Dialogue processing model training method, dialogue task processing method, and virtual character dialogue method

Abstract

Embodiments of the present disclosure provide a dialogue processing model training method, a dialogue task processing method, and a virtual character dialogue method. The dialogue processing model training method comprises: acquiring sample dialogue corpora; analyzing the sample dialogue corpora on the basis of at least two different corpus analysis dimensions, to obtain a plurality of pieces of sample dialogue data, wherein the sample dialogue data corresponding to the different corpus analysis dimensions has different levels of complexity; and on the basis of the plurality of pieces of sample dialogue data and sample response results respectively corresponding to the plurality of pieces of sample dialogue data, adjusting model parameters of a dialogue processing model to obtain a trained dialogue processing model. The at least two different corpus analysis dimensions are used to construct the sample dialogue data having different levels of complexity, such that the dialogue processing model can learn how to process dialogue data having various levels of complexity, thereby enhancing adaptability of the dialogue processing model in different application scenarios and improving robustness and flexibility of the dialogue processing model.

Description

Dialogue processing model training methods, dialogue task processing methods, and virtual character dialogue methods

[0001] This disclosure claims priority to Chinese Patent Application No. 202510399163.X, filed with the China National Intellectual Property Administration on March 28, 2025, entitled “Dialogue Processing Model Training Method, Dialogue Task Processing Method and Virtual Role Dialogue Method”, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This disclosure relates to the field of artificial intelligence technology, and in particular to a method for training a dialogue processing model, a method for processing dialogue tasks, and a method for dialogue using virtual characters. Background Technology

[0003] With the rapid development of artificial intelligence technology, dialogue processing models have been widely applied in various fields such as customer service, smart home control, and virtual assistants. Dialogue processing models aim to achieve effective communication between humans and machines through Natural Language Processing (NLP) technology, thereby improving user experience and service efficiency.

[0004] Currently, dialogue questions are typically obtained from specific platforms, and annotation personnel directly label the answers to these questions to obtain question-answer pairs. These question-answer pairs are then used to train dialogue processing models. However, because the dialogue questions on specific platforms are usually quite homogeneous, the training data for the models lacks diversity, making it difficult for the models to adapt to complex and ever-changing real-world application scenarios, thus reducing the models' robustness and flexibility. Summary of the Invention

[0005] In view of the above, embodiments of this disclosure provide a method for training a dialogue processing model. One or more embodiments of this disclosure also relate to a dialogue task processing method, a virtual character dialogue method, a dialogue processing model evaluation method, a data determination method for dialogue tasks, an information processing method based on a dialogue processing model, a task platform, a dialogue processing model training device, a dialogue task processing device, a virtual character dialogue device, a dialogue processing model evaluation device, a data determination device for dialogue tasks, an information processing device based on a dialogue processing model, a computing device, an electronic device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0006] According to a first aspect of the present disclosure, a method for training a dialogue processing model is provided, comprising:

[0007] Obtain sample dialogue data;

[0008] The sample dialogue data is analyzed based on at least two different corpus analysis dimensions to obtain multiple sample dialogue data. The complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different.

[0009] Based on multiple sample dialogue data and the corresponding sample response results, the model parameters of the dialogue processing model are adjusted to obtain the trained dialogue processing model.

[0010] According to a second aspect of the present disclosure, a dialogue task processing method is provided, comprising:

[0011] Obtain the pending problem data for the target dialogue task;

[0012] The question data to be processed is input into the dialogue processing model to obtain the target response result. The dialogue processing model is trained based on the dialogue processing model training method.

[0013] According to a third aspect of the present disclosure, a virtual character dialogue method is provided, comprising:

[0014] Obtain virtual character question data for virtual character dialogue tasks;

[0015] The virtual character's question data is input into the dialogue processing model to obtain the virtual character's response. The dialogue processing model is trained based on the dialogue processing model training method.

[0016] According to a fourth aspect of the present disclosure, a method for evaluating a dialogue processing model is provided, comprising:

[0017] Obtain model evaluation data, which includes multiple sample dialogue data and sample response results corresponding to the multiple sample dialogue data. The multiple sample dialogue data are obtained by analyzing the sample dialogue data based on at least two different corpus analysis dimensions. The complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different.

[0018] Input multiple sample dialogue data into the dialogue processing model to be evaluated to obtain the predicted response results corresponding to the multiple sample dialogue data respectively.

[0019] Based on the sample responses and predicted responses, the model evaluation results of the dialogue processing model to be evaluated are generated.

[0020] According to a fifth aspect of the present disclosure, a data determination method for a dialogue task is provided, comprising:

[0021] Obtain sample dialogue data;

[0022] The sample dialogue data is analyzed based on at least two different corpus analysis dimensions to obtain multiple sample dialogue data. The complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different.

[0023] Based on multiple sample dialogue data and the corresponding sample response results, the task data for the target dialogue task is determined.

[0024] According to a sixth aspect of the present disclosure, an information processing method based on a dialogue processing model is provided, applied to a task platform, comprising:

[0025] Receive model requests sent by terminal devices;

[0026] Based on the model request, the target dialogue processing model is determined from multiple dialogue processing models, wherein the multiple dialogue processing models are trained based on the dialogue processing model training method.

[0027] According to a seventh aspect of the present disclosure, a task platform is provided, including a request interface and a response unit;

[0028] The request interface is used to receive model requests sent by terminal devices. The model request includes at least one of the following: scene identifier of the target dialogue scene, scene input data of the target dialogue scene, and model specification parameters.

[0029] The response unit is used to determine the target dialogue processing model from multiple dialogue processing models based on the model request, wherein the multiple dialogue processing models are trained based on the dialogue processing model training method.

[0030] According to an eighth aspect of the present disclosure, a dialogue processing model training apparatus is provided, comprising:

[0031] The first acquisition module is configured to acquire sample dialogue data;

[0032] The first analysis module is configured to analyze the sample dialogue corpus according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data, wherein the complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different.

[0033] The first adjustment module is configured to adjust the model parameters of the dialogue processing model based on multiple sample dialogue data and the sample response results corresponding to the multiple sample dialogue data, so as to obtain the trained dialogue processing model.

[0034] According to a ninth aspect of the present disclosure, a dialogue task processing apparatus is provided, comprising:

[0035] The second acquisition module is configured to acquire the pending question data of the target dialogue task;

[0036] The first input module is configured to input the question data to be processed into the dialogue processing model to obtain the target response result, wherein the dialogue processing model is trained based on the dialogue processing model training method.

[0037] According to a tenth aspect of the present disclosure, a virtual character dialogue device is provided, comprising:

[0038] The third acquisition module is configured to acquire virtual character question data for virtual character dialogue tasks.

[0039] The second input module is configured to input virtual character question data into the dialogue processing model to obtain the virtual character's response result. The dialogue processing model is trained based on the dialogue processing model training method.

[0040] According to an eleventh aspect of the present disclosure, a dialogue processing model evaluation apparatus is provided, comprising:

[0041] The fourth acquisition module is configured to acquire model evaluation data, which includes multiple sample dialogue data and sample response results corresponding to the multiple sample dialogue data. The multiple sample dialogue data are obtained by analyzing the sample dialogue data based on at least two different corpus analysis dimensions. The complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different.

[0042] The third input module is configured to input multiple sample dialogue data into the dialogue processing model to be evaluated, and obtain the predicted response results corresponding to the multiple sample dialogue data respectively.

[0043] The generation module is configured to generate model evaluation results for the dialogue processing model to be evaluated based on the sample response results and the predicted response results.

[0044] According to a twelfth aspect of the present disclosure, a data determination apparatus for a dialogue task is provided, comprising:

[0045] The fifth acquisition module is configured to acquire sample dialogue data;

[0046] The second analysis module is configured to analyze the sample dialogue corpus according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data, wherein the complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different.

[0047] The first determining module is configured to determine the task data for the target dialogue task based on multiple sample dialogue data and the sample response results corresponding to the multiple sample dialogue data.

[0048] According to a thirteenth aspect of the present disclosure, an information processing apparatus based on a dialogue processing model is provided, applied to a task platform, comprising:

[0049] The receiving module is configured to receive model requests sent by the terminal device;

[0050] The second determination module is configured to determine the target dialogue processing model from multiple dialogue processing models based on a model request, wherein the multiple dialogue processing models are trained based on a dialogue processing model training method.

[0051] According to a fourteenth aspect of the present disclosure, a computing device is provided, comprising:

[0052] Memory and processor;

[0053] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the methods provided in the first, second, third, fourth, fifth, or sixth aspects described above.

[0054] According to a fifteenth aspect of the present disclosure, an electronic device is provided, comprising:

[0055] The memory and processor are connected via a bus;

[0056] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the methods provided in the first, second, third, fourth, fifth, or sixth aspects described above.

[0057] According to a sixteenth aspect of the present disclosure, a computer-readable storage medium is provided that stores a computer program / instructions which, when executed by a processor, implement the steps of the methods provided in the first, second, third, fourth, fifth, or sixth aspects described above.

[0058] According to a seventeenth aspect of the present disclosure, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the methods provided in the first, second, third, fourth, fifth, or sixth aspects described above.

[0059] This disclosure provides a method for training a dialogue processing model, comprising: acquiring sample dialogue corpora; analyzing the sample dialogue corpora according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data, wherein the complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different; adjusting the model parameters of the dialogue processing model according to the multiple sample dialogue data and the sample response results corresponding to the multiple sample dialogue data, thereby obtaining a trained dialogue processing model. By analyzing the sample dialogue corpora from at least two different corpus analysis dimensions to obtain sample dialogue data of varying complexity, the diversity and coverage of the dialogue processing model's learning are increased, enabling the dialogue processing model to learn how to handle dialogue data of various complexities, enhancing the adaptability of the dialogue processing model in different application scenarios, and improving the robustness and flexibility of the dialogue processing model. Attached Figure Description

[0060] Figure 1 is a flowchart of a dialogue processing model training method provided in an embodiment of this disclosure;

[0061] Figure 2 is a flowchart of a dialogue task processing method provided in an embodiment of this disclosure;

[0062] Figure 3 is a flowchart of a virtual character dialogue method provided in an embodiment of this disclosure;

[0063] Figure 4 is a flowchart of a dialogue processing model evaluation method provided in an embodiment of this disclosure;

[0064] Figure 5 is a flowchart of a data determination method for a dialogue task provided in an embodiment of this disclosure;

[0065] Figure 6 is a flowchart of a data determination and application method for dialogue tasks provided in an embodiment of this disclosure;

[0066] Figure 7 is a flowchart of an information processing method based on a dialogue processing model provided in an embodiment of this disclosure;

[0067] Figure 8 is a schematic diagram of the structure of a task platform provided in an embodiment of this disclosure;

[0068] Figure 9 is a schematic diagram of the structure of a dialogue processing model training device provided in an embodiment of this disclosure;

[0069] Figure 10 is a schematic diagram of the structure of a dialogue task processing device provided in an embodiment of the present disclosure;

[0070] Figure 11 is a schematic diagram of the structure of a virtual character dialogue device provided in an embodiment of this disclosure;

[0071] Figure 12 is a schematic diagram of the structure of a dialogue processing model evaluation device provided in an embodiment of the present disclosure;

[0072] Figure 13 is a schematic diagram of a data determination device for a dialogue task provided in an embodiment of the present disclosure;

[0073] Figure 14 is a schematic diagram of the structure of an information processing device based on a dialogue processing model provided in an embodiment of the present disclosure;

[0074] Figure 15 is a structural block diagram of a computing device provided in an embodiment of this disclosure;

[0075] Figure 16 is a structural block diagram of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0076] Numerous specific details are set forth in the following description to provide a full understanding of this disclosure. However, this disclosure can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this disclosure. Therefore, this disclosure is not limited to the specific implementations disclosed below.

