A biological data simulation method, device, and medium based on a large language model
By using a large language model to identify and match the user's natural language requests and biological data, the target capability interface is obtained and targeted experience text is retrieved. This solves the problem of over-generalization in the selection of analysis tools in existing technologies and achieves high accuracy and reliability of biological data simulation results.
Patent Information
- Application Number
- CN202611253026.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-18
- Publication Date
- 2026-11-10
AI Technical Summary
In existing biological data simulation methods based on large language models, the single semantic vector retrieval mechanism is prone to overgeneralization in the selection of analysis tools, resulting in low accuracy in the selection and arrangement of analysis tools.
By using a large language model to perform intent recognition on user-input natural language requests and biological data, structured intent recognition results are obtained and matched with the initial capability interface of the AI virtual cell model to obtain the target capability interface. The key feature values of the target capability interface are used to recall targeted target orchestration experience text from the orchestration experience database to generate interface orchestration results. The target capability interface is then called sequentially to obtain interface response data. Finally, biological data simulation results are generated through semantic parsing.
It significantly improves the accuracy and reliability of the orchestration results of the large language model generation interface, avoids overgeneralization of recall results, and improves the accuracy of biological data simulation results.
Smart Images

Figure CN123072508A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of large language models, and in particular to a method, apparatus and medium for simulating biological data based on large language models. Background Technology
[0002] In recent years, the rapid development of single-cell omics, high-throughput perturbation experiments, space omics, and multimodal biological data has required researchers to perform multiple tasks in a single analysis process, including data parsing, cell state identification, differential analysis, virtual perturbation inference, result interpretation, and report generation. Traditional bioinformatics platforms typically rely on fixed scripts, workflows, or service interfaces. Users need to know the data format, parameter meanings, analysis steps, and intermediate file locations in advance, and manually transfer results between different tools. With the development of large language model technology, it is now possible to use large language models to understand the natural language requests made by researchers and convert these requests into external tool calls or analysis processes to obtain biological data simulation results. Specifically, this involves: users uploading biological data files and making natural language requests; inputting user requests, historical dialogues, and descriptions of available tools into the language model agent module; the language model agent module selecting one or more analysis tools and generating tool orchestration results; the tool execution module reading user data and running analysis scripts, submitting the required inputs to the AI virtual cell model inference service; and organizing the text, tables, graphs, or file links returned by the AI virtual cell model inference service into user-friendly responses.
[0003] In the above method, when the language model agent module selects one or more analysis tools and generates tool orchestration results, it relies on semantic vector similarity to retrieve several analysis tool selection experiences from the experience database as the decision context of the large language model. However, the single semantic vector retrieval mechanism often suffers from overgeneralization during matching, easily recalling semantically similar but untargeted analysis tool selection experiences. This leads to low quality context, introduces interference noise into the large language model, misleads its reasoning process, and ultimately results in low accuracy of analysis tool selection and tool orchestration results. Summary of the Invention
[0004] To address the aforementioned technical problem, the technical solution adopted in this application is as follows: According to a first aspect of this application, a biological data simulation method based on a large language model is provided, the method comprising the following steps: S1. Perform intent recognition on the user's input natural language request and biometric data through a large language model to obtain structured intent recognition results.
[0005] S2. For each initial capability interface of the AI virtual cell model, match the intent recognition result with the capability description data corresponding to the initial capability interface, and obtain the target capability interface that matches the intent recognition result.
[0006] S3. Based on the key feature values corresponding to the target capability interface, determine at least one target arrangement experience text from the arrangement experience database, which includes several historical arrangement experience texts.
[0007] S4. Input the context information containing the target orchestration experience text into the large language model to obtain the interface orchestration results output by the large language model; the interface orchestration results include the execution order of each target capability interface.
[0008] S5. Call each target capability interface sequentially according to the execution order of each target capability interface in the interface orchestration result to obtain the interface response data returned by each target capability interface.
[0009] S6. Use a large language model to perform semantic parsing and natural language generation on the interface response data to obtain the natural language text corresponding to the biological data simulation results.
[0010] According to a second aspect of this application, a non-transitory computer-readable storage medium is provided, wherein at least one instruction or at least one program is stored in the non-transitory computer-readable storage medium, and the at least one instruction or at least one program is loaded and executed by a processor to implement the above-described biological data simulation method based on a large language model.
[0011] According to a third aspect of this application, an electronic device is provided, including a processor and the aforementioned non-transitory computer-readable storage medium.
[0012] This application has at least the following beneficial effects: This application provides a method, device, and medium for biological data simulation based on a large language model. The method uses a large language model to perform intent recognition on user-input natural language requests and biological data to obtain structured intent recognition results. For each initial capability interface of an AI virtual cell model, the intent recognition result is matched with the capability description data corresponding to the initial capability interface to obtain a target capability interface that matches the intent recognition result. Based on the key feature values corresponding to the target capability interface, at least one target arrangement experience text is determined from an arrangement experience database including several historical arrangement experience texts. Contextual information containing the target arrangement experience text is input into the large language model to obtain the interface arrangement result output by the large language model. Each target capability interface is called sequentially according to its execution order in the interface arrangement result to obtain the interface response data returned by each target capability interface. The interface response data is semantically parsed and natural language generated using the large language model to obtain the natural language text corresponding to the biological data simulation result. As can be seen, this application abstracts the AI virtual cell model into several initial capability interfaces, locks the target capability interface by matching the intent recognition result with the capability description data, and further uses the key feature values of the target capability interface to recall targeted target orchestration experience text from the orchestration experience database, so that the large language model can generate interface orchestration results based on the context information containing the target orchestration experience text; avoiding overgeneralization of the recall results, and significantly improving the accuracy and reliability of the interface orchestration results generated by the large language model. Attached Figure Description
[0013] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0014] Figure 1 A flowchart illustrating a biological data simulation method based on a large language model, provided for an embodiment of this application. Detailed Implementation
[0015] The technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0016] It is worth noting that in the following description, specific details such as particular system structures and techniques are set forth for illustrative purposes rather than for limiting purposes, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary details.
