Character role configuration data generation method and related device

Abstract

The invention provides a character role configuration data generation method and a related device. The method comprises the following steps: acquiring primary character role configuration data corresponding to the digital virtual doctor; wherein the primary character role configuration data is used for representing character character characteristics planned to be set for the digital virtual doctor; matching the primary character role configuration data in a specified doctor role configuration data set, and if the doctor role configuration data associated with the primary character role configuration data is not obtained, matching the primary character role configuration data with the specified doctor role configuration data set; taking the primary character role configuration data as target task role configuration data corresponding to the digital virtual doctor; wherein the specified doctor role configuration data set comprises a plurality of doctor role configuration data; different doctor role configuration data respectively represent role character characteristics set by different digital virtual doctors. According to the invention, the role character uniqueness of the digital virtual doctor can be improved to a certain extent.

Description

Technical Field

[0001] This application relates to the field of virtual digital human technology, and in particular to a method and related apparatus for generating character configuration data. Background Technology In the field of digital human technology, especially in digital virtual doctor applications aimed at medical and health education, configuring unique and fitting personas (i.e., "character designs") for different virtual doctors is key to generating differentiated content. In related technologies, such persona configuration data typically relies on manual design and input by operators based on their understanding of the target doctor's background.

[0002] However, in related technologies, since the configuration process relies on decentralized manual operations, when constructing roles for a large number of digital virtual doctors, there are cases where the role configuration data overlaps or is highly similar in personality traits, making it impossible to effectively distinguish between different digital virtual doctors, thereby affecting the diversity and uniqueness of the overall content library. Summary of the Invention

[0003] In view of this, one or more embodiments of this application provide a method and related apparatus for generating character configuration data, which can enhance the uniqueness of the personality of a digital virtual doctor to a certain extent.

[0004] In a first aspect, one or more embodiments of this application propose a method for generating character configuration data for a digital virtual doctor, comprising: obtaining primary character configuration data corresponding to the digital virtual doctor; wherein the primary character configuration data is used to characterize the personality traits of the role to be set for the digital virtual doctor; matching the primary character configuration data in a specified set of doctor role configuration data, and, if no doctor role configuration data associated with the primary character configuration data is obtained, using the primary character configuration data as the target task role configuration data corresponding to the digital virtual doctor; wherein the specified set of doctor role configuration data includes multiple doctor role configuration data; different doctor role configuration data respectively characterize the personality traits of the role set for different digital virtual doctors.

[0005] Secondly, one or more embodiments of this application propose an apparatus for generating character role configuration data for a digital virtual doctor, comprising: an acquisition module, configured to acquire primary character role configuration data corresponding to the digital virtual doctor; wherein the primary character role configuration data is used to characterize the personality traits of the role to be set for the digital virtual doctor; and a matching module, configured to match the primary character role configuration data in a specified set of doctor role configuration data, and, if no doctor role configuration data associated with the primary character role configuration data is obtained, to use the primary character role configuration data as the target task role configuration data corresponding to the digital virtual doctor; wherein the specified set of doctor role configuration data includes multiple doctor role configuration data; different doctor role configuration data respectively characterize the personality traits of the roles set for different digital virtual doctors.

[0006] Thirdly, one or more embodiments of this application provide a computer device including a memory and a processor, wherein the memory stores at least one computer program, which is loaded and executed by the processor to implement the method as described above.

[0007] Fourthly, one or more embodiments of this application provide a computer program product including computer instructions that, when executed by a processor, implement the method as described above.

[0008] Fifthly, one or more embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the processor to implement the method as described above.

[0009] As can be seen from the above embodiments, multiple embodiments in this application obtain primary character configuration data and match it in a specified doctor character configuration data set. When no related data is matched, the primary data is used as the target task character configuration data, thereby achieving the uniqueness screening of the personality characteristics of the digital virtual doctor character and achieving the effect of improving the uniqueness of the personality of the digital virtual doctor character. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating a method for generating character configuration data for a digital virtual doctor according to one embodiment of this application.

[0011] Figure 2 This is a schematic diagram of a module for generating character configuration data for a digital virtual doctor, provided in one embodiment of this application.

[0012] Figure 3 This is a schematic diagram of a computer device provided in one embodiment of this application. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments.

[0014] In the description of the embodiments of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0015] 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, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant 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.

[0016] In related technologies, digital virtual doctors for medical and health science popularization scenarios typically require pre-configured character profile data to represent the personality traits intended for the digital virtual doctor, such as rigorous, approachable, humorous, or authoritative. Specifically, operators often manually write and input the corresponding character profile data based on their understanding of the target doctor's background information, areas of expertise, and platform positioning. This ensures that the digital virtual doctor reflects the corresponding character style when generating science popularization content or interacting with users. While this primarily manual design approach can basically meet the basic character profile requirements when the number of virtual doctors is small, the quality and distinctiveness of the character profile data highly depend on the subjective experience and effort invested by the operators. As the number of digital virtual doctors to be supported gradually increases, the efficiency and consistency of manual configuration become difficult to guarantee.