[0077] The terminology used in one or more embodiments of this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the scope of one or more embodiments of this disclosure. The singular forms “a,” “the,” and “the” used in one or more embodiments of this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” used in one or more embodiments of this disclosure refers to and includes any or all possible combinations of one or more associated listed items. The term “at least one” in one or more embodiments of this disclosure means “one or more,” and “a plurality of” means “two or more.” The term “comprising” is an open-ended description and should be understood as “including but not limiting,” and may include other content in addition to what has been described.

[0078] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this disclosure, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this disclosure, and similarly, second may also be referred to as first. Depending on the context, the word “if” as used herein may be interpreted as “when”, “in response to a determination”, or “when…”.

[0079] Furthermore, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this disclosure are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0080] In one or more embodiments of this disclosure, a large model refers to a deep learning model with a large number of model parameters, typically containing hundreds of millions, tens of billions, hundreds of billions, trillions, or even tens of trillions of model parameters. A large model can also be called a foundation model. It is pre-trained using large-scale unlabeled corpora to produce a pre-trained model with hundreds of millions of parameters. Such models can adapt to a wide range of downstream tasks and have good generalization ability. Examples include Large Language Models (LLMs) and multi-modal pre-training models.

[0081] In practical applications, large models only require a small number of samples to fine-tune the pre-trained model before they can be applied to different tasks. Large models can be widely used in fields such as natural language processing 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. The main application scenarios of large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, and intelligent design.

[0082] First, the terms and concepts involved in one or more embodiments of this disclosure will be explained.

[0083] World knowledge: refers to the understanding of information related to the world setting, historical events, plot development, etc., of characters within the context of intellectual property (IP) such as games.

[0084] Deep self-attention (Transformer) models are network structures based on multi-head self-attention mechanisms, primarily used for processing sequential data. A Transformer model consists of repeatedly stacked encoder and decoder units. This design allows the Transformer to efficiently learn long-term dependencies, making it suitable for various natural language processing tasks, including machine translation, text summarization, and question answering systems.

[0085] Bidirectional Encoder Representations from Transformers (BERT) is a pre-trained NLP model. By learning from large amounts of unlabeled text data, this model can capture deep semantic information from text and achieves significant performance improvements on numerous NLP tasks.

[0086] The Text-to-Text Transfer Transformer (T5) model is an NLP model. The main characteristic of the T5 model is that it unifies all NLP tasks into a text-to-text format, meaning both the input and output are text sequences. This design makes the model more adaptable to various tasks, such as translation, question answering, and summarization.

[0087] Supervised fine-tuning (SFT) is a method of further training a model based on a pre-trained model. In this method, the model is trained on a dataset containing pairs of inputs and desired outputs so that it learns how to generate responses closer to human-level performance. Supervised fine-tuning is often used to adapt a model to task-specific or domain-specific data, thereby improving its performance on those tasks.

[0088] Multi-dimensional evaluation refers to the process of quantitatively or qualitatively evaluating a dialogue processing model or its generated results from multiple perspectives (such as "points", "lines", "surfaces", and "timelines"), including aspects such as content accuracy, role consistency, and interaction rationality.

[0089] The F1 score is a metric used in information retrieval and machine learning to evaluate the performance of classification models. It can be defined as the harmonic mean of precision and recall.

[0090] L1 norm loss function: a loss function used in machine learning and statistics for regression problems.

[0091] This disclosure provides a method for training a dialogue processing model, and also relates to a dialogue task processing method, a virtual character dialogue method, a dialogue processing model evaluation method, a data determination method for dialogue tasks, an information processing method based on a dialogue processing model, a task platform, a dialogue processing model training device, a dialogue task processing device, a virtual character dialogue device, a dialogue processing model evaluation device, a data determination device for dialogue tasks, an information processing device based on a dialogue processing model, a computing device, an electronic device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0092] Referring to Figure 1, Figure 1 shows a flowchart of a dialogue processing model training method provided in an embodiment of this disclosure, which specifically includes the following steps:

[0093] Step 102: Obtain sample dialogue data.

[0094] It should be noted that the sample dialogue corpus refers to the original dialogue data that has not yet been manually or automatically annotated. Taking a dialogue task in a virtual character dialogue scenario as an example, the sample dialogue corpus can be data such as original text, dialogue records, and plot descriptions related to the virtual character. The sample dialogue corpus can include data of different modalities, such as text data, image data, video data, audio data, etc. The sample dialogue corpus can also be corpus from different data sources, such as sample dialogue corpus from knowledge documents (such as encyclopedia documents), sample dialogue corpus from film, animation, original IP works and their derivative content, thereby ensuring that the sample dialogue corpus has broad applicability and generalization ability, effectively reducing data bias.

[0095] In practical applications, there are various ways to obtain sample dialogue data, and the specific method chosen depends on the actual situation. This disclosure does not impose any limitations on these methods. In one possible implementation, sample dialogue data sent by a user through a client can be received. In another possible implementation, sample dialogue data can be read from other data acquisition devices or databases.

[0096] Step 104: Analyze the sample dialogue data according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data. The complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different.

[0097] It should be noted that sample dialogue data refers to dialogue data obtained by analyzing sample dialogue corpora through specific corpus analysis dimensions. For example, a complex sample dialogue may involve multiple topic shifts and a long dialogue chain, while a simple sample dialogue may only involve a single topic and a brief exchange. Sample dialogue data can be understood as sample dialogue questions based on sample dialogue corpora. Sample dialogue data can be of different types, such as questions in open-ended questions, multiple-choice questions, true / false questions, etc.

[0098] Corpus analysis dimensions refer to the standards or perspectives used to analyze sample dialogue corpora. At least two different corpus analysis dimensions result in analyses of sample dialogue corpora with varying focuses, emphases, perspectives, or depths. Examples of corpus analysis dimensions include role analysis, plot analysis, content analysis, context analysis, and sentiment analysis. By using at least two different corpus analysis dimensions, sample dialogue corpora can be comprehensively understood and analyzed from multiple angles, yielding sample dialogue data of varying complexity and diversity. This allows the sample dialogue data to more profoundly reflect the overall character and detailed features of the corpus. Each corpus analysis dimension helps reveal different aspects of the sample dialogue corpus or different dimensions of the same aspect, thus obtaining sample dialogue data that covers various aspects of the dialogue roles and world knowledge within the sample dialogue data. This results in more comprehensive and richer sample dialogue data.

[0099] Content analysis focuses on a deep dive into the themes, information content, and expression methods of the sample dialogue corpus. It considers not only "what was said" but also "how it was said," including language style and narrative techniques. For example, the sample dialogue data obtained by analyzing the sample dialogue corpus according to content analysis can be used to summarize the main points of an article or to evaluate the effectiveness of different narrative techniques.

[0100] Contextual analysis focuses on studying the environment in which sample dialogues were generated and its impact on the meaning of those dialogues. Context can be historical, cultural, social, or even personal background factors. Contextual analysis can help understand the deeper meaning of sample dialogues and the author's intentions. For example, the sample dialogue data obtained by analyzing sample dialogues according to the contextual analysis dimension can be used to explore the information that a certain text may convey in its specific historical context or to analyze the impact of cultural differences on text interpretation.

[0101] Sentiment analysis focuses on identifying and classifying the emotional tendencies in sample dialogue corpora, such as positive, negative, or neutral emotions. Sentiment analysis can be further refined to specific emotion types, such as joy, anger, and sadness, and examine the distribution and changes of these emotions in the sample dialogue corpora. The sample dialogue data obtained by analyzing sample dialogue corpora based on sentiment analysis dimensions can be used to identify the emotional tone of a dialogue or to analyze the reasons for changes in characters' emotions and their impact on the plot.

[0102] In practical applications, there are various ways to analyze sample dialogue corpora based on at least two different corpus analysis dimensions to obtain multiple sample dialogue data. The specific method chosen depends on the actual situation, and this disclosure does not impose any limitations on this approach. In the first possible implementation, the sample dialogue corpora can be directly analyzed based on at least two different corpus analysis dimensions to obtain multiple sample dialogue data, such as using text analysis tools or text analysis models (e.g., pre-trained large models, deep learning models trained based on different corpus analysis dimensions and corresponding dialogue data). In the second possible implementation, to make the sample dialogue data more standardized, reference dialogue data that meets the dialogue requirements can be obtained. During the analysis of the sample dialogue corpora based on at least two different corpus analysis dimensions, multiple sample dialogue data can be constructed using the reference dialogue data as an example. In the third possible implementation, the sample dialogue corpora can be processed before analysis (e.g., cleaning, track splitting, segmentation, alignment, etc.). The processed sample dialogue corpora can then be analyzed based on at least two different corpus analysis dimensions to obtain multiple sample dialogue data.

[0103] In one optional embodiment of this disclosure, the above-described analysis of sample dialogue data based on at least two different corpus analysis dimensions to obtain multiple sample dialogue data may include the following steps:

[0104] Obtain reference dialogue data from at least two different dimensions of corpus analysis;

[0105] Based on at least two different corpus analysis dimensions and reference dialogue data, the sample dialogue corpus is analyzed to obtain multiple sample dialogue data.

[0106] It should be noted that reference dialogue data refers to dialogue data used as a standard or example when analyzing sample dialogue corpora. Reference dialogue data typically has a clear format and analytical direction. Reference dialogue data can be real dialogue records or simulated dialogue data.

[0107] In practical applications, there are multiple ways to obtain reference dialogue data with at least two different corpus analysis dimensions, and the specific method chosen depends on the actual situation. This disclosure does not impose any limitations on this approach. One possible implementation of this disclosure is to read reference dialogue data that meets dialogue requirements and different corpus analysis dimensions from external dialogue resources. Another possible implementation of this disclosure is that reference dialogue data with different corpus analysis dimensions can be generated by professionals through simulation.

[0108] The solution of this disclosure, by introducing high-quality reference dialogue data as a reference during the analysis of sample dialogue corpora according to at least two different corpus analysis dimensions, makes the sample dialogue data more accurate.

[0109] In one optional embodiment of this disclosure, before analyzing the sample dialogue corpus according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data, the following steps may be included:

[0110] Modality recognition is performed on the sample dialogue corpus to obtain the corpus modality of the sample dialogue corpus;

[0111] Based on the corpus modality, determine the processing strategy for the sample dialogue corpus, and process the sample dialogue corpus based on the processing strategy to obtain the processed sample dialogue corpus.

[0112] The sample dialogue corpus was analyzed based on at least two different corpus analysis dimensions to obtain multiple sample dialogue data, including:

[0113] The processed sample dialogue corpus is analyzed based on at least two different corpus analysis dimensions to obtain multiple sample dialogue data.

[0114] It should be noted that modality recognition refers to the process of determining the modality of a sample dialogue corpus. The corpus modality refers to the form or type of the sample dialogue corpus determined after modality recognition. Corpus modalities include, but are not limited to, text, speech, images, or video. For example, if the sample dialogue corpus is recorded as speech, its corpus modality is speech. Processing strategies for sample dialogue corpora include at least one of cleaning, track splitting, segmentation, and alignment. A processing strategy refers to a plan or scheme for processing the sample dialogue corpus. Different processing strategies typically correspond to different modalities of data; in other words, different processing strategies may be applicable to different modalities of data. Therefore, when processing sample dialogue corpora, a suitable processing strategy can be determined based on the corpus modality. Based on this processing strategy, the sample dialogue corpus is processed to obtain the processed sample dialogue corpus. Cleaning refers to removing unnecessary noise, redundant words, or other unwanted elements from the sample dialogue corpus. Cleaning is applicable to text, audio, images, and video. Track splitting is the process of separating different mixed sound sources into independent tracks. For example, in music recording, vocals and instrument sounds can be separated into different tracks. Track splitting applies to both audio and video. Segmentation refers to the process of dividing a sample dialogue corpus into smaller parts or segments. These segments are usually based on natural breakpoints (such as sentence ends, musical paragraph changes, etc.). Segmentation applies to text, audio, images, and video. Alignment refers to the process of precisely matching sample dialogue corpora of different modalities along a timeline. For example, in speech, synchronizing the generated text with the speaker's actual pronunciation time. Alignment applies to text, audio, images, and video.