[0017] It should be understood that in this application specification and the appended claims, the terms "comprising," "including," "including but not limited to," "including but not limited to," "mainly composed of," or "mainly made of" all indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.
[0018] It should be understood that in this application specification and the appended claims, the use of the terms "consisting of" or "component of" indicates the presence of the described feature, integral, step, operation, element and / or component, but excludes the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.
[0019] It should be understood that, in this specification and the appended claims, the term "and" indicates a combination in which multiple of the associated listed items exist simultaneously. For example, "A, B, C, and D" means a combination in which "A and B and C and D exist simultaneously".
[0020] It should be understood that in this application specification and the appended claims, the use of the term "or" indicates a combination in which one of the associated listed items exists alone. For example, "A, B, C or D" refers to the four combinations of "A alone", "B alone", "C alone", and "D alone".
[0021] It should be understood that, in this application specification and the appended claims, the term "and / or" indicates any combination of one or more of the associated listed items, as well as all possible combinations, and includes such combinations. For example, "A and / or B" refers to the three combinations of "A alone," "B alone," or "A and B simultaneously." For example, "A, B, and / or C" refers to the seven combinations of "A alone," "B alone," "C alone," "A and B simultaneously," "A and C simultaneously," "B and C simultaneously," and "A, B, and C simultaneously."
[0022] It should be understood that, in this specification and the appended claims, the term "if" is used to indicate, depending on the context, "in the case of," "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [the described condition or event] is detected" can be interpreted, depending on the context, as "once determined," "in response to determination," "once [the described condition or event] is detected," or "in response to detection of [the described condition or event]."
[0023] It should be understood that in this application specification and the appended claims, the terms "greater than", "less than", "exceeding", etc. are understood to exclude the number itself; and the terms "above", "below", "within", etc. are understood to include the number itself.
[0024] It should be understood that in this application specification and the appended claims, the terms "the," "the," "the," "the," "the," "the described," "the mentioned," etc., may be understood, depending on the context, to refer to the content mentioned above.
[0025] It should be understood that in this application specification and the appended claims, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance, nor are they used to describe a specific order or sequence.
[0026] It should be understood that in this application specification and appended claims, the designations such as "S100," "S200," and "S300" are used only for distinguishing descriptions and should not be construed as indicating or implying a specific order or sequence of execution of steps or processes. For example, "S100, acquire data A; S200, acquire data B; S300, acquire C based on A and B;" is merely an example providing a feasible execution order, not a necessary one. Those skilled in the art can determine the actual and feasible execution order based on the function and internal logic of each step. For example, "step S100 is executed before step S200, then step S300 is executed," "step S200 is executed before step S100, then step S300 is executed," or "steps S100 and S200 are executed in parallel, then step S300 is executed."
[0027] It should be understood that in this application specification and appended claims, if numbering or quantity in the form of “f(i); i=1, 2, ..., n;” is used, the specific value corresponding to the term “f(i)” should be understood as a value that changes with the value of i, and not as a fixed value. For example, in different practical scenarios, “f(1)” and “f(2)”, “f(1)” may be equal to “f(2)” or may not be equal to “f(2)”, and unless otherwise specified, there is no necessary size relationship between “f(1)” and “f(2)”.
[0028] It should be understood that in this application specification and appended claims, if numbers or symbols in the form of superscript are used, and the context or other corresponding locations provide a specific interpretation of them, they should be understood according to their corresponding explanatory description, and cannot be directly and simply understood as concepts such as "exponent" in mathematics or "atomic number" in chemistry. Similarly, if numbers or symbols in the form of subscript are used, they should be understood in the same way, without further explanation.
[0029] It should be understood that in the description of this application and the appended claims, if terms such as "A=(...)" are used, unless otherwise specifically interpreted, they should be understood as "A includes...", and not as "A is composed of...". Furthermore, unless otherwise specifically interpreted, the parentheses "(" and ")" are only used in conjunction with "=" to indicate what items or data A includes, and do not refer to any specific data structure, nor do they limit the type of its elements, whether the number of elements is fixed, whether the elements are ordered, whether the elements are repeatable, what query method is required to query its elements, or what access order is required to access its elements.
[0030] It should be understood that in the description of this application and the appended claims, if common processing functions such as "max(...)", "min(...)", and "avg(...)" are used, the context or other corresponding locations should provide a specific interpretation of these terms. If no specific interpretation exists, those skilled in the art should use relevant technical terms from computer science and technology for connection. For example, "max(...)" refers to a function for determining the maximum value, "min(...)" refers to a function for determining the minimum value, and "avg(...)" refers to a function for determining the average value.
[0031] Embodiments of this application provide a biological data simulation method based on a large language model, the method comprising the following steps, such as... Figure 1 As shown: S1. Perform intent recognition on the user's input natural language request and biometric data through a large language model to obtain structured intent recognition results.
[0032] Specifically, the intent recognition result includes several fields and the corresponding field value for each field.
[0033] Specifically, the fields in the intent recognition results include: user goal, data objects related to the user goal, constraints, expected output, and the capability type of the AI virtual cell model that may be needed.
[0034] Specifically, step S1 includes: After receiving the user's natural language request and biological data, the user's natural language request and biological data, together with the preset prompt words corresponding to the intent recognition task, are input into the large language model to drive the large language model to perform the intent recognition task and output a structured intent recognition result.
[0035] In another specific embodiment, the biological data includes uploaded file references, which are links or identifiers generated after a user uploads a file to the server and can be invoked by other programs.