[0017] With the rapid development of digital human technology and artificial intelligence technologies such as large language models, some systems have begun to explore using pre-set script templates, tag libraries, or general generation models to generate character configuration data for different digital virtual doctors in batches, reducing the workload of relying entirely on manual, one-by-one writing. In such solutions, different combinations of personality tags or slight adjustments to descriptive text are often used to create multiple seemingly different character configuration data sets, thus supporting the deployment and use of more digital virtual doctors. The introduction of large language models has improved the efficiency of character configuration data production to some extent, enabling platforms to quickly complete the basic character configuration information for more digital virtual doctors.

[0018] However, despite the introduction of templated configurations or generative modeling techniques, the diversity of character configuration data in terms of personality traits remains insufficient when building personas for a large number of digital virtual doctors. On the one hand, under time pressure or template constraints, operators often tend to reuse existing persona configuration data, making only minor textual modifications, resulting in highly similar personality settings across different digital virtual doctors. On the other hand, general generative models, lacking fine-grained constraints, may repeatedly generate content around a few common personality descriptions, making it difficult to effectively differentiate between multiple digital virtual doctors in terms of personality traits and expression styles. As the scale of digital virtual doctors on the platform expands, this repetition or high similarity in personality traits will become increasingly prominent, leading to homogenization of the overall content library in terms of presentation and character style, affecting users' perception of differences between different digital virtual doctors and their content experience.

[0019] In summary, the related technologies still suffer from the problem that the character configuration data of digital virtual doctors lacks sufficient uniqueness and distinctiveness in terms of character personality traits, which needs further improvement.

[0020] In several embodiments provided in this application, a method for generating character configuration data for a digital virtual doctor can be applied to electronic devices with certain computing power and network access capabilities. This electronic device can be a desktop computer, laptop computer, tablet computer, smartphone, or a server. Specifically, the electronic device includes a processor, memory, and a network access module for network communication. The server can be an electronic device with strong data processing capabilities; of course, a server can also refer to a server cluster formed by multiple electronic devices, or a quantum server built using a quantum computer.

[0021] One embodiment of this application provides an application scenario example of a method for generating digital virtual doctor character configuration data. The generation device can be deployed on the server side of an internet healthcare platform to automatically generate character configuration data for multiple digital virtual doctors on the platform. Through uniqueness filtering and script mapping, it drives a digital human video script generation engine to generate personalized digital human video scripts. In this scenario example, taking a consumer healthcare platform as an example, the platform provides contracted doctors with digital virtual doctor images for continuously outputting medical and health science popularization content in homepage recommendations, short video feeds, science popularization columns, and in-hospital e-commerce shopping guides. The platform aims for different digital virtual doctors to have clear and distinct character settings.

[0022] For example, the platform recently launched a new digital virtual doctor, "Dr. Zhang," specializing in endocrinology. The corresponding offline doctor is an associate chief physician in the endocrinology department of a certain hospital, specializing in the diagnosis and treatment of type 2 diabetes and gestational diabetes. The operators complete the basic onboarding for Dr. Zhang in the "Doctor Management" backend, only needing to enter the doctor's identity identifier, title, department, practice direction, and expected content format (such as "short video science popularization," "text and image product recommendations," etc.). Then, a command to generate the character configuration data with one click is triggered, and the generation device immediately starts the character configuration data generation method provided in this application.

[0023] In this scenario example, the generation device first acquires input data related to Dr. Zhang from multiple domain-related data sources. This input data collectively constitutes the joint features used for content generation decisions. On one hand, the generation device obtains Dr. Zhang's professional title, department, and practice direction information from the doctor's basic information service and transaction configuration system, using these as static attribute features to define the basic scope of Dr. Zhang's digital virtual doctor content generation, such as "Endocrinology Department," "Associate Chief Physician," and "Specializes in long-term blood glucose management of type 2 diabetes and follow-up of gestational diabetes." On the other hand, the generation device mines hot disease keywords and historical popular content keywords related to the endocrinology field from the content operation platform and content effect statistics system based on historical content performance data, using these as content-oriented features, such as "Type 2 Diabetes Dietary Misconceptions," "Blood Glucose Fluctuations and Complication Risks," and "Three Common Reasons for Failure to Control Blood Sugar." Through the above acquisition process, the joint features simultaneously include static attribute features and content-oriented features, ensuring that the subsequently generated character configuration data reflects both Dr. Zhang's professional identity and the platform's operational orientation at the content level.

[0024] After obtaining the joint features, the generation device constructs decision instructions for invoking the large language model based on these features. Specifically, the generation device first selects a decision logic template corresponding to the current task from a predefined template library based on Dr. Zhang's identity and the target content scenario type (e.g., "endocrine short video science popularization"). This decision logic template defines the structure of the character configuration data to be output in the short video scenario, specifying multiple dimensions of configuration fields to be generated, such as disease direction fields, angle fields, key introduction fields, and tone style fields. Subsequently, the generation device fills the static attribute features and content-oriented features into the corresponding positions of the template according to the logical relationship defined in the decision logic template, forming a structured decision instruction. This allows the large language model to simultaneously perceive Dr. Zhang's professional positioning, the target content scenario type, and the disease and topic directions that the platform wants to focus on when receiving the decision instruction.