[0115] In practical applications, there are various ways to perform modality recognition on sample dialogue corpora to obtain their corpus modalities. The specific method chosen depends on the actual situation, and this disclosure does not impose any limitations on these methods. One possible implementation of this disclosure involves obtaining descriptive information (including corpus modality descriptions) from the sample dialogue corpora and parsing the corpus modality from this information. Another possible implementation involves inputting the sample dialogue corpora into a modality recognition model to obtain their corpus modality. This modality recognition model can be a pre-trained large model or a deep learning model trained based on the training dialogue corpora and their modality labels.

[0116] By applying the solutions of this disclosure, the sample dialogue data is cleaned, segmented, divided, and aligned, thereby improving the quality of the sample dialogue data, facilitating further analysis to obtain high-quality sample dialogue data, and improving the efficiency of the sample dialogue data analysis process and the quality of the sample dialogue data.

[0117] In one optional embodiment of this disclosure, taking at least two different corpus analysis dimensions, including role analysis and plot analysis, as an example, the above-mentioned analysis of sample dialogue corpora based on at least two different corpus analysis dimensions to obtain multiple sample dialogue data may include the following steps:

[0118] Based on the role analysis dimensions, we perform role analysis on the sample dialogue corpus to obtain sample role dialogue data.

[0119] Based on the plot analysis dimension, plot analysis is performed on the sample dialogue corpus to obtain sample plot dialogue data. Among them, the complexity of the sample character dialogue data is less than that of the sample plot dialogue data.

[0120] It's important to note that the character analysis dimension focuses on a deep dive into the attributes of the characters themselves (such as personality, behavior, and motivation) and the interactions between them in the sample dialogue corpus. For example, sample dialogue data obtained by analyzing the sample dialogue corpus according to the character analysis dimension can be used to determine a character's personality or explore the relationships between characters. Sample character dialogue data refers to dialogue questions obtained by analyzing character information in the sample dialogue corpus. The plot analysis dimension focuses on a deep dive into the overall narrative background of the story in the sample dialogue corpus and the development of the plot over time. Sample dialogue data obtained through plot analysis can help understand how the story is constructed and how various plot segments are interconnected and supportive. For example, sample dialogue data obtained by analyzing the sample dialogue corpus according to the plot analysis dimension can be questions about the main storyline or predictions of how the story will develop. Sample character dialogue data refers to data that asks questions about character-related information in the sample dialogue corpus. Sample plot dialogue data refers to data that asks questions about plot-related information in the sample dialogue corpus.

[0121] For example, the sample dialogue corpus is a novel. Sample character dialogue data could be, "In the world of this novel, who controls a certain city: A. Zhang San; B. Li Si; C. Wang Wu; D. Zhao Liu?" Sample plot dialogue data could be, "Please explain in detail the main content of the 'Battle of [specific event]' in this novel and its impact on the plot."

[0122] In practical applications, the implementation of "analyzing the sample dialogue corpus according to the role analysis dimension to obtain sample role dialogue data; and analyzing the sample dialogue corpus according to the plot analysis dimension to obtain sample plot dialogue data" can refer to the three implementation methods described above: "analyzing the sample dialogue corpus according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data". This disclosure embodiment will not elaborate further.

[0123] By applying the scheme of this disclosure embodiment, the sample dialogue corpus is subjected to role analysis and plot analysis according to the role analysis dimension and the plot analysis dimension, and sample role dialogue data and sample plot dialogue data with different complexities are obtained. This realizes the construction of multi-dimensional and multi-complexity sample dialogue data, and increases the diversity and coverage of the dialogue processing model learning.

[0124] In one optional embodiment of this disclosure, the role analysis dimension can be further subdivided based on the complexity of role-related information. Taking the role analysis dimension as including a first role analysis dimension and a second role analysis dimension as an example, the above-mentioned role analysis of the sample dialogue corpus based on the role analysis dimension to obtain sample role dialogue data may include the following steps:

[0125] Based on the first role analysis dimension, the role attribute information in the sample dialogue corpus is analyzed to obtain the first sample role dialogue data;

[0126] Based on the second role analysis dimension, the role interaction information in the sample dialogue corpus is analyzed to obtain the second sample role dialogue data. The complexity of the first sample role dialogue data is less than that of the second sample role dialogue data.

[0127] It should be noted that role attribute information refers to the basic characteristics and background information of each role in the sample dialogue corpus. Role attribute information helps to understand the identity of the roles in the sample dialogue corpus (such as customers, experts, etc.), the basic socio-statistical characteristics of the roles (such as occupation, preferences), and the basic positioning of the roles in the dialogue. Role interaction information refers to how different roles in the sample dialogue corpus interact, their interaction patterns, and the emotional tendencies and social dynamics behind these interactions. The first role analysis dimension is used to analyze the attribute information of the roles themselves in the sample dialogue corpus, aiming to understand the background of the dialogue participants. Therefore, the first role analysis dimension can be understood as a "point" analysis dimension, or also called the role attribute analysis dimension. The first sample role dialogue data focuses on examining single-point factual knowledge of roles or the world. The first sample role dialogue data can be multiple-choice questions, open-ended questions, personality classification questions, etc., targeting single-point knowledge in the sample dialogue corpus. The second role analysis dimension is used to analyze the interaction information between roles in the sample dialogue corpus, aiming to reveal the dynamic structure and interaction characteristics of the dialogue. Therefore, the second role analysis dimension can be understood as a "line" analysis dimension, or also called the role interaction analysis dimension. The second sample of character dialogue data focuses on examining the interaction logic between characters. This second sample of character dialogue data can include question-and-answer questions, dialogue generation questions, etc., that examine the interaction relationships between characters.

[0128] For example, the sample dialogue corpus is a novel. The first sample character dialogue data could be "In the world view of this novel, who is the controller of a certain city: A. Zhang San; B. Li Si; C. Wang Wu; D. Zhao Liu". The second sample character dialogue data could be "Please briefly describe the relationship between Zhang San and Li Si in this novel, and the main reason that led to their falling out".

[0129] By applying the solution of this disclosure embodiment, single-point knowledge and interaction logic analysis can be performed on the sample dialogue corpus through the first role analysis dimension and the second role analysis dimension, sample dialogue data with different complexities but complementary to each other can be obtained, which increases the diversity and coverage of the dialogue processing model learning.

[0130] In one optional embodiment of this disclosure, the plot analysis dimension can be further subdivided based on the complexity of plot-related information, so that the plot analysis dimension includes a first plot analysis dimension and a second plot analysis dimension; the above-mentioned plot analysis of the sample dialogue corpus based on the plot analysis dimension to obtain sample plot dialogue data may include the following steps:

[0131] Based on the first plot analysis dimension, the plot background information in the sample dialogue corpus is analyzed to obtain the first sample plot dialogue data.

[0132] Based on the second plot analysis dimension, the plot progress information in the sample dialogue corpus is analyzed to obtain the second sample plot dialogue data. The complexity of the first sample plot dialogue data is less than that of the second sample plot dialogue data.

[0133] It should be noted that plot background information refers to the overall environment or context in which the dialogues take place in the sample dialogue corpus, including but not limited to elements such as time, place, and cultural background, and how these elements influence the characters' behavior and the development of the plot. Plot progression information refers to the chronological order and logical relationships of the various events in the sample dialogue corpus, emphasizing the timeline between events and how they drive the story forward.

[0134] The first plot analysis dimension refers to the process of analyzing the background information of the plot in the sample dialogue data, focusing on how the characters collectively weave together a coherent story or situation. Therefore, the first plot analysis dimension can be understood as a "surface" analysis dimension, or also as a plot background analysis dimension. The first sample plot dialogue data focuses on examining the overall narrative and plot coherence of the dialogue. The first sample plot dialogue data can include questions involving the performance of the characters and the generation of the plot. The second plot analysis dimension refers to the process of analyzing the plot progression information in the sample dialogue data, focusing on tracing the development of the story according to the timeline of events. Therefore, the second plot analysis dimension can be understood as a "timeline" analysis dimension, or also as a plot progression analysis dimension. The second sample plot dialogue data focuses on examining information related to the plot progression, such as the chronological order of events, causal relationships, and their consistency with the world background. The second sample plot dialogue data can include questions designed based on the chronological order of events. Since the first plot analysis dimension focuses on the coherence of the overall plot, while the second plot analysis dimension needs to delve into the time sequence and logical relationships of specific events, requiring a more detailed analysis of how each event occurs and affects subsequent developments, the complexity of the first sample plot dialogue data is less than that of the second sample plot dialogue data.

[0135] For example, the sample dialogue corpus is a novel. The first sample plot dialogue data could be "Please explain in detail the main content of 'Battle X' in the novel and its impact on the plot." The second sample plot dialogue data could be "In the novel, before Zhang San learns that he is a wizard, which of the following events occurs first? A. Zhang San makes the glass disappear at the zoo, releasing the snake; B. Zhang San receives an acceptance letter to magic school; C. Li Si tells Zhang San his identity; D. Zhang San's hair is cut short but returns to its original shape the next day."

[0136] By applying the scheme of this disclosure embodiment, analyzing the plot narrative and plot timeline of the sample dialogue data through the first plot analysis dimension and the second plot analysis dimension, sample dialogue data with different complexities but complementary to each other can be obtained, which increases the diversity and coverage of the dialogue processing model learning.

[0137] Step 106: Based on the multiple sample dialogue data and the sample response results corresponding to the multiple sample dialogue data, adjust the model parameters of the dialogue processing model to obtain the trained dialogue processing model.

[0138] It's important to note that sample response results refer to the ideal response or correct answer expected from the sample dialogue data. Sample response results can be understood as a supervision signal during the training of the dialogue processing model, serving as the model's dialogue objective. Through sample response results, the dialogue processing model can learn how to generate appropriate responses based on the input sample dialogue data. A dialogue processing model is a model built using machine learning or deep learning techniques, designed to understand input dialogue data and generate responses. Dialogue processing models can be large models, such as BERT, T5, etc. Model parameters refer to the adjustable parameters used in the dialogue processing model. These parameters are continuously optimized during training to improve the model's prediction accuracy for new input data.

[0139] In practical applications, the process of adjusting the model parameters of a dialogue processing model based on multiple sample dialogue data and their corresponding sample response results can be considered supervised fine-tuning. Specifically, sample dialogue data can be input into the dialogue processing model to obtain the predicted response results output by the model. Based on the sample response results and the predicted response results output by the model, the dialogue loss value is calculated. The model parameters are then adjusted based on the dialogue loss value until a preset stopping condition is reached, at which point training stops, resulting in a trained dialogue processing model. Many functions can be used to calculate the dialogue loss value, such as cross-entropy loss, L1 norm loss, maximum loss, mean squared error loss, and logarithmic loss. The specific function chosen depends on the actual situation, and this embodiment does not impose any limitations on this. The preset stopping condition includes, but is not limited to, the dialogue loss value being less than or equal to a preset threshold, or the number of iterations reaching a preset number of iterations. The preset threshold and the preset number of iterations are selected based on the actual situation, and this embodiment does not impose any limitations on these.