[0036] Further, in step S1, a task envelope is constructed based on the user's input natural language request, uploaded file reference, session context, and capability selection information, and permissions and format pre-checks are performed. The task envelope includes: user request, session identifier, authorized file reference, current task type, cell state or perturbation condition, constraint condition, user-selected capability strategy, expected output type, and historical context summary. The task envelope is semantically parsed using a large language model to obtain structured intent recognition results.
[0037] S2. For each initial capability interface of the AI virtual cell model, match the intent recognition result with the capability description data corresponding to the initial capability interface, and obtain the target capability interface that matches the intent recognition result.
[0038] Specifically, based on a pre-defined unified capability protocol, the AI virtual cell model is standardized, encapsulated, and abstracted, and the underlying reasoning logic of the AI virtual cell model is mapped and exposed as multiple standardized initial capability interfaces.
[0039] Specifically, each initial capability interface is configured with corresponding capability description data; the capability description data includes several preset capability description fields and the field value corresponding to each capability description field.
[0040] Specifically, the capability description fields include: capability identifier, supported task types, input contract, output contract, availability status, and capability type.
[0041] In another specific embodiment, the capability description field also includes: resource requirements, downstream consumable declaration, product semantic type, version information, and health check method.
[0042] Specifically, each initial capability interface corresponds to several preset key feature tags; among which, as those skilled in the art know, the preset key feature tags are feature tags that are pre-set by those skilled in the art according to actual needs, used to characterize the core features that the initial capability interface depends on during runtime, such as: input parameter values, cell state, perturbation conditions, etc., which will not be elaborated here.
[0043] In another specific embodiment, the method further includes a protocolized AI virtual cell model capability mechanism. This protocolized AI virtual cell model capability mechanism is applied to the protocolized AI virtual cell model capability layer, exposing the relevant capabilities of the AI virtual cell model to the outside world through a unified capability protocol, thereby generating standardized initial capability interfaces. That is, the protocolized AI virtual cell model capability layer abstracts the AI virtual cell model inference capabilities that were originally scattered in single-point tools or independent services to a platform-level capability layer, enabling it to be stably discovered, selected, invoked, and handed over. The protocolized AI virtual cell model capability layer can access one or more inference implementations, adapt to different deployment environments, or be compatible with future added capabilities, but always maintains a unified protocol boundary. The protocolized AI virtual cell model capability layer specifically includes a capability registration interface, a capability description interface, an availability detection interface, a standardized invocation interface, and a result handover interface. The protocolized AI virtual cell model capability layer can be deployed as a local service, a remote service, a containerized service, a batch processing service, or a cloud service.
[0044] Furthermore, when new AI virtual cell model capabilities need to be integrated, the capability provider only needs to declare its task type, input contract, output contract, availability detection method, and output semantics according to the capability description protocol. Availability detection can be a synchronous check, asynchronous check, cache check, or periodic health check. In subsequent requests, the capability can be discovered through protocol matching, allowing the large language model to use the capability in the protocol field rather than the specific implementation name during planning. Since the upper layer does not depend on internal implementation names or internal service implementations, there is no need to modify the natural language task planning logic or add hard-coded branches for specific capabilities to the process.
[0045] In another specific embodiment, the protocol objects include: a task envelope, capability description data, a call context, an artifact reference, and a runtime closure; the call context includes: a user identifier, a task identifier, a session identifier, a deadline, a cancellation check function, a file parsing function, an event sending function, an artifact registration function, and a controlled input summary required for calling the AI virtual cell model; the artifact reference includes: a file identifier, a logical key, a semantic type, visibility, whether it is downloadable, its stage, artifact type, and metadata for downstream consumption; the runtime closure includes: a complete envelope formed at the end of an executable capability run, including a stage identifier, execution status, registered artifacts, hidden handover artifacts, user-visible artifacts, a structured response, and an error summary.
[0046] Specifically, in step S2, the intent recognition result is matched with the capability description data corresponding to the initial capability interface, including S21-S22: S21. Based on the field values in the intent recognition result and the capability description data corresponding to the initial capability interface, obtain the target matching score between the intent recognition result and the initial capability interface.
[0047] Specifically, the higher the target matching score, the higher the degree of matching between the intent recognition result and the initial capability interface.
[0048] Specifically, step S21 includes the following steps: S211. Obtain the semantic similarity between the user target in the intent recognition result and the supporting task type in the capability description data; wherein, the vector similarity between the semantic feature vector of the field value corresponding to the user target and the semantic feature vector of the field value corresponding to the supporting task type is taken as the semantic similarity; the value range of the vector similarity is 0 to 1, and the larger the vector similarity, the more similar the two corresponding semantic feature vectors are.
[0049] S212. When the data object and constraints in the intent recognition result satisfy the input contract in the capability description data, the input matching degree is determined to be 1; otherwise, the input matching degree is determined to be 0.
[0050] S213. When the expected output in the intent recognition result satisfies the output contract in the capability description data, the output matching degree is determined to be 1; otherwise, the output matching degree is determined to be 0.
[0051] S214. When the capability type of the AI virtual cell model that may be needed in the intent recognition result includes the capability type in the capability description data, the capability type matching degree is determined to be 1; otherwise, the capability type matching degree is determined to be 0.
[0052] S215. Based on the preset weights corresponding to semantic similarity, input matching degree, output matching degree, and ability type matching degree, perform weighted processing on semantic similarity, input matching degree, output matching degree, and ability type matching degree to obtain the target matching score; wherein, the preset weights corresponding to semantic similarity, input matching degree, output matching degree, and ability type matching degree are added together to equal 1.