[0025] The generation device invokes a large-scale language model using the aforementioned decision instructions. This large-scale language model is pre-trained on large-scale general corpora and medical and health domain corpora, and can optionally be fine-tuned using medical scenario samples. It possesses the ability to perform collaborative semantic modeling and generation of doctor identity attributes, disease themes, audience needs, and content style preferences. The large-scale language model performs collaborative analysis of joint features according to the decision instructions, outputting structured persona configuration data. Taking Dr. Zhang as an example, the generation device can obtain a structured persona configuration data, where the disease direction field includes "long-term management of type 2 diabetes," "glucose monitoring during gestational diabetes," and "comprehensive management of patients with hypertension," etc.; the approach field includes "starting from common misconceptions about blood sugar control," "explaining dietary principles through daily dietary scenarios," and "reminding patients of complication risks by combining real follow-up cases," etc.; the key information field includes "frequency and timing of pre- and post-meal glucose monitoring," "key points for staple food replacement and carbohydrate estimation," and "principles of safe medication use during pregnancy," etc.; and the tone and style field is "professional and reliable but not serious, mainly reassuring and encouraging, with appropriate use of everyday metaphors." The generation device uses this structured character configuration data as the primary character configuration data corresponding to Dr. Zhang's digital virtual doctor, which is used to characterize the personality traits and content expression habits of the role to be set for the digital virtual doctor.

[0026] Subsequently, to avoid Dr. Zhang's initial character configuration data being highly similar in personality traits to the existing digital virtual doctor character configuration data on the platform, the generation device matches the aforementioned initial character configuration data against a designated doctor character configuration data set. This designated doctor character configuration data set can be viewed as a historical database of doctor character configuration data maintained by the platform, storing the established target task character configuration data for digital virtual doctors in other departments and within the same department. The generation device inputs Dr. Zhang's initial character configuration data into a text encoding model to obtain the corresponding semantic vector representation; simultaneously, each doctor character configuration data in the designated doctor character configuration data set is also pre-processed using the same text encoding model to obtain its own semantic vector representation. The generation device uses cosine similarity and other measurement methods to perform similarity matching between the semantic vector representation of the primary character configuration data and the semantic vector representation of the existing doctor character configuration data in the set. When it is found that the similarity between a certain existing doctor character configuration data and the current primary character configuration data exceeds a preset specified similarity threshold, the generation device determines that the current primary character configuration data is highly similar to the doctor character configuration data in terms of character personality traits and configuration field combinations, that is, it determines that the doctor character configuration data associated with the primary character configuration data has been obtained.

[0027] If the associated doctor role configuration data is identified, the generation device will not directly use the current primary character role configuration data. Instead, it will trigger a process to re-acquire the primary character role configuration data corresponding to the digital virtual doctor. Specifically, when constructing a new round of decision instructions, the generation device can transform the primary character role configuration data generated in the previous round, or highly similar doctor role configuration data, into negative constraint information or exclusion conditions. For example, it can avoid using the same combination of disease direction fields and angle fields again, or prioritize the introduction of other hot disease keywords and historically popular content keywords related to endocrinology that are not yet fully covered. Based on this, the generation device calls a large language model to generate new structured character role configuration data and uses it as new primary character role configuration data. It then obtains a semantic vector representation through a text encoding model and performs similarity matching with the semantic vector representations in the specified doctor role configuration data set. The above "generation-matching-determination-regeneration" process can be iterated multiple times as needed until the current primary character role configuration data is no longer identified as having associated doctor role configuration data in the specified doctor role configuration data set, that is, there is no existing configuration record with a semantic similarity exceeding the specified similarity threshold. At this point, the generation device establishes the current primary character configuration data as the target task character configuration data corresponding to Dr. Zhang's digital virtual doctor, completing the uniqueness screening of character personality traits.

[0028] After establishing the target task role configuration data, the generation device further maps each configuration field in the target task role configuration data to the corresponding input items of the digital human video script generation engine, thereby driving the generation of digital human video scripts with personalized characteristics corresponding to the content of each configuration field. The generation device can map the disease direction field to the script topic configuration input item, which is used to specify which disease scenarios the script will focus on; map the angle of approach field to the content structure configuration input item, which is used to constrain the script to adopt an explanatory structure such as "case story - misconception deconstruction - action suggestions" or "common question and answer style"; map the key points field to the knowledge point configuration input item, which is used to prompt the core medical knowledge that must be emphasized in the script; and map the tone style field to the tone style configuration input item, which is used to control the script to present a style such as "professional and friendly, emphasizing long-term companionship management" in terms of language expression. After receiving the above input items, the digital human video script generation engine can automatically generate multiple video scripts that match the positioning of Dr. Zhang's digital virtual doctor role. For example, script texts around themes such as "Why does blood sugar fluctuate repeatedly despite hard work in controlling blood sugar" and "What are the most common misconceptions about blood sugar control during pregnancy?" The script structure, knowledge point selection, and wording style fully follow the configuration fields in the target task role configuration data.

[0029] In this scenario example, the generation device acquires joint features, constructs decision instructions, and calls a large language model to generate structured character configuration data. Then, by combining similarity matching based on text encoding models and semantic vector representations with a multi-round regeneration mechanism, it achieves unique screening of the digital virtual doctor character configuration data in the dimension of character personality characteristics. The final target task character configuration data is then directly mapped to the input items of the digital human video script generation engine. Thus, in the scenario of combining Internet healthcare and e-commerce, it automatically generates digital human video scripts that are both consistent with the doctor's professional identity and have a differentiated content style.