[0140] The solution of this disclosure increases the diversity and coverage of the dialogue processing model's learning by analyzing sample dialogue data from at least two different corpus analysis dimensions. This enables the dialogue processing model to learn how to process dialogue data of various complexities, enhances the adaptability of the dialogue processing model in different application scenarios, and improves the robustness, flexibility, realism, and coherence of the dialogue processing model.

[0141] In practical applications, before obtaining the trained dialogue processing model by adjusting the model parameters based on multiple sample dialogue data and their corresponding sample response results, it is also possible to obtain the sample response results corresponding to the multiple sample dialogue data. There are various ways to obtain the sample response results of the sample dialogue data, and the specific method chosen depends on the actual situation; this disclosure does not limit this approach. In one possible implementation, the sample response results corresponding to multiple sample dialogue data sent by an expert through a client can be received. In another possible implementation, a data annotation platform can be used to annotate the response results of the sample dialogue data to obtain the sample response results corresponding to the multiple sample dialogue data.

[0142] In one optional embodiment of this disclosure, the analyzed sample dialogue data can be further tested (diversity detection, quality detection, generalization detection, etc.), and the model parameters of the dialogue processing model can be adjusted using the tested sample dialogue data to ensure the quality of the sample dialogue data. In another optional embodiment of this disclosure, the number of sample dialogue data can be reasonably allocated according to the IP works, the number of characters, and the distribution of plots (for example, a total of 480 sample dialogue data, of which IP character knowledge, IP world knowledge, and dialogue evaluation each account for a certain proportion), to ensure that the data coverage is broad and representative.

[0143] In one optional embodiment of this disclosure, before adjusting the model parameters of the dialogue processing model based on multiple sample dialogue data and the sample response results corresponding to the multiple sample dialogue data to obtain the trained dialogue processing model, the following steps may be included:

[0144] The data annotation platform is called to annotate the response results of multiple sample dialogue data, thereby obtaining the sample response results corresponding to each of the multiple sample dialogue data.

[0145] It's important to note that a data annotation platform refers to a tool or system capable of adding response tags to sample dialogue data. A data annotation platform can be a crowdsourcing platform, which is an online service platform that uses internet technology to distribute specific tasks to a broad, unspecified group of people (i.e., a "population" or "masses") to complete them. For example, a data annotation platform can provide an interface for users to input or select correct sample responses to sample dialogue data, support collaborative annotation by multiple users, and ensure consistent annotation quality.

[0146] By applying the solution of this disclosure embodiment, and by calling a data annotation platform to annotate the response results of multiple sample dialogue data, the sample response results corresponding to each sample dialogue data can be obtained, ensuring the diversity, generalization and high quality of the sample response results, and providing accurate target output for the training of subsequent dialogue processing models.

[0147] Referring to Figure 2, which shows a flowchart of a dialogue task processing method according to an embodiment of the present disclosure, the method specifically includes the following steps:

[0148] Step 202: Obtain the pending problem data for the target dialogue task.

[0149] Step 204: Input the question data to be processed into the dialogue processing model to obtain the target response result. The dialogue processing model is trained based on the dialogue processing model training method.

[0150] It's important to clarify that the target dialogue task refers to the task that needs to be solved within the dialogue application scenario. The target dialogue task can be a dialogue task in different scenarios, such as dialogue tasks in virtual character dialogue scenarios, role-playing dialogue scenarios, emotional dialogue scenarios, customer service scenarios, and so on. The question data to be processed refers to the raw question data input into the dialogue processing model. This question data can be data of different modalities, such as text, images, audio, video, etc. The dialogue processing model is a model that can understand the user's intent based on the input data and generate an appropriate response based on that intent. The target response result is the output of the dialogue processing model after processing the question data to be processed; that is, the response to the user's input. The target response result can be a text message, a voice reply, or an instruction to perform a certain action, aimed at answering the question data to be processed.

[0151] In practical applications, there are various ways to obtain pending problem data for a target dialogue task, and the specific method chosen depends on the actual situation. This disclosure does not impose any limitations on these methods. In one possible implementation, pending problem data for a target dialogue task can be received from a user via a client. In another possible implementation, pending problem data for a target dialogue task can be read from other data acquisition devices or databases.

[0152] Furthermore, there are multiple ways to input the question data to be processed into the dialogue processing model to obtain the target response result, and the specific method selected depends on the actual situation. This disclosure does not impose any limitations on this approach. In one possible implementation, the question data to be processed can be directly input into the dialogue processing model to obtain the target response result. In another possible implementation, to enable the dialogue processing model to better understand the question data to be processed, the question data to be processed and its dialogue context can be input into the dialogue processing model to obtain the target response result.

[0153] By applying the solution of this disclosure embodiment, since the dialogue processing model trained based on the dialogue processing model training method has learned how to process dialogue data of various complexities and has a very strong adaptability in different dialogue scenarios, the dialogue processing model is used to process the problem data to be processed, ensuring the normal progress of dialogue task processing, thereby obtaining a highly accurate target response result.

[0154] In one optional embodiment of this disclosure, after inputting the question data to be processed into the dialogue processing model and obtaining the target response result, the following steps may be further included:

[0155] The quality of the target response results is evaluated to obtain the quality indicators of the target response results.

[0156] If the quality indicators do not meet the quality inspection conditions, the parameters of the dialogue processing model are adjusted to obtain the adjusted dialogue processing model.

[0157] It should be noted that quality inspection refers to the process of checking whether the target response meets quality standards or requirements. Quality indicators are metrics used to measure the quality of the target response. Quality indicators can be specific numerical values, such as 80 points, or quality levels, such as high quality or low quality. Quality inspection conditions refer to a pre-defined set of rules used to determine whether the quality indicators have reached an acceptable standard. For example, a quality indicator exceeding a quality indicator threshold (e.g., 90 points). The adjusted dialogue processing model refers to a new version of the dialogue processing model obtained after one or more rounds of parameter adjustments. The adjusted dialogue processing model can produce higher-quality target responses under the same or similar inputs. If the quality indicators do not meet the quality inspection conditions, it means the target response does not meet the requirements, and the dialogue processing model is considered to have poor dialogue processing capabilities. In this case, the model parameters of the dialogue processing model can be adjusted. If the quality indicators meet the quality inspection conditions, it means the target response meets the requirements, and the dialogue processing model is considered to have strong dialogue processing capabilities, and there is no need to adjust the model parameters. The implementation method of "adjusting the parameters of the dialogue processing model to obtain the adjusted dialogue processing model" can be referred to the dialogue processing model training method shown in Figure 1, and will not be described in detail in this embodiment.

[0158] In practical applications, there are various ways to perform quality checks on the target response and obtain its quality indicators. The specific method chosen depends on the actual situation, and this disclosure does not impose any limitations on this approach. One possible implementation of this disclosure involves performing text matching or semantic similarity calculation between the target response and sample dialogue corpus to obtain the quality indicators. Another possible implementation involves performing quality checks on the target response from multiple quality inspection dimensions (such as relevance, accuracy, fluency, etc.) to obtain the quality indicators.

[0159] By applying the solution of this disclosure embodiment, the shortcomings of the answers generated by the current dialogue processing model are identified through quality detection, and the model parameters of the dialogue processing model are adjusted accordingly. This enables the dialogue processing model to learn how to more accurately understand and respond to user questions, thereby improving the accuracy and reliability of the dialogue processing model.

[0160] In one optional embodiment of this disclosure, after inputting the question data to be processed into the dialogue processing model and obtaining the target response result, the following steps may be further included:

[0161] Receive result feedback information sent by the client, wherein the result feedback information is the information provided by the client in response to the target response based on the dialogue requirements;

[0162] Based on the feedback information, the parameters of the dialogue processing model are adjusted to obtain the adjusted dialogue processing model.

[0163] It's important to note that feedback information refers to the feedback provided by the client after evaluating the target response based on the dialogue requirements. Feedback typically includes opinions on the accuracy, relevance, and satisfaction of the answer. For example, if a user feels the target response is insufficiently detailed or completely deviates from the core of the problem data, they can relay this information to the server through the client. Dialogue requirements refer to the specific purpose or problem the user hopes to achieve through interaction with the dialogue processing model. If the client is satisfied with the target response based on the feedback information, no parameter adjustments are made to the dialogue processing model; if the client is dissatisfied with the target response based on the feedback information, the parameters are adjusted to obtain the adjusted dialogue processing model.

[0164] In practical applications, there are various ways to adjust the parameters of the dialogue processing model based on the feedback information to obtain the adjusted dialogue processing model. The specific method chosen depends on the actual situation, and this disclosure does not limit this approach. One possible implementation of this disclosure involves sending a pre-set optimization prompt message to the user, such as, "I'm very sorry for providing you with inaccurate information. Please point out the specific inaccuracy or provide the correct answer to the relevant question, and I will correct and optimize my answer as soon as possible to better serve you." The client then receives model optimization data in response to the optimization prompt message, and the parameters of the dialogue processing model are adjusted using this data to obtain the adjusted dialogue processing model. Another possible implementation involves reading model optimization data based on the feedback information from other data acquisition devices or databases, and using this data to adjust the parameters of the dialogue processing model to obtain the adjusted dialogue processing model. The implementation of "adjusting the parameters of the dialogue processing model using model optimization data to obtain the adjusted dialogue processing model" can be referred to the dialogue processing model training method shown in Figure 1, and will not be elaborated upon further in this disclosure.

[0165] By applying the solution of this disclosure embodiment and combining feedback information from human responses to the target, the dialogue processing model is optimized, a feedback loop is established to support iterative optimization of the model, and the accuracy of the dialogue processing model is improved.

[0166] The following description, in conjunction with Figure 3, uses the application of the dialogue task processing method provided in this disclosure in a virtual character dialogue scenario as an example to further illustrate the dialogue task processing method. Figure 3 shows a flowchart of a virtual character dialogue method provided in an embodiment of this disclosure, specifically including the following steps:

[0167] Step 302: Obtain the virtual character question data for the virtual character dialogue task.

[0168] Step 304: Input the virtual character's question data into the dialogue processing model to obtain the virtual character's response. The dialogue processing model is trained based on the dialogue processing model training method.

[0169] It should be noted that the virtual character dialogue task refers to the target dialogue task within a virtual character dialogue scenario. Virtual character question data refers to the question data prepared for processing in order to complete the virtual character dialogue task. The virtual character response result refers to the answer or output generated by the dialogue processing model after the virtual character question data is input. The virtual character response result conforms to the virtual character's identity characteristics, the dialogue background, and the contextual requirements of the current dialogue.

[0170] In practical applications, the implementation methods of steps 302 and 304 are the same as those of steps 202 and 204, and will not be described again in this embodiment.

[0171] By applying the solution of this disclosure embodiment, since the dialogue processing model trained based on the dialogue processing model training method has learned how to process dialogue data of various complexities and has a very strong adaptability in different dialogue scenarios, the dialogue processing model is used to process the virtual character question data, ensuring the normal progress of the virtual character dialogue task processing, thereby obtaining highly accurate virtual character response results.

[0172] With the rapid development of dialogue generation technology, accurately evaluating the performance of dialogue processing models in terms of character portrayal, plot coherence, and world-building has become a pressing issue. Traditional evaluation methods for dialogue processing models often focus on simple objective metrics, such as the accuracy of multiple-choice questions, while neglecting key dimensions such as character personality, inter-character interaction, plot coherence, and world-building. Alternatively, they rely solely on heuristic rules such as text length and keyword matching, lacking a comprehensive assessment of multimodal factors including semantics, sentiment, and character consistency, making them ill-suited for complex dialogue scenarios. Meanwhile, while subjective human evaluation can reflect some details, it suffers from low efficiency and inconsistent standards, failing to comprehensively measure the actual performance of dialogue generation models in character world modeling.

