such as text-based sentiment classification and text summarization. These applications can be applied to various scenarios, including digital assistants, intelligent robots, search, online education, office software, e-commerce, and intelligent design. In step S502, the first server receives a request for evaluating the multi-round interaction capabilities of multiple language models from a client-side device. In step S503, the first server obtains pre-annotated multi-round input information from at least one conversation, as well as pre-annotated response information corresponding to each round of input information. In this step, the first server obtains the pre-annotated multi-round interaction information from at least one conversation in the same manner as in step S201. For details, refer to the relevant content of the previous embodiment and will not be repeated here.In this embodiment, the first server may also obtain pre-annotated multi-round interaction capabilities covering each session. Multi-round interaction capabilities include at least the following categories: context-dependent, context-independent, and context-consistent. Context-dependent means that the preceding context information in a multi-round interaction may be helpful (positively impact) the response to the current input information. Context-independent means that the preceding context information in a multi-round interaction has a negative impact on the response to the current input information. Context-consistent means that the response to the current input information and the preceding context information in a multi-round interaction should remain consistent. The multi-round interaction capabilities covered by each session can be manually annotated by annotators. Examples of multi-round interaction information covering different categories of multi-round interaction capabilities are shown in Table 2 and are not further described here. In step S504, the first server determines the preceding context information for each round of input information in the current session based on the pre-annotated multi-round interaction information of any session. The server then inputs the input information and the preceding context information of each round of input information in the current session into each language model. The language model then outputs a response result for each round of input information. This step is similar to the specific implementation of step S202 above. The human-computer interaction models to be evaluated in step S504 are multiple language models specified by the user. The specific implementation is described in the previous embodiments and will not be further described here. In step S505, the first server determines evaluation information for the multiple language models in various multi-round interaction capability dimensions based on the input information for each round in at least one session, the context of each input round, the response results output by each language model for each input round, and the multi-round interaction capabilities covered by each session. In this step, the first server determines interaction quality information for the human-computer interaction model in each session based on the input information for each round in at least one session, the context of the input information, and the response results. Evaluation information for the multi-round interaction capabilities of the human-computer interaction model is determined based on the interaction quality information for each human-computer interaction model in each session and the multi-round interaction capabilities covered by each session. The first server determines the interaction quality information of the human-computer interaction model in each session based on the input information of each round in at least one session, the context of the input information, and the response result. This is consistent with the specific implementation of step S507. For specific implementations, refer to the relevant content in the previous embodiment and are not further described here. Furthermore, the first server may average the interaction quality information of the human-computer interaction model in sessions covering each type of multi-round interaction capability based on the multi-round interaction capabilities covered by the at least one session and the interaction quality information of the human-computer interaction model in each session, thereby obtaining evaluation information of the human-computer interaction model in each multi-round interaction capability dimension. For specific implementations, refer to the relevant content in step S508 and are not further described here.In step S506, the first server outputs evaluation information of the multiple language models in various multi-round interaction capability dimensions to the client-side device. In step S507, the client-side device receives the evaluation information of the multiple language models in various multi-round interaction capability dimensions sent by the first server. In step S508, the client-side device outputs the evaluation information. In an optional embodiment, the client-side device may perform a weighted sum / weighted average of the evaluation information of each language model in each multi-round interaction capability dimension based on the weight coefficients corresponding to each multi-round interaction capability, obtain comprehensive evaluation information of the multi-round interaction capability of each language model, and output the comprehensive evaluation information of the multi-round interaction capability of each language model. Furthermore, the client-side device may select one of the language models as a target model based on the comprehensive evaluation information of the multi-round interaction capability of each language model, and output information about the target model. Optionally, the client-side device may select one of the language models as an optimized version based on the comprehensive evaluation information of the multi-round interaction capability of each language model, and update the optimized version of the language model. In an optional embodiment, the client device can compare the evaluation information of multiple language models in various multi-round interaction capability dimensions and output the comparison results. The comparison results are used to guide the user to select a language model with better multi-round interaction capability. In this embodiment, for multiple language models, pre-annotated multi-round interaction information from at least one conversation is obtained. The multi-round interaction information for a conversation includes multi-round input information, pre-annotated response information for each round of input information, and the multi-round interaction capabilities