[0030] Please see Figure 1 One embodiment of this application provides a method for generating character configuration data for a digital virtual doctor. This method can be applied to a generation device, which can be an electronic device with certain computing power and network access capabilities. Of course, in some embodiments, the generation device can also be software running on an electronic device. A method for generating literature summary text may include the following steps.

[0031] Step S110: Obtain the primary character configuration data corresponding to the digital virtual doctor; wherein, the primary character configuration data is used to characterize the personality traits of the role to be set for the digital virtual doctor.

[0032] Step S120: Match the primary character configuration data with the specified doctor character configuration data set, and if no doctor character configuration data associated with the primary character configuration data is obtained, use the primary character configuration data as the target task character configuration data corresponding to the digital virtual doctor; wherein, the specified doctor character configuration data set includes multiple doctor character configuration data; different doctor character configuration data respectively represent the character personality characteristics set by different digital virtual doctors.

[0033] In this embodiment, the generation device can automatically generate character configuration data for different digital virtual doctors for subsequent content generation and interaction. The character configuration data is used to characterize the personality traits of the proposed digital virtual doctor, such as rigorous, approachable, humorous, and authoritative, to constrain the overall style of the digital virtual doctor in medical and health science popularization scenarios.

[0034] In this embodiment, the generation device can acquire the primary character configuration data corresponding to the digital virtual doctor. The primary character configuration data is used to characterize the personality traits intended for the target digital virtual doctor, and may include at least personality tags describing the digital virtual doctor's role positioning, expression style descriptions, and personality orientation information related to the target content scene. Specifically, when the digital virtual doctor completes basic registration or initial configuration, the generation device can generate one or more primary character configuration data associated with the digital virtual doctor based on the digital virtual doctor's identity information and preset role configuration rules. In some embodiments, the primary character configuration data can also be input by operators through a configuration interface, or generated and provided by other transaction systems based on historical content performance and user feedback. This embodiment does not limit the specific generation method of the primary character configuration data, only requiring that the primary character configuration data can be used to characterize the personality traits intended for the digital virtual doctor and can serve as input for subsequent role configuration uniqueness filtering.

[0035] In this embodiment, to avoid duplication of character personality configurations among different digital virtual doctors, the generation device can match the initial character configuration data with a designated doctor character configuration data set. The designated doctor character configuration data set can store multiple doctor character configuration data sets, each associated with a digital virtual doctor, representing the established character configuration content of the corresponding digital virtual doctor. Different doctor character configuration data sets describe the character personality traits set by different digital virtual doctors, such as differences in disease direction, content expression style, communication tone, and information presentation preferences. During matching, the generation device can determine the correlation between the initial character configuration data and each doctor character configuration data set in the designated doctor character configuration data set to determine whether there is any doctor character configuration data that is the same as or highly similar to the initial character configuration data. In this embodiment, the correlation determination can be based on preset matching rules or other implementation methods; the specific implementation is not limited.

[0036] In this embodiment, when the generation device does not obtain doctor role configuration data associated with the primary character role configuration data in the specified doctor role configuration data set, the primary character role configuration data can be directly established as the target task role configuration data corresponding to the target digital virtual doctor. The target task role configuration data represents the target character role configuration result that has passed uniqueness screening and can be referenced by subsequent digital human content generation engines, script generation engines, or interactive dialogue systems to drive the target digital virtual doctor to output content according to predetermined role personality traits in specific medical and health science popularization scenarios. Through the above processing flow, based on the generation device obtaining the primary character role configuration data and matching it in the specified doctor role configuration data set, the primary character role configuration data is established as the target task role configuration data only when no associated doctor role configuration data is matched. This allows the character role configuration data of the digital virtual doctor to pass uniqueness screening at the level of role personality traits, thereby improving the uniqueness and distinguishability of different digital virtual doctor character personality traits.

[0037] In this application, multiple embodiments obtain primary character configuration data and match it in a specified doctor character configuration data set. When no related data is matched, the primary data is used as the target task character configuration data, thereby achieving the uniqueness screening of the personality characteristics of the digital virtual doctor character and achieving the effect of enhancing the uniqueness of the digital virtual doctor character's personality.

[0038] In some implementations, the generation device can acquire input data associated with the digital virtual doctor from multiple domain-related data sources; wherein the input data collectively constitutes joint features for content generation decisions, the joint features including: static attribute features defining the basic scope of the digital virtual doctor's content generation, and content orientation features derived from historical content performance data; based on the joint features, a decision instruction for invoking a large language model is constructed, wherein the decision instruction causes the large language model to perform collaborative analysis of the static attribute features and the content orientation features to generate structured character role configuration data matching the role positioning of the digital virtual doctor; the decision instruction is used to invoke the large language model, and the structured character role configuration data generated by the large language model is used as the initial character role configuration data.