[0173] Based on this, this disclosure proposes a dialogue processing model evaluation scheme, constructing multi-dimensional model evaluation data covering role knowledge and world knowledge, automatically evaluating the role and world knowledge of the dialogue processing model, and achieving a comprehensive quantitative evaluation of the dialogue processing model. This approach considers both the accuracy of objective data and the feedback of subjective experience, thereby improving model iteration efficiency. Referring to Figure 4, Figure 4 shows a flowchart of a dialogue processing model evaluation method provided by an embodiment of this disclosure, specifically including the following steps:

[0174] Step 402: Obtain model evaluation data, which includes multiple sample dialogue data and sample response results corresponding to the multiple sample dialogue data. The multiple sample dialogue data are obtained by analyzing the sample dialogue corpus based on at least two different corpus analysis dimensions. The complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different.

[0175] Step 404: Input multiple sample dialogue data into the dialogue processing model to be evaluated, and obtain the predicted response results corresponding to the multiple sample dialogue data respectively.

[0176] Step 406: Based on the sample response results and the predicted response results, generate the model evaluation results for the dialogue processing model to be evaluated.

[0177] It should be noted that model evaluation data refers to the dataset used to assess the performance of the dialogue processing model under evaluation. Model evaluation data quantifies performance metrics such as accuracy and recall of the dialogue processing model under evaluation. Predicted response results refer to the responses generated by the dialogue processing model under evaluation based on its trained knowledge after inputting sample dialogue data. Sample response results refer to the expected correct responses corresponding to the predicted response results, serving as the standard for judging the accuracy of the dialogue processing model's responses. The dialogue processing model under evaluation refers to a pre-trained deep learning model for handling dialogue tasks, but its performance and applicability need to be evaluated. The dialogue processing model under evaluation can be a model trained using the dialogue processing model training method shown in Figure 1, or a model trained using other methods. Model evaluation results refer to a series of evaluation metrics generated by comparing the differences between sample response results and predicted response results, such as accuracy, recall, and F1 score. These metrics combined constitute the model evaluation results. The model evaluation results reflect the performance of the dialogue processing model under evaluation in handling the target dialogue task.

[0178] In practical applications, there are various ways to obtain model evaluation data, and the specific method should be selected according to the actual situation. This disclosure does not impose any limitations on these methods. In one possible implementation, model evaluation data for the target dialogue task can be read from other data acquisition devices or databases. In another possible implementation, the sample dialogue corpus can be analyzed according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data. Based on the multiple sample dialogue data and the corresponding sample response results, model evaluation data for the target dialogue task can be constructed.

[0179] The solution using the embodiments of this disclosure includes at least two types of sample dialogue data with different complexities in the model evaluation data. Therefore, the model evaluation data can be used to evaluate the dialogue processing model to be evaluated from multiple dimensions. During the evaluation process, the interaction of text, context, emotion and time information is fully considered. From multiple perspectives such as role knowledge, interaction logic, overall narrative, historical consistency and plot coherence, a comprehensive and automated evaluation of the dialogue processing model to be evaluated is achieved, which improves the comprehensiveness and reliability of the evaluation and provides more accurate feedback for the optimization of the dialogue processing model to be evaluated.

[0180] In one optional embodiment of this disclosure, after generating the model evaluation result of the dialogue processing model to be evaluated based on the sample response result and the predicted response result, the following steps may be further included:

[0181] If the model evaluation results do not meet the model evaluation conditions, the parameters of the dialogue processing model to be evaluated are adjusted to obtain the adjusted dialogue processing model to be evaluated.

[0182] It should be noted that model evaluation criteria refer to a series of pre-set standards or thresholds when evaluating the dialogue processing model to be evaluated. These criteria are used to determine whether the performance of the dialogue processing model to be evaluated has reached the expected goals, or whether it is suitable for deployment in real-world application scenarios. Model evaluation criteria can be set based on various indicators and project requirements, such as an accuracy rate of over 90%.

[0183] In practical applications, if the model evaluation results meet the model evaluation conditions, it means that the performance of the dialogue processing model under evaluation has reached the expected goal or can be deployed in practical applications, and there is no need to adjust the parameters of the dialogue processing model under evaluation. If the model evaluation results do not meet the model evaluation conditions, it means that the performance of the dialogue processing model under evaluation has not reached the expected goal and cannot be deployed in practical applications. In this case, the parameters of the dialogue processing model under evaluation can be adjusted. The implementation method of "adjusting the parameters of the dialogue processing model under evaluation to obtain the adjusted dialogue processing model under evaluation" can be referred to the dialogue processing model training method shown in Figure 1, and will not be described in detail in this embodiment.

[0184] By applying the solution of this disclosure embodiment, when the model evaluation result does not meet the model evaluation conditions, the ability of the dialogue processing model to be evaluated to grasp sample dialogue data of different complexities is continuously improved, realizing the iterative optimization of the dialogue processing model to be evaluated, and continuously improving the dialogue processing model to be evaluated in terms of character restoration, plot coherence and worldview consistency.

[0185] Referring to Figure 5, Figure 5 shows a flowchart of a data determination method for a dialogue task provided by an embodiment of this disclosure, specifically including the following steps:

[0186] Step 502: Obtain sample dialogue data.

[0187] Step 504: Analyze the sample dialogue corpus according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data. The complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different.

[0188] Step 506: Determine the task data for the target dialogue task based on the multiple sample dialogue data and the sample response results corresponding to the multiple sample dialogue data.

[0189] It should be noted that the implementation methods of steps 502 and 504 are the same as those of steps 102 to 104, and will not be described again in this embodiment. The task data of the target dialogue task consists of dialogue data pairs composed of sample dialogue data and sample response results corresponding to the sample dialogue data. The task data of the target dialogue task can be model evaluation data used for evaluating the dialogue processing model, or model training data used for training the dialogue processing model.

[0190] In one optional embodiment of this disclosure, after determining the task data for the target dialogue task based on multiple sample dialogue data and the sample response results corresponding to each sample dialogue data, the quantity of target task data matching the processor performance can be determined from the task data of the target dialogue task based on the processor performance of the task platform (model training platform or model evaluation platform) used to process the target dialogue task. Specifically, quality inspection can be performed on each task data to obtain data quality indicators, and the task data can be sorted according to the data quality indicators; the processor performance for executing the target dialogue task can be determined, and the target task data can be filtered from the sorted task data, wherein the quantity of target task data is adapted to the processor performance. This scheme can avoid the processor processing too much task data when processing the target dialogue task, which would lead to excessive processor pressure and thus task platform overload. In actual implementation, the processor performance may include computing power (such as clock frequency, number of cores, number of threads, etc.), memory management (cache size and hierarchy, memory bandwidth, memory latency, etc.), scalability and compatibility, etc. Of course, other indicators can also be used to indicate the processor's processing performance, and this disclosure embodiment does not limit this. Clock frequency, usually expressed in GHz, refers to the number of cycles a processor can execute per second. A higher clock frequency means the processor can complete more instructions in the same amount of time. Modern processors typically have multiple cores, each of which can execute program instructions independently. Multi-core processors can run multiple tasks simultaneously or process different parts of a single task in parallel, thereby improving efficiency. The processor's internal cache can be used to store frequently accessed data and instructions, reducing the time required to read from main memory. A larger cache usually improves performance. Processor performance indicates the current processing power of the processor on a task platform. The higher the processor performance, the more target task data it can process; the lower the processor performance, the less target task data it can process. Different processor performance levels can be pre-configured to correspond to different target task data volumes.

[0191] By applying the solution of this disclosure, sample dialogue data with different complexities are obtained by analyzing sample dialogue data from at least two different corpus analysis dimensions, which increases the diversity, generalization and coverage of task data for the target dialogue task, and provides high-quality task data for the process of evaluating or training dialogue processing models.

[0192] Referring to Figure 6, which illustrates a flowchart of a data determination and application method for dialogue tasks according to an embodiment of this disclosure, the method specifically includes: acquiring sample dialogue corpora; processing the sample dialogue corpora to obtain processed sample dialogue corpora, wherein the processing methods include, but are not limited to, cleaning, track splitting, segmentation, and alignment; analyzing the processed sample dialogue corpora based on multiple corpus analysis dimensions to obtain multiple sample dialogue data, wherein the complexity of the sample dialogue data corresponding to different corpus analysis dimensions varies, such as role attribute analysis dimension ("point" analysis dimension), role interaction analysis dimension ("line" analysis dimension), plot background analysis dimension ("surface" analysis dimension), and plot progress analysis dimension ("timeline" analysis dimension), etc.; generating sample response results for the multiple sample dialogue data respectively; and determining the task data for the target dialogue task based on the multiple sample dialogue data and the sample response results corresponding to the multiple sample dialogue data respectively; and after determining the task data for the target dialogue task, the task data can be used for dialogue processing model training or dialogue processing model evaluation. The dialogue processing model training and dialogue processing model evaluation processes will be described below.

[0193] Dialogue processing model training: Based on multiple sample dialogue data and the sample response results corresponding to the multiple sample dialogue data, adjust the model parameters of the dialogue processing model to obtain the trained dialogue processing model.

[0194] Dialogue processing model evaluation: Input multiple sample dialogue data into the dialogue processing model to obtain the predicted response results corresponding to each sample dialogue data; generate the model evaluation results of the dialogue processing model based on the sample response results and the predicted response results.

[0195] In one optional embodiment of this disclosure, a trained dialogue processing model can be used to process a target dialogue task: acquiring the question data to be processed for the target dialogue task; inputting the question data to be processed into the dialogue processing model to obtain the target response result. Furthermore, the dialogue processing model can be optimized based on the target response result. Specifically, a quality check is performed on the target response result to obtain a quality index; if the quality index does not meet the quality check conditions, the parameters of the dialogue processing model are adjusted to obtain an adjusted dialogue processing model.

[0196] In another optional embodiment of this disclosure, after generating the model evaluation results of the dialogue processing model, the dialogue processing model can be optimized based on the model evaluation results. Specifically, if the model evaluation results do not meet the model evaluation conditions, the parameters of the dialogue processing model are adjusted to obtain an adjusted dialogue processing model.

[0197] Referring to Figure 7, which shows a flowchart of an information processing method based on a dialogue processing model according to an embodiment of this disclosure, the information processing method based on the dialogue processing model is applied to a task platform and specifically includes the following steps:

[0198] Step 702: Receive the model request sent by the terminal device.

[0199] Step 704: Based on the model request, determine the target dialogue processing model from multiple dialogue processing models, wherein the multiple dialogue processing models are trained based on the dialogue processing model training method.

[0200] It should be noted that the multiple dialogue processing models can be dialogue processing models with different model specifications and parameters, adapted to different dialogue scenarios. The target dialogue processing model is a dialogue processing model suitable for the target dialogue scenario. The target dialogue scenario can be different dialogue scenarios, such as emotional dialogue scenarios, role-playing dialogue scenarios, customer service dialogue scenarios, etc. The model request includes a scenario identifier of the target dialogue scenario, scenario input data of the target dialogue scenario, and at least one of the model specification parameters. Based on the model request, there are multiple ways to determine the target dialogue processing model from multiple dialogue processing models, and the specific method is selected according to the actual situation. This disclosure does not limit this method. In the first possible implementation of this disclosure, the corresponding target dialogue processing model can be searched from the dialogue processing models included in the first model library based on the scenario identifier included in the model request; in the second possible implementation of this disclosure, the target dialogue processing model can be trained and obtained based on the scenario input data included in the model request; in the third possible implementation of this disclosure, the target dialogue processing model can be constructed based on the model request.