covered by the conversation. Based on the pre-annotated multi-round interaction information from any conversation, contextual information for each round of input information in the conversation is determined. The input information and contextual information for each round of input information in the conversation are then input into each language model, and each language model outputs a response for each round of input information. Based on the input information for each round of input information in the at least one conversation, the contextual information for each round of input information, the response results for each round of input information output by each language model, and the multi-round interaction capabilities covered by each conversation, evaluation information for the multiple language models in various multi-round interaction capability dimensions is determined. This method not only enables fair and objective evaluation of the multi-round interaction capabilities of different language models, but also enables comprehensive and accurate evaluation of the multi-round interaction capabilities of language models from multiple different multi-round interaction capability dimensions, thereby improving the quality of multi-round interaction capability evaluation of language models and, therefore, improving the quality of human-computer interaction based on language models. FIG6 is a schematic diagram of the structure of a server provided in an embodiment of the present disclosure. As shown in FIG6 , the server includes a memory 601 and a processor 602. The memory 601 is used to store computer-executable instructions and can be configured to store various other data to support operations on the server.Processor 602 is communicatively coupled to memory 601 and configured to execute computer-executable instructions stored in memory 601 to implement the technical solution of the first server in any of the aforementioned method embodiments. Its specific functions and technical effects are similar and will not be further described here. Optionally, as shown in FIG6 , the cloud server also includes other components, such as a firewall 603, a load balancer 604, a communication component 605, and a power supply component 606. FIG6 schematically illustrates only some of the components and does not imply that the cloud server only includes the components shown in FIG6 . The present disclosure also provides an end-side device, comprising a memory and a processor. The memory is configured to store computer-executable instructions and may be configured to store various other data to support operations on the end-side device. The processor is communicatively coupled to the memory and configured to execute the computer-executable instructions stored in the memory to implement the technical solution implemented by the end-side device in any of the aforementioned method embodiments. Its specific functions and technical effects are similar and will not be further described here. The present disclosure also provides a processor. FIG7 shows a schematic structural diagram of a processor used to implement the present disclosure. As shown in Figure 7, the processor 700 is configured to run a program, wherein the program, when executed by the processor, performs the method in the above-described embodiment. In the embodiment of the present disclosure, the processor 700 may execute a program for executing the data processing method for human-computer interaction. Optionally, the processor 700 may be configured to perform the following steps: obtaining pre-annotated multi-round input information from at least one conversation, and pre-annotated response information corresponding to each round of input information; determining context information for each round of input information in the conversation based on the pre-annotated multi-round input information from any conversation and the pre-annotated response information corresponding to each round of input information; inputting each round of input information and the context information of each round of input information into a human-computer interaction model; and outputting a response result for each round of input information through the human-computer interaction model; determining evaluation information for the multi-round interaction capability of the human-computer interaction model based on the input information, context information, and response result of each round of input information in at least one conversation; and outputting the evaluation information.Optionally, the processor 700 may be configured to perform the following steps: receiving a multi-round interaction capability evaluation request for multiple language models sent by a terminal-side device; obtaining pre-annotated multi-round interaction information of at least one session, wherein the multi-round interaction information includes multi-round input information and pre-annotated response information of each round of input information; determining context information of each round of input information in the session based on the pre-annotated multi-round input information of any one session and the pre-annotated response information of each round of input information, inputting the input information of each round of the session and the pre-annotated response information of each round of input information into each language model, and outputting a response result of each round of input information through each language model; determining evaluation information of the multiple language models in various multi-round interaction capability dimensions based on the input information of each round of the at least one session, the context information of each round of input information, and the response result of each round of input information output by each language model, as well as the multi-round interaction capability covered by each session; and outputting the evaluation information of the multiple language models in various multi-round interaction capability dimensions to the terminal-side device. Optionally, the processor 700 may be configured to perform the following steps: obtaining a model to be evaluated provided by a terminal-side device; obtaining, in response to a multi-round interaction capability evaluation request for the model to be evaluated sent by the terminal-side device, multi-round input information of at least one pre-annotated session, and pre-annotated response information corresponding to each round of input information; determining context information of each round of input information in the session based on the multi-round input information of any pre-annotated session and the pre-annotated response information corresponding to each round of input information, inputting each round of input information in the session and the context