[0039] In this embodiment, to automatically generate the initial character configuration data corresponding to the digital virtual doctor, the generation device can first acquire input data associated with the target digital virtual doctor from multiple domain-related data sources. These multiple domain-related data sources may include a doctor's basic information database, a transaction configuration system, a content operation platform, and a content effect statistics system, etc. The acquired input data collectively constitutes the joint features used for content generation decisions. The joint features include at least two types of information: one is static attribute features, used to define the basic scope of digital virtual doctor content generation, such as attribute information related to professional title, department, practice direction, and platform positioning; the other is content-oriented features, used to characterize the digital virtual doctor's preferences or advantages at the content level, such as trending diseases, common entry points, or high-conversion topics derived from historical content performance data. Through the combination of static attribute features and content-oriented features, the joint features can simultaneously reflect the digital virtual doctor's identity positioning and content orientation.

[0040] In this embodiment, the generation device can construct a decision instruction for invoking a large language model based on joint features. The decision instruction can be a structured task instruction, used to embed static attribute features and content-oriented features into the request to invoke the large language model according to a predetermined format, enabling the large language model to perform collaborative analysis of the two types of features when generating results. Specifically, the decision instruction may include a target description describing the positioning of the digital virtual doctor role, field-based expressions of each element in the joint features, and constraint information on the structure of the output results, to guide the large language model to output character configuration data that conforms to the expected structure. The large language model can be a large-scale pre-trained model that has been pre-trained and / or fine-tuned on large-scale general corpora and medical and health-related corpora, possessing the ability to semantically model and generate doctor identity information, disease themes, and content style requirements, thereby supporting the intelligent generation of character configuration data based on the decision instruction. In this embodiment, the specific model architecture of the large language model is not improved, so the specific algorithm architecture of the large language model will not be described in detail. Those skilled in the art can select a large language model with an appropriate algorithm architecture based on the known technology, without further elaboration.

[0041] In this embodiment, the generation device can invoke a large language model using the decision instruction. The large language model analyzes the joint features according to the decision instruction and generates structured character configuration data. The structured character configuration data may include multiple configuration fields for describing character personality traits and content orientation, and is organized in a predefined data structure for easy subsequent processing.

[0042] In some implementations, the generating device can select a decision logic template from a predefined template library based on the identity of the digital virtual doctor and the target content scene type, wherein the target content scene type is used to characterize the category of content to be generated by the digital virtual doctor; and fill the static attribute features and the content guidance features into the corresponding positions of the decision logic template according to the logical association defined in the decision logic template to form the decision instruction.

[0043] In this embodiment, after acquiring the joint features associated with the target digital virtual doctor, the generation device can construct decision instructions for invoking a large language model based on these joint features. To this end, the generation device can pre-maintain a template library containing multiple decision logic templates. These templates constrain how the input information related to the digital virtual doctor and the output format of the large language model are organized to adapt to different usage scenarios. Specifically, the generation device can select a decision logic template matching the current task from the template library based on the digital virtual doctor's identity and the target content scenario type. The identity is used to uniquely identify the target digital virtual doctor, and the target content scenario type characterizes the category of content to be generated by the digital virtual doctor, such as short video science popularization, graphic interpretation, Q&A replies, or live scripts. Different decision logic templates corresponding to different target content scenario types can differentiate the focus of input elements and the output structure. This allows the subsequently constructed decision instructions to be customized for different content scenarios.

[0044] In this embodiment, after determining the target decision logic template, the generation device can fill the static attribute features and content-oriented features from the joint features into the corresponding positions in the template according to the logical relationships defined in the decision logic template, thereby forming a decision instruction. The decision logic template can predefine the display order, grouping method, and logical dependencies of different features in the decision instruction. For example, static attribute features can be used to describe the basic positioning of the digital virtual doctor's role, and content-oriented features can be used to highlight the differentiated advantages of the digital virtual doctor in terms of content direction and expression preferences. The above features can be embedded into the same task description through text placeholders or structured fields. The decision instruction obtained by the generation device after filling in the features according to the decision logic template can be used as input for calling a large language model. This allows the large language model to perform collaborative analysis of static attribute features and content-oriented features, based on an understanding of the digital virtual doctor's identity and the target content scene type, thereby generating structured character configuration data that matches the target digital virtual doctor's role positioning.

[0045] In some implementations, the static attribute features include at least one of the professional title information, department information, and practice direction information corresponding to the digital virtual doctor; the content orientation features include at least one of the hot disease keywords and historical popular content keywords determined based on historical content performance data; the structured character configuration data includes configuration fields of multiple dimensions, and the configuration fields of multiple dimensions include at least: disease direction field, angle of approach field, key introduction field, and tone style field.

[0046] In this embodiment, static attribute features may specifically include at least one of the following: professional title information, department information, and practice direction information corresponding to the digital virtual doctor. Professional title information represents the professional qualification level of the corresponding doctor in the real world, such as chief physician, associate chief physician, attending physician, etc., and can be used to limit the professional authority of the digital virtual doctor in content output. Department information represents the clinical department category to which the digital virtual doctor belongs, such as cardiology, endocrinology, dermatology, etc., and can be used to limit the scope of diseases mainly handled by the digital virtual doctor. Practice direction information represents a more specific treatment direction of the digital virtual doctor within the corresponding department, such as "long-term management of type 2 diabetes" or "prevention and control of myopia in adolescents," and can be used to further narrow the professional focus of content generation. By incorporating at least one of the professional title information, department information, and practice direction information into the static attribute features, the joint features can provide clear identity positioning constraints for character configuration when calling large-scale language models.