[0201] For example, based on the scene identifier of the target dialogue scenario, at least one pre-trained dialogue processing model can be searched from a first model library. Then, based on the model specification parameters, an intermediate dialogue processing model can be selected from the at least one dialogue processing model. Finally, based on the scene input data of the target dialogue scenario, the selected intermediate dialogue processing model is trained to obtain a target dialogue processing model suitable for user needs. The at least one dialogue processing model can be trained according to the dialogue processing model training method shown in Figure 1, which will not be elaborated further in this embodiment.

[0202] The solution applied in this disclosure is adapted to user needs to obtain the target dialogue processing model, realizing personalized model services, providing users with an efficient, flexible and easy-to-use model service method, and improving user experience.

[0203] In one optional embodiment of this disclosure, determining the target dialogue processing model from multiple dialogue processing models based on a model request may include the following steps:

[0204] If the model request includes a scene identifier of the target dialogue scenario, the target dialogue processing model that is suitable for the target dialogue scenario is searched from the first model library based on the scene identifier. The first model library stores multiple dialogue processing models that are suitable for different dialogue scenarios.

[0205] When the model request includes scene input data of the target dialogue scenario, the dialogue processing model to be trained that is suitable for the target dialogue scenario is determined from multiple dialogue processing models, and the dialogue processing model to be trained is trained based on the scene input data to obtain the target dialogue processing model.

[0206] If the model request includes model specification parameters, the target dialogue processing model corresponding to the model specification parameters is searched from the second model library, which stores multiple dialogue processing models with different model specification parameters.

[0207] It should be noted that scene identifiers refer to unique or specific labels used to distinguish different dialogue scenarios. The first model library is a database for storing and managing various pre-trained deep learning models. Multiple dialogue processing models adapted to different dialogue scenarios cover various application scenarios and needs. The first model library allows users to select a suitable dialogue processing model according to their needs, or directly call the appropriate dialogue processing model through the application programming interface to execute the target dialogue task. The multiple dialogue processing models adapted to different dialogue scenarios are various models stored in the first model library suitable for different dialogue scenarios. Each dialogue processing model is optimized for a specific application environment, and any given dialogue processing model is a model trained according to the dialogue processing model training method shown in Figure 1. For example, based on the scene identifier "virtual character dialogue" of the target dialogue scenario, a target dialogue processing model adapted to the virtual character dialogue scenario can be found in the first model library.

[0208] Different dialogue processing models are adapted to different dialogue scenarios. For example, dialogue processing model 1 is suitable for dialogue scenarios 1 and 2, while dialogue processing model 2 is suitable for dialogue scenarios 2 and 3. A dialogue processing model to be trained refers to a model among multiple dialogue processing models that is suitable for the target dialogue scenario, but whose performance can be further optimized. If the target dialogue scenario is dialogue scenario 1, then the dialogue processing model to be trained is dialogue processing model 1 adapted to dialogue scenario 1. A dialogue processing model to be trained may not only be applicable to the target dialogue scenario but also to other dialogue scenarios, making it a general dialogue processing model applicable to different dialogue scenarios. The dialogue processing model to be trained can be used to perform the target dialogue task, but the results may not be very good. In this case, the dialogue processing model to be trained can be optimized based on the scenario input data of the target dialogue scenario. For example, optimizing the dialogue processing model to be trained based on the scenario input data of a virtual character dialogue scenario can yield a target dialogue processing model suitable for virtual character dialogue scenarios. The scenario input data of the target dialogue scenario can be understood as the model training data of the dialogue processing model in the target dialogue scenario (including sample dialogue data of at least two complexities and the sample response results corresponding to each sample dialogue data). The dialogue processing model to be trained is trained based on the scene input data of the target dialogue scenario. The method of obtaining the target dialogue processing model can be referred to the dialogue processing model training method shown in Figure 1, and will not be described again in this embodiment.

[0209] Model specifications refer to the various parameters that define the structure and behavior of a model. These parameters can be broadly categorized into two types: model parameters (learnable parameters) and hyperparameters. Model parameters are those automatically adjusted during model training using the backpropagation algorithm, including but not limited to weight matrices and bias terms. For example, in a simple fully connected layer, the weight matrix is ​​a two-dimensional tensor connecting neurons in the input and output layers; the bias term is a one-dimensional vector providing additional offset values ​​for each output neuron. Hyperparameters are parameters set before model training begins, used to control the model's learning process and architecture. Hyperparameters include, but are not limited to, the learning rate and the number of neurons per layer, selected based on specific requirements.

[0210] The solution applied in this disclosure embodiment, based on scenario requirements, accurately identifies the target dialogue processing model adapted to the corresponding scenario through scenario identification, making the processing of the target dialogue task more accurate and more in line with the target dialogue scenario; based on scenario requirements, a general dialogue processing model to be trained is further trained through scenario input data to obtain a target dialogue processing model adapted to the target dialogue scenario, making the target dialogue processing model more in line with the target dialogue scenario, thereby improving user experience and task processing quality; based on model specification parameters, the corresponding target dialogue processing model can be accurately found, ensuring the efficient and stable operation of the target dialogue processing model and improving user experience.

[0211] In one optional embodiment of this disclosure, after determining the target dialogue processing model from multiple dialogue processing models based on model requests, the following steps may be further included:

[0212] Deploy the target dialogue processing model, and based on the target dialogue processing model, build a dialogue task processing interface so that the terminal device can schedule the target dialogue processing model to execute the target dialogue task through the dialogue task processing interface.

[0213] It should be noted that the dialogue task processing interface is an interactive programming interface for the terminal device to schedule the target dialogue processing model to process the target dialogue task, and it is usually provided in the form of an application programming interface (API). Through the dialogue task processing interface, the user can input the question data to be processed for the target dialogue task, and then proceed with the target dialogue task processing. The target dialogue task can be of different types, such as a dialogue processing model training task, a dialogue processing model-based dialogue processing task, a dialogue processing model evaluation task, a dialogue processing model training data determination task, a dialogue processing model evaluation data determination task, and so on. For example, in the embodiment shown in Figure 1, the target dialogue task is a dialogue processing model training task.

[0214] In practical applications, there are various ways to deploy the target dialogue processing model, and the specific method chosen depends on the actual situation. This disclosure does not impose any limitations on this approach. One possible implementation of this disclosure is to deploy the target dialogue processing model on cloud-side devices using infrastructure provided by a cloud service provider. Another possible implementation is to deploy the target dialogue processing model on edge devices using a lightweight framework. For example, the target dialogue processing model can be deployed on a distributed system, and a dialogue task processing interface can be built based on the target dialogue processing model and provided to terminal devices, enabling the terminal devices to schedule the target dialogue processing model to execute target dialogue tasks.

[0215] By applying the solution provided in the embodiments of this disclosure, deploying a target dialogue processing model, and constructing a dialogue task processing interface based on the target dialogue processing model, terminal devices can efficiently call the target dialogue processing model, thereby improving the processing quality and response speed of the target dialogue task.

[0216] Referring to Figure 8, Figure 8 shows a schematic diagram of the structure of a task platform provided in an embodiment of the present disclosure. The task platform 800 includes a request interface 802 and a response unit 804.

[0217] Request interface 802 is used to receive a model request sent by a terminal device, wherein the model request includes at least one of the following: scene identifier of the target dialogue scene, scene input data of the target dialogue scene, and model specification parameters.

[0218] The response unit 804 is used to determine the target dialogue processing model from multiple dialogue processing models based on the model request, wherein the multiple dialogue processing models are trained based on the dialogue processing model training method.

[0219] In one optional embodiment of this disclosure, the task platform further includes a dialogue task processing interface, which is constructed based on the target dialogue processing model.

[0220] The dialogue task processing interface is used for terminal devices to schedule and execute target dialogue tasks.

[0221] The above is an illustrative scheme of a task platform according to this embodiment. It should be noted that the technical solution of this task platform and the technical solution of the information processing method based on the dialogue processing model described above belong to the same concept. For details not described in detail in the technical solution of the task platform, please refer to the description of the technical solution of the information processing method based on the dialogue processing model described above.

[0222] Corresponding to the above-described embodiments of dialogue processing model training methods, this disclosure also provides embodiments of dialogue processing model training devices. Figure 9 shows a schematic diagram of the structure of a dialogue processing model training device provided in one embodiment of this disclosure. As shown in Figure 9, the device includes:

[0223] The first acquisition module 902 is configured to acquire sample dialogue data;

[0224] The first analysis module 904 is configured to analyze the sample dialogue corpus according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data, wherein the complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different.

[0225] The first adjustment module 906 is configured to adjust the model parameters of the dialogue processing model based on multiple sample dialogue data and the sample response results corresponding to the multiple sample dialogue data, so as to obtain the trained dialogue processing model.

[0226] Optionally, at least two different corpus analysis dimensions include a role analysis dimension and a plot analysis dimension; the first analysis module 904 is further configured to perform role analysis on the sample dialogue corpus according to the role analysis dimension to obtain sample role dialogue data; and to perform plot analysis on the sample dialogue corpus according to the plot analysis dimension to obtain sample plot dialogue data, wherein the complexity of the sample role dialogue data is less than the complexity of the sample plot dialogue data.

[0227] Optionally, the role analysis dimensions include a first role analysis dimension and a second role analysis dimension; the first analysis module 904 is further configured to analyze the role attribute information in the sample dialogue corpus according to the first role analysis dimension to obtain first sample role dialogue data; and to analyze the role interaction information in the sample dialogue corpus according to the second role analysis dimension to obtain second sample role dialogue data, wherein the complexity of the first sample role dialogue data is less than the complexity of the second sample role dialogue data.

[0228] Optionally, the plot analysis dimension includes a first plot analysis dimension and a second plot analysis dimension; the first analysis module 904 is further configured to analyze the plot background information in the sample dialogue corpus according to the first plot analysis dimension to obtain first sample plot dialogue data; and to analyze the plot progress information in the sample dialogue corpus according to the second plot analysis dimension to obtain second sample plot dialogue data, wherein the complexity of the first sample plot dialogue data is less than the complexity of the second sample plot dialogue data.

[0229] Optionally, the first analysis module 904 is further configured to acquire reference dialogue data of at least two different corpus analysis dimensions; and to analyze the sample dialogue corpus based on the at least two different corpus analysis dimensions and the reference dialogue data to obtain multiple sample dialogue data.

[0230] Optionally, the device further includes: an identification module configured to perform modal recognition on the sample dialogue corpus to obtain the corpus modality of the sample dialogue corpus; determine a processing strategy for the sample dialogue corpus based on the corpus modality, and process the sample dialogue corpus based on the processing strategy to obtain processed sample dialogue corpus; and a first analysis module 904 further configured to analyze the processed sample dialogue corpus according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data.

[0231] Optionally, the device further includes an annotation module, configured to call a data annotation platform to annotate the response results of multiple sample dialogue data, and obtain the sample response results corresponding to the multiple sample dialogue data respectively.

[0232] By applying the solution of this disclosure embodiment, sample dialogue data with different complexities are obtained by analyzing sample dialogue data from at least two different corpus analysis dimensions. This increases the diversity and coverage of the dialogue processing model's learning, enabling the dialogue processing model to learn how to process dialogue data of various complexities. This enhances the adaptability of the dialogue processing model in different application scenarios and improves the robustness and flexibility of the dialogue processing model.