information of each round of input information into the model to be evaluated, and outputting a response result of each round of input information through the model to be evaluated; determining evaluation information of the multi-round interaction capability of the model to be evaluated based on the input information of each round of input information, the context information of each round of input information, and the response result of each round of input information in at least one session; and outputting the evaluation information to the terminal-side device. The present disclosure also provides a computer-readable storage medium storing computer-executable instructions. When executed by a processor, these instructions are used to implement the technical solution executed by the first server in any of the above-described method embodiments. The specific functions and technical effects achieved are not described in detail here. The present disclosure also provides a computer-readable storage medium storing computer-executable instructions. When executed by a processor, these instructions are used to implement the technical solution executed by the end-side device in any of the above-described method embodiments. The specific functions and technical effects achieved are not described in detail here. A computer storage medium may also be referred to as a computer-readable storage medium. It may include a data signal transmitted in baseband or as part of a carrier wave, which carries readable program code.This propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable storage medium can transmit, propagate, or transfer a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. The program code contained in the computer-readable storage medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical cable, radio frequency, or any suitable combination thereof. The present disclosure also provides a computer program product comprising a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the operations corresponding to the human-computer interaction data processing method in the above-described method embodiment. The present disclosure also provides a computer program that, when executed by a processor, implements the operations corresponding to the human-computer interaction data processing method in the above-described method embodiment. Optionally, the computer program, when executed by a processor, implements program code for the following steps: obtaining pre-annotated multi-round input information of at least one conversation, and pre-annotated response information corresponding to each round of input information; determining context information of each round of input information in the conversation based on the pre-annotated multi-round input information of any conversation and the pre-annotated response information corresponding to each round of input information, inputting each round of input information and the context information of each round of input information in the conversation into a human-computer interaction model, and outputting a response result of each round of input information through the human-computer interaction model; determining evaluation information of the multi-round interaction capability of the human-computer interaction model based on each round of input information, the context information of each round of input information, and the response result of each round of input information in at least one conversation; and outputting the evaluation information. Optionally, the computer program, when executed by a processor, implements program code for the following steps: receiving a multi-round interaction capability evaluation request for multiple language models sent by an end-side device; obtaining pre-annotated multi-round input information of at least one conversation, and pre-annotated response information of each round of input information; determining context information of each round of input information in the conversation based on the pre-annotated multi-round input information of any conversation and the pre-annotated response information of each round of input information, inputting the input information of each round of the conversation and the pre-annotated response information of each round of input information into each language model, and outputting a response result of each round of input information through each language model; determining evaluation information of multiple language models in various multi-round interaction capability dimensions based on the input information of each round of at least one conversation, the context information of each round of input information, and the response result of each round of input information output by each language model, as well as the multi-round interaction capability covered by each conversation; and outputting the evaluation information to the end-side device.Optionally, the computer program implements program code for the following steps when executed by a processor: obtaining a model to be evaluated provided by a server-side device; obtaining, in response to a multi-round interaction capability evaluation request for the model to be evaluated sent by a client-side device, multi-round input information of at least one pre-annotated session, and pre-annotated response information corresponding to each round of input information; determining context information of each round of input information in the session based on the multi-round input information of any pre-annotated session and the pre-annotated response information corresponding to each round of input information, inputting each round of input information in the session and the context information of each round of input information into the model to be evaluated, and outputting a response result of each round of input information through the model to be evaluated; determining evaluation information of the multi-round interaction capability of the model to be evaluated based on the input information of each round, the context information of each round of input information, and the response result of each round of input information in at least one session; and outputting the evaluation information of the multi-round interaction capability of the model to be evaluated to the client-side device. The present disclosure also provides a computer program product, comprising: a computer program stored in a readable storage medium; at least one processor of a first server can read the computer program from the readable storage medium; the at least one processor executes the computer program, causing the first server to execute the technical solution implemented by the first server in any of the aforementioned method embodiments. The specific functions and technical effects achieved are not detailed here. The present disclosure also