[0053] In another specific embodiment, step S21 includes: Obtain the input contract satisfaction, output contract matching degree with expected result, state score of whether the result can be consumed by downstream steps, user or policy constraint satisfaction, and capability matching degree corresponding to the initial capability interface; and perform weighted processing on the input contract satisfaction, output contract matching degree with expected result, state score of whether the result can be consumed by downstream steps, user or policy constraint satisfaction, and capability matching degree corresponding to the initial capability interface according to preset weight parameters to obtain the target matching score, which is not greater than 1 and not less than 0.
[0054] S22. If the target matching score is not less than the preset threshold corresponding to the matching score, it is determined that the intent recognition result is successfully matched with the capability description data corresponding to the initial capability interface, and the initial capability interface is determined to be the target capability interface that matches the intent recognition result.
[0055] In step S22, if the target matching score is less than the preset threshold corresponding to the matching score, it is determined that the intent recognition result fails to match the capability description data corresponding to the initial capability interface.
[0056] Specifically, the preset threshold is 0.7.
[0057] Specifically, in step S2, before step S21, the following is also included: S20. When the initial capability interface is currently unavailable, it is determined that the intent recognition result fails to match the capability description data corresponding to the initial capability interface; when the initial capability interface is currently available, steps S21-S22 are executed.
[0058] Specifically, the field value corresponding to the available status in the capability description data is either identifier 0 or identifier 1.
[0059] Furthermore, when the value of the field corresponding to the available status in the capability description data corresponding to the initial capability interface is 0, it indicates that the initial capability interface is currently in an unavailable state; when the value of the field corresponding to the available status in the capability description data corresponding to the initial capability interface is 1, it indicates that the initial capability interface is currently in an available state.
[0060] Specifically, when the initial capability interface is invoked, the field value corresponding to the available status in the capability description data corresponding to the initial capability interface is updated to identifier 0; when the initial capability interface is invoked, the field value corresponding to the available status in the capability description data corresponding to the initial capability interface is updated to identifier 1.
[0061] In a specific embodiment, if the target matching score corresponding to all initial capability interfaces is less than the preset threshold corresponding to the matching score, the initial capability interfaces corresponding to the target matching scores are sorted in descending order of target matching scores, and the top F initial capability interfaces after sorting are selected as F target capability interfaces that match the intent recognition result; F is the preset number of target capability interfaces.
[0062] Through the above steps, based on the field values in the intent recognition result and the capability description data corresponding to the initial capability interface, semantic similarity, input matching degree, output matching degree, and capability type matching degree are determined. Based on the preset weights corresponding to these four factors, a weighted average is applied to obtain the target matching score. The initial capability interface corresponding to the target matching score that is not less than a preset threshold is determined as the target capability interface matching the intent recognition result. The target matching score is further determined by obtaining the target matching score through semantic vectors, rule constraints, and a multi-dimensional weighted scoring strategy, significantly improving the accuracy of target capability interface selection.
[0063] S3. Based on the key feature values corresponding to the target capability interface, determine at least one target arrangement experience text from the arrangement experience database, which includes several historical arrangement experience texts.
[0064] Specifically, step S3 includes S31-S34: S31. Extract the key feature values of each preset key feature label corresponding to each target capability interface from the user-input biometric data and intent recognition results.
[0065] S32. For each target capability interface, extract key feature values from all preset key feature labels corresponding to the target capability interface to obtain the initial data feature vector set A={A1, A2, ..., A...} for the target capability interface. i , ..., A m}, A i For the i-th target capability interface J i The corresponding initial data feature vector, 1≤i≤m, where m is the number of target capability interfaces.
[0066] Specifically, in step S32, for each key feature value, if the data type of the key feature value is numerical, then normalization or standardization processing is performed on the key feature value to obtain a numerical feature representation; if the data type of the key feature value is text, then vectorization encoding processing is performed on the key feature value to obtain a text feature vector; if the key feature value is null, then imputation or status marking is performed on the key feature value based on a preset missing value handling method to obtain a missing feature representation; according to the arrangement order of the preset key feature labels corresponding to the target capability interface, the numerical feature representations, text feature vectors or missing feature representations corresponding to the key feature values of all preset key feature labels corresponding to the target capability interface are concatenated to obtain the initial data feature vector corresponding to the target capability interface.
[0067] S33. Obtain the final data feature vector set B corresponding to the j-th historical arrangement experience text in the arrangement experience database. j The comprehensive vector similarity C between A and B j Where 1≤j≤n, and n is the number of historical arrangement experience texts in the arrangement experience database; each historical arrangement experience text corresponds to a final data feature vector set; the final data feature vector set includes the final data feature vector corresponding to each initial capability interface; the final data feature vector is obtained based on the key feature values of all preset key feature labels corresponding to the initial capability interface; C j Meets the following conditions: C j =∑ m i=1 (P) i ×Q ij ), P i For intent recognition results and J i The target matching score between; Q ij For A i With D ij Vector similarity between them; D ij For B j China and J i The corresponding final data feature vector.
[0068] Specifically, 0≤Q ij ≤1, Q ij The larger A is i With D ij The more similar they are.
[0069] In one specific embodiment, A i With D ij The vector distance between them is converted to a value between 0 and 1, and this value is used as Q. ijThe larger the vector distance, the closer the value between 0 and 1 converted from the vector distance is to 0; the smaller the vector distance, the closer the value between 0 and 1 converted from the vector distance is to 1.
[0070] S34, according to C j Define the objectives and compile experience texts.
[0071] Specifically, in step S34, according to C j Identify target layout experience texts, including: When C j >C 0 At that time, C j The corresponding historical layout experience text is used as the target layout experience text, C 0 This is the preset similarity threshold.
[0072] Specifically, C 0 The value range is from 0.7 to 0.9, preferably, C 0 It is 0.8.