[0233] The above is an illustrative scheme of a dialogue processing model training device according to this embodiment. It should be noted that the technical solution of this dialogue processing model training device and the technical solution of the dialogue processing model training method described above belong to the same concept. For details not described in detail in the technical solution of the dialogue processing model training device, please refer to the description of the technical solution of the dialogue processing model training method described above.

[0234] Corresponding to the above-described embodiments of dialogue task processing methods, this disclosure also provides embodiments of dialogue task processing apparatus. Figure 10 shows a schematic diagram of the structure of a dialogue task processing apparatus provided in one embodiment of this disclosure. As shown in Figure 10, the apparatus includes:

[0235] The second acquisition module 1002 is configured to acquire the pending question data of the target dialogue task;

[0236] The first input module 1004 is configured to input the question data to be processed into the dialogue processing model to obtain the target response result, wherein the dialogue processing model is trained based on the dialogue processing model training method.

[0237] Optionally, the device further includes: a second adjustment module configured to perform quality detection on the target response result to obtain a quality index of the target response result; and to adjust the parameters of the dialogue processing model to obtain an adjusted dialogue processing model if the quality index does not meet the quality detection conditions.

[0238] Optionally, the device further includes: a third adjustment module, configured to receive result feedback information sent by the client, wherein the result feedback information is information provided by the client in response to the target response based on the dialogue requirements; and to adjust the parameters of the dialogue processing model according to the result feedback information to obtain the adjusted dialogue processing model.

[0239] By applying the solution of this disclosure embodiment, since the dialogue processing model trained based on the dialogue processing model training method has learned how to process dialogue data of various complexities and has a very strong adaptability in different dialogue scenarios, the dialogue processing model is used to process the problem data to be processed, ensuring the normal progress of dialogue task processing, thereby obtaining a highly accurate target response result.

[0240] The above is an illustrative scheme of a dialogue task processing device according to this embodiment. It should be noted that the technical solution of this dialogue task processing device and the technical solution of the above-described dialogue task processing method belong to the same concept. For details not described in detail in the technical solution of the dialogue task processing device, please refer to the description of the technical solution of the above-described dialogue task processing method.

[0241] Corresponding to the above-described virtual character dialogue method embodiments, this disclosure also provides virtual character dialogue device embodiments. Figure 11 shows a schematic diagram of the structure of a virtual character dialogue device provided in one embodiment of this disclosure. As shown in Figure 11, the device includes:

[0242] The third acquisition module 1102 is configured to acquire virtual character question data of virtual character dialogue tasks;

[0243] The second input module 1104 is configured to input virtual character question data into the dialogue processing model to obtain the virtual character's response result, wherein the dialogue processing model is trained based on the dialogue processing model training method.

[0244] By applying the solution of this disclosure embodiment, since the dialogue processing model trained based on the dialogue processing model training method has learned how to process dialogue data of various complexities and has a very strong adaptability in different dialogue scenarios, the dialogue processing model is used to process the virtual character question data, ensuring the normal progress of the virtual character dialogue task processing, thereby obtaining highly accurate virtual character response results.

[0245] The above is an illustrative scheme of a virtual character dialogue device according to this embodiment. It should be noted that the technical solution of this virtual character dialogue device and the technical solution of the virtual character dialogue method described above belong to the same concept. For details not described in detail in the technical solution of the virtual character dialogue device, please refer to the description of the technical solution of the virtual character dialogue method described above.

[0246] Corresponding to the above-described embodiments of the dialogue processing model evaluation method, this disclosure also provides embodiments of the dialogue processing model evaluation device. Figure 12 shows a schematic diagram of the structure of a dialogue processing model evaluation device provided in one embodiment of this disclosure.

[0247] As shown in Figure 12, the device includes:

[0248] The fourth acquisition module 1202 is configured to acquire model evaluation data, wherein the model evaluation data includes multiple sample dialogue data and sample response results corresponding to the multiple sample dialogue data respectively. The multiple sample dialogue data are obtained by analyzing the sample dialogue data based on at least two different corpus analysis dimensions, and the complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different.

[0249] The third input module 1204 is configured to input multiple sample dialogue data into the dialogue processing model to be evaluated, and obtain the predicted response results corresponding to the multiple sample dialogue data respectively.

[0250] The generation module 1206 is configured to generate model evaluation results for the dialogue processing model to be evaluated based on the sample response results and the predicted response results.

[0251] Optionally, the device further includes a fourth adjustment module, configured to adjust the parameters of the dialogue processing model to be evaluated when the model evaluation result does not meet the model evaluation conditions, so as to obtain an adjusted dialogue processing model to be evaluated.

[0252] The solution using the embodiments of this disclosure includes at least two types of sample dialogue data with different complexities in the model evaluation data. Therefore, the model evaluation data can be used to perform multi-dimensional fusion evaluation of the dialogue processing model to be evaluated. During the evaluation process, the interaction of text, context, emotion, and time information is fully considered. From multiple perspectives such as role knowledge, interaction logic, overall narrative, historical consistency, and plot coherence, a comprehensive and automated evaluation of the dialogue processing model to be evaluated is achieved, improving the comprehensiveness and reliability of the evaluation and providing more accurate feedback for the optimization of the dialogue processing model to be evaluated.

[0253] The above is an illustrative scheme of a dialogue processing model evaluation device according to this embodiment. It should be noted that the technical solution of this dialogue processing model evaluation device and the technical solution of the dialogue processing model evaluation method described above belong to the same concept. For details not described in detail in the technical solution of the dialogue processing model evaluation device, please refer to the description of the technical solution of the dialogue processing model evaluation method described above.

[0254] Corresponding to the above-described embodiments of the data determination method for dialogue tasks, this disclosure also provides embodiments of a data determination apparatus for dialogue tasks. Figure 13 shows a schematic diagram of the structure of a data determination apparatus for dialogue tasks provided in one embodiment of this disclosure. As shown in Figure 13, the apparatus includes:

[0255] The fifth acquisition module 1302 is configured to acquire sample dialogue data;

[0256] The second analysis module 1304 is configured to analyze the sample dialogue data according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data, wherein the complexity of the sample dialogue data corresponding to different corpus analysis dimensions is different.

[0257] The first determining module 1306 is configured to determine the task data of the target dialogue task based on multiple sample dialogue data and the sample response results corresponding to the multiple sample dialogue data respectively.

[0258] By applying the solution of this disclosure, sample dialogue data with different complexities are obtained by analyzing sample dialogue data from at least two different corpus analysis dimensions, which increases the diversity, generalization and coverage of task data for the target dialogue task, and provides high-quality task data for the process of evaluating or training dialogue processing models.

[0259] The above is an illustrative scheme of a data determination device for a dialogue task according to this embodiment. It should be noted that the technical solution of this data determination device for a dialogue task and the technical solution of the data determination method for a dialogue task described above belong to the same concept. For details not described in detail in the technical solution of the data determination device for a dialogue task, please refer to the description of the technical solution of the data determination method for a dialogue task described above.

[0260] Corresponding to the above-described information processing method embodiments based on a dialogue processing model, this disclosure also provides embodiments of an information processing apparatus based on a dialogue processing model. Figure 14 shows a schematic diagram of the structure of an information processing apparatus based on a dialogue processing model provided in one embodiment of this disclosure. As shown in Figure 14, the apparatus is applied to a task platform and includes:

[0261] The receiving module 1402 is configured to receive model requests sent by the terminal device;

[0262] The second determining module 1404 is configured to determine a target dialogue processing model from multiple dialogue processing models based on a model request, wherein the multiple dialogue processing models are trained based on a dialogue processing model training method.

[0263] Optionally, the second determining module 1404 is further configured to: when the model request includes a scene identifier of the target dialogue scene, search for a target dialogue processing model adapted to the target dialogue scene from a first model library based on the scene identifier, wherein the first model library stores multiple dialogue processing models adapted to different dialogue scenes; when the model request includes scene input data of the target dialogue scene, determine a dialogue processing model to be trained adapted to the target dialogue scene from multiple dialogue processing models, and train the dialogue processing model to be trained based on the scene input data to obtain the target dialogue processing model; when the model request includes model specification parameters, search for the target dialogue processing model corresponding to the model specification parameters from a second model library, wherein the second model library stores multiple dialogue processing models with different model specification parameters.

[0264] Optionally, the device further includes: a deployment module configured to deploy a target dialogue processing model and, based on the target dialogue processing model, construct a dialogue task processing interface so that the terminal device can schedule the target dialogue processing model to execute target dialogue tasks through the dialogue task processing interface.

[0265] The solution applied in this disclosure is adapted to user needs to obtain the target dialogue processing model, realizing personalized model services, providing users with an efficient, flexible and easy-to-use model service method, and improving user experience.

[0266] The above is an illustrative scheme of an information processing device based on a dialogue processing model according to this embodiment. It should be noted that the technical solution of this information processing device based on a dialogue processing model and the technical solution of the information processing method based on a dialogue processing model described above belong to the same concept. For details not described in detail in the technical solution of the information processing device based on a dialogue processing model, please refer to the description of the technical solution of the information processing method based on a dialogue processing model described above.

[0267] Figure 15 shows a structural block diagram of a computing device 1500 provided in one embodiment of the present disclosure.

[0268] The computing device 1500 includes a memory 1510 and a processor 1520; the memory 1510 is used to store computer programs / instructions, and the processor 1520 is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor 1520, they implement the steps of the above-mentioned dialogue processing model training method, dialogue task processing method, virtual character dialogue method, dialogue processing model evaluation method, data determination method for dialogue tasks, or information processing method based on dialogue processing model.

[0269] In one or more embodiments of this disclosure, the computing device can be understood as an integrated smart terminal, including but not limited to a server, desktop computer, personal computer (PC), all-in-one model machine, mobile phone, tablet computer or other portable smart terminal, etc., and the computing device may have the model described in the above embodiments of this disclosure pre-installed.

[0270] Specifically, this computing device can pre-install various types of models, including but not limited to models in natural language processing, visual processing, speech processing, code processing, and multimodal task processing, thus providing diverse model selection. In different product forms, this computing device can support one or more model usage methods, including but not limited to model training, model invocation, model fine-tuning, model deployment, model inference, and application. In some product forms, this computing device also supports model management, including but not limited to multi-type model management (supporting the management of discriminative, generative, and other types of models), model version control (supporting the control of different model versions), and model evaluation (evaluating model performance and effectiveness based on model evaluation tools). In other product forms, this computing device can also create applications based on models, providing application programming interface (API) invocation capabilities. Models can be invoked into created applications through the API interface, and application management tools are provided for application management and monitoring.

[0271] Furthermore, the computing device may also include data management (supporting the creation and management of model tuning datasets), a training center (providing abundant training resources to help users learn and master artificial intelligence technology), and basic control capabilities (providing enterprise-level basic control capabilities to ensure the security and efficient operation of the system). Through the above functions, it provides a comprehensive and integrated device for artificial intelligence development, training, deployment, and application.

[0272] Figure 16 shows a structural block diagram of an electronic device 1600 provided according to an embodiment of the present disclosure.

[0273] The memory 1610 and the processor 1620 are connected via a bus 1630;

[0274] The memory 1610 is used to store computer programs / instructions, and the processor 1620 is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor 1620, they implement the steps of the above-mentioned dialogue processing model training method, dialogue task processing method, virtual character dialogue method, dialogue processing model evaluation method, data determination method for dialogue tasks, or information processing method based on dialogue processing model.

[0275] Specifically, the components of the electronic device 1600 include, but are not limited to, a memory 1610 and a processor 1620. The processor 1620 and the memory 1610 can be connected via a bus 1630.