provides a computer program product, comprising: a computer program stored in a readable storage medium; at least one processor of an end-side device can read the computer program from the readable storage medium; the at least one processor executes the computer program, causing the end-side device to execute the technical solution implemented by the end-side device in any of the aforementioned method embodiments. The specific functions and technical effects achieved are not detailed here. The present disclosure also provides a chip, comprising: a processing module and a communication interface; the processing module can execute the technical solution implemented by the first server or end-side device in the aforementioned method embodiments. Optionally, the chip further includes a storage module (e.g., a memory), the storage module being configured to store instructions, and the processing module being configured to execute the instructions stored in the storage module. Execution of the instructions stored in the storage module causes the processing module to execute the technical solution executed by the first server or the end-side device in any of the aforementioned method embodiments. The memory may be an object storage service (OSS).The aforementioned memory can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk. The aforementioned communication component is configured to facilitate wired or wireless communication between the device housing the communication component and other devices. The device housing the communication component can access a wireless network based on a communication standard, such as a mobile hotspot (WiFi), a second-generation mobile communication system (2G), a third-generation mobile communication system (3G), a fourth-generation mobile communication system (4G) / long-term evolution (LTE), a fifth-generation mobile communication system (5G), or a combination thereof. In an exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication component also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies. The aforementioned power supply component provides power to various components of the device in which the power supply component resides. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which the power supply component resides. Those skilled in the art will appreciate that embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, compact disc read-only memory (CD-ROM), optical storage, etc.) containing computer-usable program code. The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions.These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, such that the instructions, when executed by the processor of the computer or other programmable data processing device, produce means for implementing the functions specified in one or more flows in the flowchart and / or one or more blocks in the block diagram. These computer program instructions can also be stored in a computer-readable memory that directs the computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means, which implement the functions specified in one or more flows in the flowchart and / or one or more blocks in the block diagram. These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing the computer or other programmable device to execute a series of operational steps to produce a computer-implemented process, such that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more flows in the flowchart and / or one or more blocks in the block diagram. In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory. Memory may include non-permanent storage in computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media. Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can be implemented using any method or technology for information storage. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change RAM (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.It should be noted that the user information (including but not limited to user device information, user attribute information, etc.) and data (including but not limited to data used for analysis, storage, and display) involved in this disclosure are all authorized by the user or fully authorized by all parties. The collection, use, and processing of the relevant data must comply with relevant laws, regulations, and standards, and corresponding operation portals are provided for users to choose to authorize or reject. In addition, some processes described in the above embodiments and figures include multiple operations that appear in a specific order. However, it should be understood that these operations may not be executed in the order in which they appear herein or in parallel. The sequence numbers are merely used to distinguish different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the terms "first" and "second" in this document are used to distinguish different messages, devices, modules, etc., and do not indicate a sequential order, nor do they limit the "first" and "second" to different types. "Multiple" means two or more, unless otherwise specifically defined. Those skilled in the art will readily envision other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the following claims. It should be understood that the present disclosure is not limited to the precise structure described above and illustrated in the accompanying drawings, and that various modifications and variations may be made without departing from the scope of the present disclosure. The scope of the present disclosure is limited solely by the appended claims. Industrial Applicability The solution provided by the embodiments of the present disclosure can be applied in the process of data processing to obtain pre-annotated multi-round input information of at least one conversation and pre-annotated response information corresponding to each round of input information; determine the context of each round of input information in the conversation based on the pre-annotated multi-round input information of any conversation and the pre-annotated response information corresponding to each round of input information; input each round of input information in the conversation and the context of each round of input information into a human-computer interaction model, and output the response result of each round of input information through the human-computer interaction model; determine evaluation information of the multi-round interaction capability of the human-computer interaction model based on each round of input information, the context of each round of input information, and the response result of each round of input information in at least one conversation; and output the evaluation information, thereby solving the technical problem of low multi-round interaction quality of the iteratively updated human-computer interaction model.