[0047] In this implementation, content-oriented features can specifically include at least one of the following: trending disease keywords determined based on historical content performance data and historical best-selling content keywords. Trending disease keywords can be statistically analyzed based on metrics such as platform historical content views, completion rates, interaction rates, and conversion rates to extract disease names or disease-related topics with high user interest, representing disease areas that are more worthy of focus at the current stage. Historical best-selling content keywords can be extracted from high-performing content samples in the past, such as titles, hashtags, or frequently occurring core phrases, representing highly attractive content entry points that have been validated in existing operational practices. By incorporating trending disease keywords and historical best-selling content keywords into content-oriented features, the combined features not only reflect the professional identity of the digital virtual doctor but also its potential advantages in content operation, thereby guiding the large-scale language model to balance medical professionalism and content effectiveness when generating character configuration data.

[0048] In this embodiment, after receiving joint features organized based on decision instructions, the large-scale language model can generate structured character configuration data. This structured character configuration data includes configuration fields across multiple dimensions, at least including disease direction, perspective, key information, and tone / style fields. The disease direction field records diseases or combinations of diseases suitable for the digital virtual doctor's focus, such as "gestational diabetes follow-up management" or "long-term hypertension control," and corresponds to the aforementioned department information, practice direction information, and hot disease keywords. The perspective field characterizes the recommended expression perspective for the digital virtual doctor when creating content or providing popular science explanations, such as "complication risk reminder," "lifestyle intervention suggestion," or "medication misconception clarification." The key information field identifies knowledge points or information modules that should be highlighted in specific content generation, such as "initial examination items," "core principles of dietary control," and "follow-up visit time points." The tone / style field constrains the language style of the digital virtual doctor's output content, such as "rigorous and professional," "friendly and reassuring," "humorous and relaxed," or "authoritative explanation." By combining the above-mentioned multiple dimensions of configuration fields, the structured character configuration data can meticulously depict the personality traits of the digital virtual doctor from four aspects: the scope of disease content, the content organization method, the information focus, and the expression style. This data can then be directly referenced by the subsequent script generation engine or content generation module as primary character configuration data or target task character configuration data, driving the digital virtual doctor to output differentiated content consistent with its role positioning in internet healthcare and e-commerce scenarios.

[0049] In some implementations, the generating device may input the primary character configuration data into a text encoding model to obtain a semantic vector representation of the primary character configuration data; perform similarity matching between the semantic vector representation of the primary character configuration data and the semantic vector representation of existing doctor character configuration data in the specified doctor character configuration data set; and if the similarity exceeds a specified similarity threshold, identify the doctor character configuration data associated with the primary character configuration data.

[0050] In this embodiment, after the generation device obtains the initial character configuration data corresponding to the digital virtual doctor, in order to determine whether the initial character configuration data overlaps or is highly similar to existing character configuration content in terms of character personality characteristics, the generation device can match the initial character configuration data with a specified doctor character configuration data set. For this purpose, the generation device can use a text encoding model to perform semantic modeling on the character configuration data. The text encoding model can be a deep learning-based text representation model, used to map the input text data into semantic vector representations in a high-dimensional vector space, making semantically similar texts close to each other in the vector space, thereby facilitating subsequent calculation and comparison of semantic similarity based on numerical operations.

[0051] In this embodiment, the generation device can input the initial character configuration data into a text encoding model to obtain a semantic vector representation of the initial character configuration data. The semantic vector representation is used to numerically characterize the overall semantic information of the character personality traits and related configuration fields contained in the initial character configuration data. Each existing piece of doctor character configuration data in the specified doctor character configuration data set can also be processed in advance or near real-time through the text encoding model to obtain its corresponding semantic vector representation, and stored in a vector index structure or other searchable data structure, so that it can be uniformly measured with the semantic vector representation of the initial character configuration data during the matching stage.

[0052] In this embodiment, the generation device can perform similarity matching between the semantic vector representation of the initial character configuration data and the semantic vector representation of existing doctor character configuration data in a specified doctor character configuration data set. Similarity matching can be based on cosine similarity, Euclidean distance, or other vector similarity measures to calculate the closeness between the initial character configuration data and each doctor character configuration data in the semantic space. The generation device can pre-set a specified similarity threshold to define the boundary for judging "semantically high similarity." When the similarity between the semantic vector representation of any existing doctor character configuration data and the semantic vector representation of the initial character configuration data exceeds the specified similarity threshold, the generation device can determine that there already exists doctor character configuration data associated with the initial character configuration data in the specified doctor character configuration data set; that is, there is at least one existing configuration record that is semantically highly similar to the initial character configuration data in terms of character personality traits and configuration dimensions. Through the above-mentioned similarity matching mechanism based on text encoding models and semantic vector representations, the matching process of character configuration data is elevated from the literal text level to the semantic level, providing a reliable basis for subsequent unique screening of digital virtual doctor character configuration data.