[0276] Electronic device 1600 may also include access device 1640, which enables electronic device 1600 to communicate with database 1650 storing data via one or more networks 1660. Examples of such networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. Access device 1640 may include one or more of any type of wired or wireless network interface (e.g., Network Interface Card (NIC)), such as IEEE 802.11 Wireless Local Area Networks (WLAN) wireless interface, Wi-MAX (World Interoperability for Microwave Access) interface, Ethernet interface, Universal Serial Bus (USB) interface, cellular network interface, Bluetooth interface, Near Field Communication (NFC) interface, and so on.

[0277] In one embodiment of this disclosure, the aforementioned components of the electronic device 1600, as well as other components not shown in FIG. 16, may also be connected to each other, for example, via a bus. It should be understood that the electronic device structural block diagram shown in FIG. 16 is merely for illustrative purposes and is not intended to limit the scope of this disclosure. Those skilled in the art can add or replace other components as needed.

[0278] Electronic device 1600 can be any type of stationary or mobile electronic device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable electronic devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary electronic devices such as desktop computers or PCs. Electronic device 1600 can also be a mobile or stationary electronic device.

[0279] The above is an illustrative scheme of an electronic device according to this embodiment. It should be noted that the technical solution of this electronic device belongs to the same concept as the aforementioned dialogue processing model training method, dialogue task processing method, virtual character dialogue method, dialogue processing model evaluation method, data determination method for dialogue tasks, and information processing method based on dialogue processing models. Details not described in detail in the technical solution of the electronic device can be found in the descriptions of the aforementioned dialogue processing model training method, dialogue task processing method, virtual character dialogue method, dialogue processing model evaluation method, data determination method for dialogue tasks, or information processing method based on dialogue processing models.

[0280] An embodiment of this disclosure also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the above-described dialogue processing model training method, dialogue task processing method, virtual character dialogue method, dialogue processing model evaluation method, data determination method for dialogue tasks, or information processing method based on a dialogue processing model.

[0281] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the technical solutions of the aforementioned dialogue processing model training method, dialogue task processing method, virtual character dialogue method, dialogue processing model evaluation method, data determination method for dialogue tasks, and information processing method based on dialogue processing models. Details not described in detail in the technical solution of the storage medium can be found in the descriptions of the aforementioned technical solutions of the dialogue processing model training method, dialogue task processing method, virtual character dialogue method, dialogue processing model evaluation method, data determination method for dialogue tasks, or information processing method based on dialogue processing models.

[0282] An embodiment of this disclosure also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-described dialogue processing model training method, dialogue task processing method, virtual character dialogue method, dialogue processing model evaluation method, data determination method for dialogue tasks, or information processing method based on a dialogue processing model.

[0283] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product belongs to the same concept as the aforementioned dialogue processing model training method, dialogue task processing method, virtual character dialogue method, dialogue processing model evaluation method, data determination method for dialogue tasks, and information processing method based on dialogue processing models. Details not described in detail in the technical solution of the computer program product can be found in the descriptions of the aforementioned dialogue processing model training method, dialogue task processing method, virtual character dialogue method, dialogue processing model evaluation method, data determination method for dialogue tasks, or information processing method based on dialogue processing models.

[0284] The foregoing has described specific embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0285] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0286] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of this disclosure are not limited to the described order of actions, because according to the embodiments of this disclosure, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments of this disclosure.

[0287] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0288] The preferred embodiments disclosed above are merely illustrative of this disclosure. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments of this disclosure. These embodiments are selected and specifically described in this disclosure to better explain the principles and practical applications of the embodiments of this disclosure, thereby enabling those skilled in the art to better understand and utilize this disclosure. This disclosure is limited only by the claims and their full scope and equivalents.

Claims

1. A method for training a dialogue processing model, comprising: Obtain sample dialogue data; The sample dialogue corpus is analyzed according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data, wherein the complexity of the sample dialogue data corresponding to the different corpus analysis dimensions is different. Based on the multiple sample dialogue data and the corresponding sample response results, the model parameters of the dialogue processing model are adjusted to obtain the trained dialogue processing model.

2. The method according to claim 1, wherein the at least two different corpus analysis dimensions include a character analysis dimension and a plot analysis dimension; The sample dialogue corpus is analyzed according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data, including: Based on the aforementioned role analysis dimensions, role analysis is performed on the sample dialogue corpus to obtain sample role dialogue data; Based on the aforementioned plot analysis dimension, plot analysis is performed on the sample dialogue corpus to obtain sample plot dialogue data, wherein the complexity of the sample character dialogue data is less than the complexity of the sample plot dialogue data.

3. The method according to claim 2, wherein the role analysis dimension includes a first role analysis dimension and a second role analysis dimension; The step of performing role analysis on the sample dialogue corpus according to the aforementioned role analysis dimensions to obtain sample role dialogue data includes: Based on the first role analysis dimension, the role attribute information in the sample dialogue corpus is analyzed to obtain the first sample role dialogue data; Based on the second role analysis dimension, the role interaction information in the sample dialogue corpus is analyzed to obtain the second sample role dialogue data, wherein the complexity of the first sample role dialogue data is less than that of the second sample role dialogue data.

4. The method according to claim 2 or 3, wherein the plot analysis dimension includes a first plot analysis dimension and a second plot analysis dimension; The step of performing plot analysis on the sample dialogue corpus according to the plot analysis dimension to obtain sample plot dialogue data includes: Based on the first plot analysis dimension, the plot background information in the sample dialogue corpus is analyzed to obtain the first sample plot dialogue data; Based on the second plot analysis dimension, the plot progress information in the sample dialogue corpus is analyzed to obtain second sample plot dialogue data, wherein the complexity of the first sample plot dialogue data is less than that of the second sample plot dialogue data.

5. The method according to claim 1, wherein analyzing the sample dialogue corpus according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data includes: Obtain reference dialogue data for at least two different corpus analysis dimensions; Based on the at least two different corpus analysis dimensions and the reference dialogue data, the sample dialogue corpus is analyzed to obtain multiple sample dialogue data.

6. The method according to any one of claims 1 to 5, wherein before analyzing the sample dialogue corpus according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data, the method further includes: Modality recognition is performed on the sample dialogue corpus to obtain the corpus modality of the sample dialogue corpus; Based on the corpus modality, a processing strategy for the sample dialogue corpus is determined, and based on the processing strategy, the sample dialogue corpus is processed to obtain the processed sample dialogue corpus. The sample dialogue corpus is analyzed according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data, including: The processed sample dialogue corpus is analyzed based on at least two different corpus analysis dimensions to obtain multiple sample dialogue data.

7. The method according to any one of claims 1 to 6, wherein before adjusting the model parameters of the dialogue processing model based on the plurality of sample dialogue data and the sample response results corresponding to the plurality of sample dialogue data respectively, and obtaining the trained dialogue processing model, the method further includes: The data annotation platform is invoked to annotate the response results of the multiple sample dialogue data, thereby obtaining the sample response results corresponding to the multiple sample dialogue data respectively.

8. A dialogue task processing method, comprising: Obtain the pending problem data for the target dialogue task; The question data to be processed is input into the dialogue processing model to obtain the target response result, wherein the dialogue processing model is trained based on the method described in any one of claims 1 to 7.

9. The method according to claim 8, after inputting the question data to be processed into the dialogue processing model to obtain the target response result, further comprising: The target response results are subjected to quality testing to obtain the quality indicators of the target response results; If the quality indicators do not meet the quality detection conditions, the parameters of the dialogue processing model are adjusted to obtain the adjusted dialogue processing model.

10. The method according to claim 8, further comprising, after inputting the question data to be processed into the dialogue processing model and obtaining the target response result: Receive result feedback information sent by the client, wherein the result feedback information is information provided by the client in response to the target response based on the dialogue requirements; Based on the feedback information, the parameters of the dialogue processing model are adjusted to obtain the adjusted dialogue processing model.

11. A method for virtual character dialogue, comprising: Obtain virtual character question data for virtual character dialogue tasks; The virtual character's question data is input into the dialogue processing model to obtain the virtual character's response, wherein the dialogue processing model is trained based on the method described in any one of claims 1 to 7.

12. A method for evaluating a dialogue processing model, comprising: Obtain model evaluation data, wherein the model evaluation data includes multiple sample dialogue data and sample response results corresponding to the multiple sample dialogue data respectively. The multiple sample dialogue data are obtained by analyzing the sample dialogue corpus based on at least two different corpus analysis dimensions, and the complexity of the sample dialogue data corresponding to the different corpus analysis dimensions is different. Input the multiple sample dialogue data into the dialogue processing model to be evaluated to obtain the predicted response results corresponding to the multiple sample dialogue data respectively. Based on the sample response results and the predicted response results, the model evaluation results of the dialogue processing model to be evaluated are generated.

13. The method according to claim 12, further comprising, after generating the model evaluation result of the dialogue processing model to be evaluated based on the sample response result and the predicted response result: If the model evaluation results do not meet the model evaluation conditions, the parameters of the dialogue processing model to be evaluated are adjusted to obtain the adjusted dialogue processing model to be evaluated.

14. A data determination method for a dialogue task, comprising: Obtain sample dialogue data; The sample dialogue corpus is analyzed according to at least two different corpus analysis dimensions to obtain multiple sample dialogue data, wherein the complexity of the sample dialogue data corresponding to the different corpus analysis dimensions is different. Based on the multiple sample dialogue data and the sample response results corresponding to the multiple sample dialogue data, the task data for the target dialogue task is determined.

15. An information processing method based on a dialogue processing model, applied to a task platform, comprising: Receive model requests sent by terminal devices; Based on the model request, a target dialogue processing model is determined from a plurality of dialogue processing models, wherein the plurality of dialogue processing models are trained based on the method described in any one of claims 1 to 7.

16. The method according to claim 15, wherein determining the target dialogue processing model from multiple dialogue processing models based on the model request comprises: If the model request includes a scene identifier of the target dialogue scenario, a target dialogue processing model suitable for the target dialogue scenario is searched from the first model library based on the scene identifier. The first model library stores multiple dialogue processing models suitable for different dialogue scenarios. When the model request includes scene input data of the target dialogue scenario, a dialogue processing model to be trained that is suitable for the target dialogue scenario is determined from the plurality of dialogue processing models, and the dialogue processing model to be trained is trained based on the scene input data to obtain the target dialogue processing model. If the model request includes model specification parameters, the target dialogue processing model corresponding to the model specification parameters is searched from the second model library, wherein the second model library stores multiple dialogue processing models with different model specification parameters.

17. The method according to claim 15 or 16, further comprising, after determining the target dialogue processing model from multiple dialogue processing models based on the model request: Deploy the target dialogue processing model, and based on the target dialogue processing model, construct a dialogue task processing interface so that the terminal device can schedule the target dialogue processing model to execute target dialogue tasks through the dialogue task processing interface.

18. A task platform, comprising a request interface and a response unit; The request interface is used to receive model requests sent by the terminal device, wherein... The model request includes at least one of the following: the scene identifier of the target dialogue scene, the scene input data of the target dialogue scene, and the model specification parameters. The response unit is configured to determine a target dialogue processing model from multiple dialogue processing models based on the model request, wherein the multiple dialogue processing models are trained based on the method described in any one of claims 1 to 7.

19. The task platform according to claim 18, further comprising a dialogue task processing interface, wherein the dialogue task processing interface is constructed based on the target dialogue processing model; The dialogue task processing interface is used for the terminal device to schedule and execute the target dialogue task.

20. A computing device, comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 17.

21. An electronic device, comprising: A memory and a processor, the memory and the processor being connected via a bus; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 17.

22. A computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 17.

23. A computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 17.

Images