Claims

Claims 1. A data processing method for human-computer interaction, wherein: include: Acquire multiple rounds of input information of at least one conversation that is pre-marked, and pre-marked response information corresponding to each round of the input information; Determine context information of each round of input information in the conversation according to pre-annotated multiple rounds of input information of any conversation and pre-annotated response information corresponding to each round of input information, input each round of input information in the conversation and the context information of each round of input information into a human-computer interaction model, and output a response result of each round of input information through the human-computer interaction model; Determining evaluation information of a multi-round interaction capability of the human-computer interaction model according to the input information of each round in the at least one conversation, context information of the input information of each round, and a response result of the input information of each round; Output the evaluation information.

2. The method according to claim 1, wherein determining context information of each round of input information in the conversation based on multiple rounds of input information of any pre-annotated conversation and pre-annotated response information corresponding to each round of input information, inputting each round of input information in the conversation and the context information of each round of input information into a human-computer interaction model, and outputting a response result of each round of input information through the human-computer interaction model, comprises: According to multiple rounds of input information of any pre-annotated conversation and pre-annotated response information corresponding to each round of the input information, the input information of each round in the conversation is processed in turn as follows: the context information of the conversation is used as the context information of the current round of input information, wherein the context information of the conversation includes the historical interaction information of the conversation; the input information of the current round and the context information of the conversation are input into the human-computer interaction model, and the response result of the input information of the current round is output through the human-computer interaction model; the context information of the conversation is updated, and the pre-annotated response information corresponding to the input information of the current round is used to replace the response result of the input information of the current round in the context information of the updated conversation.

3. The method according to claim 1, wherein determining context information of each round of input information in the session based on multiple rounds of input information of any pre-annotated session and pre-annotated response information corresponding to each round of input information comprises: According to the pre-annotated multiple rounds of input information of any conversation and the pre-annotated response information corresponding to each round of input information, for any round of input information in the conversation, the input information before the current round and 29 The pre-marked response information corresponding to the input information before the current round is used as the context information of the input information of the current round.

4. The method according to claim 1, wherein determining the evaluation information of the multi-round interaction capability of the human-computer interaction model according to each round of input information, context information of each round of input information and response results of each round of input information in at least one conversation comprises: Determine the interaction quality information of the human-computer interaction model in each session based on the input information in each round of at least one session, the context information of the input information in each round, and the response result of the input information in each round; determine the evaluation information of the multi-round interaction capability of the human-computer interaction model based on the interaction quality information of the human-computer interaction model in each session.

5. The method according to claim 4, after obtaining the pre-annotated input information of multiple rounds of at least one conversation and the pre-annotated response information corresponding to each round of the input information, the method further comprises: The multi-round interaction capabilities covered by each conversation are marked, wherein the multi-round interaction capabilities include at least one of the following: context-related, context-independent, and context-consistent.

6. The method according to claim 5, wherein determining the evaluation information of the multi-round interaction capability of the human-computer interaction model according to the interaction quality information of the human-computer interaction model in each session comprises: According to the multi-round interaction capabilities covered by the at least one session and the interaction quality information of the human-computer interaction model in each session, and according to the interaction quality information of the human-computer interaction model in the sessions covering various types of multi-round interaction capabilities, the evaluation information of the human-computer interaction model in various multi-round interaction capability dimensions is calculated.

7. The method according to claim 6, after obtaining the evaluation information of the human-computer interaction model in various multi-round interaction capability dimensions by calculation, the method further comprises: According to the weight coefficients corresponding to the various multi-round interaction capabilities, the evaluation information of the human-computer interaction model in the various multi-round interaction capability dimensions is weighted and integrated to obtain the comprehensive evaluation information of the multi-round interaction capability of the human-computer interaction model; According to the comprehensive evaluation information of the multi-round interaction capability of the human-computer interaction model, it is determined whether the human-computer interaction model meets the online conditions; Outputting online prompt information of the human-computer interaction model and / or evaluation information of the human-computer interaction model, wherein the online prompt information indicates whether the human-computer interaction model meets the online condition.

8. The method according to claim 7, further comprising: displaying a weight configuration interface for each type of multi-round interaction capability; 30 Obtain weight coefficients corresponding to various types of multi-round interaction capabilities configured on the weight configuration interface.