[0053] In some implementations, the generating device may, upon determining that doctor role configuration data associated with the primary character role configuration data has been obtained, re-execute the step to obtain primary character role configuration data corresponding to the digital virtual doctor; wherein the re-obtained primary character role configuration data is at least partially different from the primary character role data in the previous round.

[0054] In this embodiment, when the doctor role configuration data associated with the primary character role configuration data is obtained through text encoding model and semantic vector representation similarity matching, in order to prevent the digital virtual doctor from using configuration content that is highly similar to the existing doctor role configuration data in terms of character personality traits, the generation device can further execute a process of re-acquiring the primary character role configuration data. Specifically, when the similarity between the semantic vector representation of the primary character role configuration data and the semantic vector representation of at least one doctor role configuration data in the specified doctor role configuration data set exceeds a specified similarity threshold, the generation device determines that the current primary character role configuration data fails to meet the requirement of unique character personality traits, thereby triggering the regeneration of the primary character role configuration data for the digital virtual doctor.

[0055] In this embodiment, when reacquiring the initial character configuration data, the generation device can again construct decision instructions based on the joint features associated with the digital virtual doctor, and call a large language model to generate new structured character configuration data, using this new structured character configuration data as the new initial character configuration data. Compared with the initial character configuration data generated in the previous round, the new initial character configuration data differs from at least some of the configuration fields. For example, at least one of the disease direction field, approach angle field, key introduction field, or tone style field is adjusted, thus making the character personality traits of the new initial character configuration data different from the previous round. In some embodiments, when constructing new decision instructions, the generation device can embed the initial character configuration data from the previous round or the associated doctor character configuration data as negative constraints or exclusion conditions into the decision instructions, so as to guide the large language model to avoid excessive overlap with existing configurations when generating new structured character configuration data.

[0056] In this embodiment, the generation device can iteratively execute the above-mentioned process of "reacquiring primary character configuration data - semantic similarity matching" until no doctor character configuration data associated with the current primary character configuration data is identified in the specified doctor character configuration data set. At this point, the current primary character configuration data can be used as the target task character configuration data corresponding to the target digital virtual doctor, and enter the configuration establishment process after the aforementioned uniqueness screening is completed. Through the above-mentioned primary character configuration data regeneration mechanism driven by similarity judgment results, the character configuration data of the digital virtual doctor can gradually move away from the semantic distribution of the existing doctor character configuration data through multiple rounds of constraints and iterative optimization during the generation process, thereby further improving the uniqueness and distinguishability of the target task character configuration data in the dimension of character personality characteristics.

[0057] In some implementations, the generating device can map each configuration field in the target task role configuration data to the corresponding input item of the digital human video script generation engine to drive the generation of a digital human video script with personalized features corresponding to the content of each configuration field.

[0058] In this embodiment, the generation device can further use the target task role configuration data to drive the digital human content production chain. Specifically, the generation device can map each configuration field included in the target task role configuration data to the corresponding input items of the digital human video script generation engine. The digital human video script generation engine can be a software module running in an electronic device, used to automatically generate video script text for driving the digital human image to broadcast based on structured input parameters. Its input items can include content theme parameters, content organization parameters, and expression style parameters, etc., used to control the content scope and presentation of the generated script.

[0059] In this embodiment, the generation device can map the disease direction field, approach angle field, key introduction field, and tone style field in the target task role configuration data to the theme configuration input item, content structure input item, key information input item, and tone style input item in the digital human video script generation engine, respectively, based on a pre-configured mapping relationship. For example, the disease direction field can serve as a script theme configuration parameter to limit the disease or combination of diseases the script revolves around; the approach angle field can serve as a content structure configuration parameter to indicate the explanation perspective adopted by the script; the key introduction field can serve as a key information configuration parameter to indicate the knowledge points that need to be prioritized in the script; and the tone style field can serve as a tone style configuration parameter to constrain the language style and expression method adopted by the script. After receiving the input items obtained from the mapping of the above configuration fields, the digital human video script generation engine can automatically combine and generate the paragraph structure, knowledge point arrangement, language style, and emotional tone of the script according to the corresponding input meaning, thereby outputting a digital human video script with personalized characteristics corresponding to the content of each configuration field. Through the above field-level mapping and script generation process, the target task role configuration data is not just a static description of the character, but can directly drive the digital human video script generation engine to generate differentiated video scripts that match the character's personality traits and content orientation in actual Internet medical scenarios.

[0060] Please see Figure 2 One or more embodiments of this application also provide an apparatus for generating character configuration data for a digital virtual doctor. The apparatus for generating character configuration data may include: an acquisition module and a matching module.

[0061] The acquisition module can be used to acquire the primary character configuration data corresponding to the digital virtual doctor; wherein, the primary character configuration data is used to characterize the personality traits of the role to be set for the digital virtual doctor.

[0062] The matching module can be used to match the primary character configuration data with a specified set of doctor character configuration data, and if no doctor character configuration data associated with the primary character configuration data is obtained, the primary character configuration data can be used as the target task character configuration data corresponding to the digital virtual doctor; wherein, the specified set of doctor character configuration data includes multiple doctor character configuration data; different doctor character configuration data respectively represent the character personality characteristics set by different digital virtual doctors.