[0073] In one specific embodiment, in step S34, according to C j Identify target layout experience texts, including: According to C1, C2, ..., C j , ..., C n Sort all historical layout experience texts in the layout experience database in descending order, and extract the top E historical layout experience texts after sorting as E target layout experience texts; E is the preset number of layout experience texts to be filtered.
[0074] Through the above steps, key feature values of each preset key feature label corresponding to each target capability interface are extracted from the user-input biometric data and intent recognition results; feature extraction is performed on the key feature values of all preset key feature labels corresponding to the target capability interface to obtain the initial data feature vector set corresponding to the target capability interface; based on the initial data feature vector set and the final data feature vector set corresponding to historical arrangement experience texts in the arrangement experience database, a comprehensive vector similarity is obtained; the target arrangement experience text is determined based on the comprehensive vector similarity, which can recall targeted target arrangement experience texts and avoid overgeneralization of the recall results.
[0075] Specifically, after step S2 and before step S3, the following steps are also included: If the number of target capability interfaces matching the intent recognition result is greater than 1, proceed to step S3; if the number of target capability interfaces matching the intent recognition result is equal to 1, call the target capability interface to obtain the interface response data returned by the target capability interface and proceed to step S6.
[0076] S4. Input the context information containing the target orchestration experience text into the large language model to obtain the interface orchestration results output by the large language model; the interface orchestration results include the execution order of each target capability interface.
[0077] Specifically, before step S4, context information is obtained, and the target arrangement experience text is concatenated into the context information to obtain the context information of the target arrangement experience text; wherein, when the length of the context information is greater than the preset length, the context information is compressed according to the length of the context information so as to update the context information.
[0078] Specifically, in step S4, while inputting the contextual information containing the target orchestration experience text into the large language model, the preset prompt words corresponding to the interface orchestration result generation task are also sent to the large language model to drive the large language model to execute the interface orchestration result generation task and output the interface orchestration result.
[0079] S5. Call each target capability interface sequentially according to the execution order of each target capability interface in the interface orchestration result to obtain the interface response data returned by each target capability interface.
[0080] In another specific embodiment, in step S5, after obtaining the interface response data returned by each target capability interface, a runtime closure is generated; the user-downloadable artifact, the hidden artifact for front-end display, and the downstream handover artifact are distinguished by visibility and semantic fields, respectively.
[0081] S6. Use a large language model to perform semantic parsing and natural language generation on the interface response data to obtain the natural language text corresponding to the biological data simulation results.
[0082] Specifically, in step S6, the interface response data returned by each target capability interface, the current context information, and the preset prompt words corresponding to the semantic parsing and natural language generation tasks are input into the large language model to drive the large language model to perform the semantic parsing and natural language generation tasks and output the natural language text corresponding to the biological data simulation results.
[0083] Through the above steps, the AI virtual cell model is abstracted into several initial capability interfaces. Target capability interfaces are identified by matching intent recognition results with capability description data. Furthermore, key feature values of the target capability interfaces are used to recall targeted orchestration experience text from the orchestration experience database. This allows the large language model to generate interface orchestration results based on contextual information containing the target orchestration experience text, avoiding overgeneralization of the recalled results and significantly improving the accuracy and reliability of the interface orchestration results generated by the large language model. Further, each target capability interface is sequentially called according to its execution order in the interface orchestration results to obtain the interface response data returned by each target capability interface. The large language model performs semantic parsing and natural language generation on the interface response data to obtain the natural language text corresponding to the biological data simulation results, significantly improving the accuracy of the natural language text corresponding to the biological data simulation results.
[0084] In another specific embodiment, in step S6, the large language model reads the structured result summary and product references in the running closure, generates a user-visible answer, and uses the user-visible answer as the natural language text corresponding to the biological data simulation result; wherein, for large-volume tables, graphs, binary files and complete structured data, they are only transmitted through product references or lateral response channels and do not directly enter the context of the large language model.
[0085] Furthermore, if the user continues to ask follow-up questions, the planning continues based on the task envelope summary, trusted product references, and executable context in the conversation, rather than inferring the AI virtual cell model input or backend execution parameters from the natural language display text.
[0086] Specifically, after step S6, steps S10-S40 are also included: S10. The large language model generates relevant data for the task based on the interface orchestration results, generates experience text related to interface orchestration, and stores this experience text as historical orchestration experience text L in the orchestration experience database.
[0087] Specifically, in step S10, the relevant data of the interface orchestration result generation task and the preset prompt words corresponding to the experience text generation task are input into the large language model to drive the large language model to execute the experience text generation task and generate experience text related to interface orchestration.
[0088] S20. Extract the key feature values of each preset key feature tag corresponding to each initial capability interface from the user-input biometric data, intent recognition results, interface response data returned by each target capability interface, and intermediate process data generated during the invocation of each target capability interface.
[0089] S30. For each initial capability interface, extract the key feature values of all preset key feature labels corresponding to the initial capability interface to obtain the final data feature vector corresponding to the initial capability interface.
[0090] Specifically, in step S30, the same method used in step S32 to extract key feature values of all preset key feature tags corresponding to the target capability interface to obtain the initial data feature vector corresponding to the target capability interface is adopted to extract key feature values of all preset key feature tags corresponding to the initial capability interface to obtain the final data feature vector corresponding to the initial capability interface.
[0091] S40. Store the final data feature vector set, which is composed of the final data feature vectors corresponding to all initial capability interfaces, as the final data feature vector set corresponding to L in the orchestration experience database.
[0092] Specifically, when the key feature value of the preset key feature label cannot be extracted, the key feature value of the preset key feature label is set to an empty value.