9. The method according to claim 4, wherein determining the interaction quality information of the human-computer interaction model in each session according to each round of the input information in the at least one session, context information of each round of the input information, and a response result of each round of the input information comprises: Determine evaluation information of the response results of the input information in each round according to the input information in each round of at least one conversation, the context information of the input information in each round, and the response results of the input information in each round; and weighted average the evaluation information of the response results of the input information in each round in the same conversation according to the weight coefficients corresponding to the rounds in the same conversation to obtain interaction quality information of the human-computer interaction model in this conversation.

10. The method according to claim 1, characterized in that: The determining of the evaluation information of the multi-round interaction capability of the human-computer interaction model according to the input information of each round in the at least one conversation, the context information of each round of the input information and the response result of each round of the input information includes: for any round of the input information in any conversation, outputting the input information of this round, the context information of this round of the input information and the response result of this round of the input information output by multiple human-computer interaction models through an interactive interface, and not displaying the corresponding relationship between the response result and each human-computer interaction model on the interactive interface; receiving the ranking result of the response result of the same input information output by each human-computer interaction model specified in the interactive interface; and calculating the relative evaluation information of the response quality of each human-computer interaction model according to the ranking result of the response result of the same input information output by each human-computer interaction model.

11. The method according to any one of claims 1 to 10, further comprising: According to the evaluation information of the multi-round interaction capabilities of the multiple human-computer interaction models, one of the human-computer interaction models is selected as the target model, and the information of the target model is output to the terminal device; or, according to the evaluation information of the multi-round interaction capabilities of the multiple different versions of the human-computer interaction model, one of the multiple different versions is selected as the optimized version, and the optimized version of the human-computer interaction model is updated.

12. A data processing method for human-computer interaction, wherein: Applied to the server, including: receiving a multi-round interactive capability evaluation request for multiple language models sent by a terminal device: obtaining multi-round input information of at least one conversation marked in advance, and pre-marked response information corresponding to each round of the input information; According to the pre-annotated multiple rounds of input information of any conversation and the pre-annotated response information of each round of the input information, determine the context information of each round of the input information in the conversation, input the input information of each round in the conversation and the context information of each round of the input information into each language model, and output the response result of each round of the input information through each language model; according to the input information of each round in the at least one conversation, the context information of each round of the input information and the response result of each round of the input information output by each language model, and the multi-round interaction capability covered by each conversation, determine the evaluation information of the multiple language models in various multi-round interaction capability dimensions; output the evaluation information to the terminal side device.

13. A data processing method for human-computer interaction, wherein: Applied to a server, comprising: obtaining a model to be evaluated provided by a terminal device; in response to a multi-round interaction capability evaluation request for the model to be evaluated sent by the terminal device, obtaining multi-round input information of at least one pre-annotated session and pre-annotated response information corresponding to each round of the input information; determining context information of each round of the input information in the session according to the multi-round input information of any pre-annotated session and the pre-annotated response information corresponding to each round of the input information, and inputting the input information of each round in the session and the context information of each round of the input information into the model to be evaluated, and outputting a response result of each round of the input information through the model to be evaluated; determining evaluation information of the multi-round interaction capability of the model to be evaluated according to the input information of each round in the at least one session, the context information of each round of the input information and the response result of each round of the input information; and outputting the evaluation information to the terminal device.

14. The method according to claim 13, wherein determining the context information of each round of input information in the conversation based on the pre-annotated multiple rounds of input information of any conversation and the pre-annotated response information corresponding to each round of input information comprises: According to multiple rounds of input information of any pre-marked conversation and pre-marked response information corresponding to each round of input information, for any round of input information in the conversation, the input information before this round and the pre-marked response information corresponding to the input information before this round are used as context information of this round of input information.

15. A server, wherein: include: A processor, and a memory in communication with the processor; the memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 14.

16. An electronic device, comprising: Memory and processor; The memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the method according to any one of claims 1 to 14 are implemented.

17. A computer program product, wherein: Comprising a computer program, which implements the steps of the method according to any one of claims 1 to 14 when executed by a processor.

18. A computer program product, wherein: The invention comprises a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 14 are implemented.

19. A computer program, wherein When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 14 are implemented.

20. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 14. 33

Images