[0063] In this embodiment, the functions and effects of the device for generating the character configuration data of the digital virtual doctor can be explained in comparison with the aforementioned embodiments, and will not be repeated here.

[0064] Please see Figure 3This application also provides a computer device comprising: a memory and a processor, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the method described above.

[0065] The memory, processor, and communication interface in the computer device can communicate with each other via the system bus and network communication.

[0066] In this embodiment, the functions and effects implemented by the computer device can be explained by referring to the foregoing embodiments, and will not be repeated here.

[0067] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the processor to implement the method as described above.

[0068] The functions and effects achieved in this embodiment can be explained by referring to other embodiments, and will not be repeated here.

[0069] This application also provides a computer program product containing instructions, including a computer program / instructions that, when executed by a processor, implement the method as described above.

[0070] The functions and effects achieved in this embodiment can be explained by referring to other embodiments, and will not be repeated here.

[0071] It is understood that the specific examples in this document are only intended to help those skilled in the art better understand the embodiments of this application, and are not intended to limit the scope of the invention.

[0072] It is understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0073] It is understood that the various implementation methods described in this application can be implemented individually or in combination, and the implementation methods in this application are not limited in this respect.

[0074] Unless otherwise stated, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this application includes any and all combinations of one or more of the associated listed items. The singular forms "a," "the," and "the" as used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0075] It is understood that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0076] It is understood that the memory in the embodiments of this application may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Specifically, non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM). It should be noted that the memory in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0077] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0078] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the aforementioned method implementations, and will not be repeated here.

[0079] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0080] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0081] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0082] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0083] The above description is merely a specific embodiment of this application, but the scope of protection of this invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A method for generating character configuration data for a digital virtual doctor, characterized in that, include: Obtain the primary character configuration data corresponding to the digital virtual doctor; wherein, the primary character configuration data is used to characterize the personality traits of the role to be set for the digital virtual doctor; The primary character configuration data is matched against a specified set of doctor character configuration data. If no doctor character configuration data associated with the primary character configuration data is found, the primary character configuration data is used as the target task character configuration data corresponding to the digital virtual doctor. The specified set of doctor character configuration data includes multiple doctor character configuration data. Different doctor character configuration data represent the character personality traits set by different digital virtual doctors.

2. The method according to claim 1, characterized in that, The process of obtaining the basic character configuration data corresponding to the digital virtual doctor includes: The system acquires input data associated with the digital virtual doctor from multiple domain-related data sources; wherein the input data collectively constitutes joint features for content generation decisions, the joint features including: static attribute features for defining the basic scope of content generation for the digital virtual doctor, and content orientation features derived from historical content performance data; Based on the joint features, a decision instruction for invoking a large language model is constructed, wherein the decision instruction enables the large language model to perform a collaborative analysis of the static attribute features and the content-oriented features to generate structured character configuration data that matches the role positioning of the digital virtual doctor. The decision instruction is used to invoke the large language model, and the structured character configuration data generated by the large language model is used as the primary character configuration data.

3. The method for generating character configuration data for a digital virtual doctor according to claim 2, characterized in that, Based on the joint features, a decision instruction for invoking a large language model is constructed, including: Based on the identity of the digital virtual doctor and the target content scene type, a decision logic template is selected from a predefined template library, wherein the target content scene type is used to characterize the category of content to be generated by the digital virtual doctor; The static attribute features and the content guidance features are filled into the corresponding positions of the decision logic template according to the logical association defined in the decision logic template to form the decision instruction.

4. The method for generating character configuration data for a digital virtual doctor according to claim 2, characterized in that, The static attribute features include at least one of the following: the professional title information, department information, and practice direction information corresponding to the digital virtual doctor; The content-oriented features include at least one of the following: hot disease keywords determined based on historical content performance data and keywords of historically popular content. The structured character configuration data includes configuration fields of multiple dimensions, including at least: disease direction field, angle of approach field, key introduction field, and tone style field.

5. The method for generating character configuration data for a digital virtual doctor according to claim 1, characterized in that, Matching the aforementioned basic character configuration data with the specified doctor character configuration data set, including: The basic character configuration data is input into a text encoding model to obtain a semantic vector representation of the basic character configuration data. Perform similarity matching between the semantic vector representation of the primary character configuration data and the semantic vector representation of the existing doctor character configuration data in the specified doctor character configuration data set; If the similarity exceeds a specified similarity threshold, doctor role configuration data associated with the primary character role configuration data is identified.

6. The method for generating character configuration data for a digital virtual doctor according to claim 5, characterized in that, The method further includes: If the doctor role configuration data associated with the primary character role configuration data is obtained, the step of obtaining the primary character role configuration data corresponding to the digital virtual doctor is repeated; wherein the re-obtained primary character role configuration data is at least partially different from the primary character role data in the previous round.

7. The method for generating character configuration data for a digital virtual doctor according to claim 1, characterized in that, The method further includes: The configuration fields in the target task role configuration data are mapped to the corresponding input items of the digital human video script generation engine to drive the generation of digital human video scripts with personalized features corresponding to the content of each configuration field.

8. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, causes the processor to implement the method as described in any one of claims 1 to 7.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to implement the method as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, Includes computer instructions that, when executed by a processor, implement the method as described in any one of claims 1 to 7.

Images