[0093] Through the above steps, after each round of biological data simulation task is completed, the large language model generates relevant data for the task based on the interface orchestration results, generates experience text related to interface orchestration, and stores this experience text as historical orchestration experience text in the orchestration experience database. From the biological data input by the user, the intent recognition results, the interface response data returned by each target capability interface, and the intermediate process data generated during the invocation of each target capability interface, the key feature values of each preset key feature tag corresponding to each initial capability interface are extracted, and feature extraction is performed on the key feature values of all preset key feature tags corresponding to the initial capability interface to obtain the final data feature vector corresponding to the initial capability interface, so as to construct the final data feature vector set corresponding to the historical orchestration experience text. This realizes the dynamic expansion and continuous iteration of the orchestration experience database, lays a solid foundation for subsequent targeted and accurate experience retrieval based on comprehensive vector similarity, and significantly improves the system's long-term adaptability in solving complex problems.
[0094] In another specific embodiment, the method further includes a runtime guarantee mechanism applied to the runtime guarantee layer. This mechanism injects runtime context such as user identifier, session identifier, task identifier, authorization file reference, cancellation check, deadline, event sending, and artifact registration into each large language model planning, AI virtual cell model call, or other capability call. It also uniformly maintains asynchronous task status, streaming events, rate limiting quotas, audit logs, resource isolation, and runtime closures. Event notifications in the runtime guarantee layer can be sent via server events, WebSocket, message queues, polling interfaces, or other event transmission methods.
[0095] Furthermore, for time-consuming analysis or inference tasks, the runtime assurance layer creates a task identifier and puts the task into a queued or running state. Users can subscribe to event streams through the task identifier and can also initiate cancellation requests when needed. The cancellation signal is written to the task state by the runtime assurance layer and checked collaboratively by the executed components through the call context. Regardless of whether the task is completed, failed, or cancelled, the system updates the final state, persists the phase snapshot, writes audit logs, and ensures that result messages in the user session do not remain in the running state for an extended period.
[0096] In this application embodiment, a non-transitory computer-readable storage medium is also provided. This non-transitory computer-readable storage medium can be disposed in an electronic device to store at least one instruction or at least one program related to implementing the method provided in any embodiment of this application. The at least one instruction or the at least one program is loaded and executed by a processor to implement the method provided in any embodiment of this application, and can achieve the same technical effect. To avoid repetition, further details are omitted here.
[0097] Examples of non-transitory computer-readable storage media include: computer read-only memory (ROM), random access memory (RAM), erasable programmable read-only memory (EPROM), compact disc read-only memory (CD-ROM), flash memory, magnetic disk, optical disk, portable computer disk, hard disk and / or solid-state drive, etc.
[0098] In this application embodiment, an electronic device is also provided, the electronic device including a processor and the non-transitory computer-readable storage medium. The processor loads and executes at least one instruction or at least one program stored in the non-transitory computer-readable storage medium related to implementing the method provided in any embodiment of this application, so as to implement the method provided in the embodiment of this application.
[0099] For example, the electronic device may be a mobile electronic device or a non-mobile electronic device that also includes other functions such as a personal digital assistant and / or music player. Further, the mobile electronic device may be any one of a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA); the non-mobile electronic device includes any one of a server, network attached storage (NAS), personal computer (PC), television (TV), ATM, or self-service machine. No specific limitations are made in the embodiments of this application.
[0100] For example, the processor can be a processor in any electronic device.
[0101] Exemplary embodiments of this application also provide a feasible structure for an electronic device. For example, the electronic device may include a processor, an external memory interface, internal memory, a universal serial bus (USB) interface (hereinafter referred to as a USB interface), a charging management module, a power management module, a battery, a first antenna, a second antenna, a mobile communication module, a wireless communication module, an audio module, a speaker, a receiver, a microphone, a headphone jack, a sensor module, buttons, a motor, an indicator, a camera, a display screen, and a subscriber identification module (SIM) card interface (hereinafter referred to as a SIM card interface), etc. The sensor module may include pressure sensors, gyroscope sensors, barometric pressure sensors, magnetic sensors, accelerometers, distance sensors, proximity sensors, fingerprint sensors, temperature sensors, touch sensors, ambient light sensors, and / or bone conduction sensors, etc.
[0102] It should be understood that the exemplary structures of the embodiments of this application do not constitute a specific limitation on the electronic device. In other embodiments of this application, the electronic device may include more or fewer components than the structure described herein, or combine some components, or split some components, or have different component arrangements. The components may be implemented in hardware, software, or a combination of software and hardware.
[0103] For example, the processor may include one or more processing units, wherein the processing units include: an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors. The controller may generate operation control signals based on instruction opcodes and timing signals to control instruction fetching and execution.
[0104] For example, the processor may further include a memory for storing instructions and data. In some embodiments of this application, the memory in the processor is a cache memory. The memory can store instructions or data that the processor has just used or that are used repeatedly. If the processor needs to use the instructions or data again, it can directly retrieve them from the memory.
[0105] For example, the processor may also include one or more interfaces. These interfaces include: an inter-integrated circuit (I2C) interface, an inter-integrated circuit sound (I2S) interface, a pulse code modulation (PCM) interface, a universal asynchronous receiver / transmitter (UART) interface, a mobile industry processor interface (MIPI), a general-purpose input / output (GPIO) interface, a SIM card interface, and / or a USB interface, etc.
[0106] The integrated circuit interface is a bidirectional synchronous serial bus, which includes a serial data line (SDA) and a serial clock line (SCL).
[0107] Integrated circuits with built-in audio interfaces can be used for audio communication. In some embodiments of this application, the processor may include multiple sets of integrated circuit-based audio interfaces. The processor can couple with an audio module through the integrated circuit-based audio interface to achieve communication between the processor and the audio module.
[0108] The pulse code modulation interface can also be used for audio communication, specifically for sampling, quantizing, and encoding analog signals. In some embodiments of this application, the audio module and the wireless communication module can be coupled through the pulse code modulation interface.
[0109] A Universal Asynchronous Receiver / Transmitter (UART) is a universal serial data bus interface used for asynchronous communication. This bus can be a bidirectional communication bus interface. It converts the data to be transmitted between serial and parallel communication. In some embodiments of this application, the processor and the wireless communication module can be connected via the UART.
[0110] Mobile industry processor interfaces can be used to connect processors to peripheral devices such as displays and cameras.
[0111] General purpose input / output interfaces can be configured via software.
[0112] Furthermore, the general-purpose input / output interface can be configured as a control signal or a data signal. In some embodiments of this application, the general-purpose input / output interface can be used to connect the processor to a camera, display screen, wireless communication module, audio module, sensor module, etc.
[0113] A USB interface is an interface that conforms to the USB standard specification, specifically including Mini USB, Micro USB, and USB Type-C interfaces. USB interfaces can be used to connect chargers to charge electronic devices, and also for transferring data between electronic devices and peripheral devices.
[0114] It should be understood that the interface connection relationships between the modules illustrated in the embodiments of this application are merely illustrative and do not constitute a limitation on the structure of the electronic device. In other embodiments of this application, the electronic device may also employ different interface connection methods or a combination of multiple interface connection methods as described in the embodiments.
[0115] The charging management module receives charging input from a charger, which can be either a wireless or wired charger. In some wired charging embodiments, the charging management module receives charging input from the wired charger via a USB interface. In some wireless charging embodiments, the charging management module receives wireless charging input via the wireless charging coil of the electronic device. While charging the battery, the charging management module can also supply power to various parts of the electronic device via the power management module.
[0116] The power management module is used to connect the battery, the charging management module, and the processor.
[0117] Wireless communication functionality in electronic devices can be achieved through a first antenna, a second antenna, a mobile communication module, a wireless communication module, a modem processor, and a baseband processor.
[0118] Mobile communication modules can enable wireless communication solutions, including 2G / 3G / 4G / 5G, for use in electronic devices.
[0119] A modem processor may include a modulator and a demodulator. The modulator modulates a low-frequency baseband signal to be transmitted into a mid-to-high frequency signal. The demodulator demodulates a received electromagnetic wave signal into a low-frequency baseband signal. The demodulator then transmits the demodulated low-frequency baseband signal to a baseband processor for processing. After processing by the baseband processor, the low-frequency baseband signal is transmitted to an application processor. The application processor outputs sound signals through audio devices (not limited to speakers, receivers, etc.) or displays images or videos on a display screen. In some embodiments of this application, the modem processor may be a separate device. In other embodiments, the modem processor may be independent of the processor and housed within the same device as the mobile communication module or other functional modules.
[0120] Wireless communication modules can enable solutions for wireless communication applications in electronic devices, including wireless local area networks (WLAN) (such as wireless fidelity (Wi-Fi) networks), Bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication (NFC), and infrared (IR) technologies.
[0121] In some embodiments of this application, the first antenna of the electronic device is coupled to the mobile communication module, and the second antenna is coupled to the wireless communication module, enabling the electronic device to communicate with networks and other devices via wireless communication technology.
[0122] Electronic devices utilize GPUs, displays, and application processors to achieve their display functions. A GPU is a microprocessor for image processing, connecting the display and the application processor. GPUs perform mathematical and geometric calculations and are used for graphics rendering. A processor may include one or more GPUs, which execute program instructions to generate or modify display information.
[0123] The display screen is used to display images, videos, etc. The display screen includes a display panel. The display panel can be a liquid crystal display (LCD), or a display panel made of materials selected from organic light-emitting diodes (OLEDs), active-matrix organic light-emitting diodes (AMOLEDs), flexible light-emitting diodes (FLEDs), minimized, microLEDs, micro-OLEDs, or quantum dot light-emitting diodes (QLEDs). In some embodiments of this application, the electronic device may include one or more display screens. In some embodiments of this application, the display screen may also integrate touch functionality and may also be referred to as a touch screen.
[0124] Electronic devices can achieve shooting functions through ISPs, cameras, video codecs, GPUs, displays, and application processors.
[0125] External storage interfaces can be used to connect external memory cards, such as Micro SD cards, to expand the storage capacity of electronic devices.
[0126] Internal memory can be used to store executable program code for a computer, which includes instructions. The processor executes the instructions stored in internal memory to perform various functional applications and data processing of electronic devices.
[0127] Electronic devices can implement audio functions through audio modules, speakers, receivers, microphones, headphone jacks, and application processors. Examples include music playback and recording.
[0128] The audio module converts digital audio information into analog audio signals for output, and also converts analog audio input into digital audio signals. The speaker, also called a "horn," converts audio electrical signals into sound signals. The receiver, also called a "handset," converts audio electrical signals into sound signals. The microphone, also called a "microphone" or "voice transducer," converts sound signals into electrical signals. The headphone jack is used to connect wired headphones.
[0129] Pressure sensors are used to sense pressure signals and convert them into electrical signals. In some embodiments of this application, the pressure sensor can be located on the display screen. A gyroscope sensor can be used to determine the motion posture of the electronic device. A barometric pressure sensor is used to measure air pressure. In some embodiments of this application, the electronic device calculates altitude using the air pressure value measured by the barometric pressure sensor, assisting in positioning and navigation. An accelerometer can detect the magnitude of acceleration of the electronic device in various directions (generally three axes). A distance sensor is used to measure distance. A fingerprint sensor is used to collect fingerprints. A touch sensor, also known as a "touch panel," can be located on the display screen, forming a touchscreen, also known as a "touch screen." A bone conduction sensor can acquire vibration signals. In some embodiments of this application, a bone conduction sensor can acquire vibration signals from vibrating bone fragments in the human vocal cords. A bone conduction sensor can also contact the human pulse to receive blood pressure signals.
[0130] The buttons include a power button and volume buttons. A motor can generate vibration alerts. Indicators can be indicator lights, used to show charging status, battery level changes, messages, missed calls, notifications, etc. A SIM card slot is used to connect a SIM card.
[0131] Embodiments of this application also provide a computer program product including program code that, when the program product is run on an electronic device, causes the electronic device to perform the steps of the methods described in this specification according to various exemplary embodiments of this application.
[0132] This application also provides a chip, which includes a processor and a communication interface. The communication interface is used to receive signals and transmit the signals to the processor. The processor processes the signals so that the methods described in the various exemplary embodiments of this application are executed.
[0133] While specific embodiments of this application have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this application. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of this application. The scope of this application is defined by the appended claims.
Claims
1. A biological data simulation method based on a large language model, characterized in that, The method includes the following steps: S1. Use a large language model to perform intent recognition on the user's input natural language request and biometric data to obtain structured intent recognition results; S2. For each initial capability interface of the AI virtual cell model, match the intent recognition result with the capability description data corresponding to the initial capability interface, and obtain the target capability interface that matches the intent recognition result. S3. Based on the key feature values corresponding to the target capability interface, determine at least one target arrangement experience text from the arrangement experience database, which includes several historical arrangement experience texts. S4. Input the context information containing the target orchestration experience text into the large language model to obtain the interface orchestration results output by the large language model; the interface orchestration results include the execution order of each target capability interface; S5. Call each target capability interface sequentially according to the execution order of each target capability interface in the interface orchestration result to obtain the interface response data returned by each target capability interface. S6. Use a large language model to perform semantic parsing and natural language generation on the interface response data to obtain the natural language text corresponding to the biological data simulation results.
2. The biological data simulation method based on a large language model according to claim 1, characterized in that, In step S2, the intent recognition result is matched with the capability description data corresponding to the initial capability interface, including: S21. Based on the field values in the intent recognition result and the capability description data corresponding to the initial capability interface, obtain the target matching score between the intent recognition result and the initial capability interface; the intent recognition result includes several fields and the field value corresponding to each field. S22. If the target matching score is not less than the preset threshold corresponding to the matching score, it is determined that the intent recognition result is successfully matched with the capability description data corresponding to the initial capability interface, and the initial capability interface is determined to be the target capability interface that matches the intent recognition result.
3. The biological data simulation method based on a large language model according to claim 1, characterized in that, Step S3 includes: S31. Extract the key feature values of each preset key feature label corresponding to each target capability interface from the user-input biometric data and intent recognition results. S32. For each target capability interface, extract key feature values from all preset key feature labels corresponding to the target capability interface to obtain the initial data feature vector set A={A1, A2, ..., A...} for the target capability interface. i , ..., A m }, A i For the i-th target capability interface J i The corresponding initial data feature vector, 1≤i≤m, where m is the number of target capability interfaces; S33. Obtain the final data feature vector set B corresponding to the j-th historical arrangement experience text in the arrangement experience database. j The comprehensive vector similarity C between A and B j Where 1≤j≤n, and n is the number of historical arrangement experience texts in the arrangement experience database; each historical arrangement experience text corresponds to a final data feature vector set; the final data feature vector set includes the final data feature vector corresponding to each initial capability interface; the final data feature vector is obtained based on the key feature values of all preset key feature labels corresponding to the initial capability interface; C j The following conditions must be met: C j =∑ m i=1 (P) i ×Q ij ), P i For intent recognition results and J i The target matching score between; Q ij For A i With D ij Vector similarity between them; D ij For B j China and J i The corresponding final data feature vector; S34, according to C j Define the objectives and compile experience texts.
4. The biological data simulation method based on a large language model according to claim 3, characterized in that, In step S34, according to C j Identify target layout experience texts, including: When C j >C 0 At that time, C j The corresponding historical layout experience text is used as the target layout experience text, C 0 This is the preset similarity threshold.
5. The biological data simulation method based on a large language model according to claim 3, characterized in that, In step S34, according to C j Identify target layout experience texts, including: According to C1, C2, ..., C j , ..., C n Sort all historical layout experience texts in the layout experience database in descending order, and extract the top E historical layout experience texts after sorting as E target layout experience texts; E is the preset number of layout experience texts to be filtered.
6. The biological data simulation method based on a large language model according to claim 1, characterized in that, Following step S6, the following is also included: S10. The large language model generates relevant data for the task based on the interface orchestration results, generates experience text related to interface orchestration, and stores the experience text as historical orchestration experience text L in the orchestration experience database. S20. Extract the key feature value of each preset key feature label corresponding to each initial capability interface from the user-input biometric data, intent recognition results, interface response data returned by each target capability interface, and intermediate process data generated during the invocation of each target capability interface. S30. For each initial capability interface, extract the key feature values of all preset key feature tags corresponding to the initial capability interface to obtain the final data feature vector corresponding to the initial capability interface. S40. Store the final data feature vector set, which is composed of the final data feature vectors corresponding to all initial capability interfaces, as the final data feature vector set corresponding to L in the orchestration experience database.
7. The biological data simulation method based on a large language model according to claim 2, characterized in that, In step S2, the following is included before step S21: S20. When the initial capability interface is currently unavailable, determine that the intent recognition result fails to match the capability description data corresponding to the initial capability interface; when the initial capability interface is currently available, execute steps S21-S22.
8. The biological data simulation method based on a large language model according to claim 3 or 6, characterized in that, If the key feature value of the preset key feature label cannot be extracted, the key feature value of the preset key feature label will be set to an empty value.
9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores at least one instruction or at least one program segment, which is loaded and executed by a processor to implement the biological data simulation method based on a large language model as described in any one of claims 1-8.
10. An electronic device, characterized in that, Includes a processor and the non-transitory computer-readable storage medium as described in claim 9.
