A decision method, system, terminal and medium of an organic semiconductor light-emitting material

Abstract

The application discloses a decision-making method, system, terminal and medium of an organic semiconductor light-emitting material, and the method comprises the following steps: integrating multi-source heterogeneous data in the field of organic semiconductor light-emitting materials, and constructing a high-quality heterogeneous knowledge base; adopting a semantic analysis model based on a material field adaptive BERT to convert a user natural language query into a structured query, and dynamically constructing a knowledge subgraph related to a task; constructing a symbolic reasoning engine based on a field rule, performing an interpretable logical reasoning, obtaining a symbolic reasoning conclusion, adopting a material field heterogeneous graph representation learning model to perform distributed representation learning and multi-task optimization, obtaining a learning reasoning conclusion, integrating the symbolic reasoning conclusion and the learning reasoning conclusion, and generating a final decision-making suggestion. Through the construction of a multi-source heterogeneous knowledge base, the design of an adaptive scheduling engine and the realization of multi-level reasoning fusion, the efficiency and accuracy of material research and development are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence-driven materials science and technology, and in particular to a decision-making method, system, terminal, and medium for organic semiconductor light-emitting materials. Background Technology

[0002] In the field of organic semiconductor light-emitting materials research and development, exploring the complex relationship between material structure and performance is crucial for accelerating the discovery of new materials and optimizing device design. However, traditional research methods often rely on single types of data or empirical rules, such as extracting limited information from molecular structure descriptors, experimental performance parameters, or patent literature. While this single-dimensional analysis is effective in certain specific scenarios, it struggles to comprehensively capture the multi-factor interactions in material design, such as the dynamic relationships between molecular configuration, fabrication process, device environment, and performance indicators. In recent years, with the widespread application of artificial intelligence in materials science, data-driven methods have gradually become a research hotspot. By constructing a knowledge base that integrates multi-source data and introducing machine learning models, it is possible to predict material properties and uncover design rules.

[0003] However, existing methods tend to focus on homogeneous data, while the heterogeneous data sources involved in the field of organic semiconductor luminescent materials, such as scientific literature, patent documents, experimental reports, and molecular structure databases, suffer from semantic gaps and format differences, making knowledge integration difficult and unable to support complex decision-making and reasoning needs. Currently, research in this field is gradually moving from single models to hybrid reasoning frameworks to combine the interpretability of symbolic reasoning with the generalization ability of neural networks. However, how to achieve efficient scheduling of heterogeneous knowledge and the fusion of logical reasoning and decision-making remains a significant challenge.

[0004] Therefore, existing technologies still have shortcomings. Summary of the Invention

[0005] To address the aforementioned deficiencies in existing technologies, this invention provides a decision-making method, system, terminal, and medium for organic semiconductor light-emitting materials. The technical solution adopted by this invention is as follows:

[0006] In a first aspect, the present invention provides a decision-making method for organic semiconductor light-emitting materials, the method comprising:

[0007] By integrating multi-source heterogeneous data in the field of organic semiconductor luminescent materials, and through structured knowledge extraction and format standardization, unstructured knowledge parsing and semantic extraction, as well as domain ontology-driven semantic alignment and quality verification, a high-quality heterogeneous knowledge base is constructed.

[0008] A semantic parsing model based on BERT adapted to the materials domain is adopted to transform user natural language queries into structured queries. An adaptive knowledge scheduling engine based on demand-aware graph neural network is designed to perform multi-hop retrieval in heterogeneous knowledge bases according to the structured queries, and to dynamically construct task-related knowledge subgraphs through a priority scheduling mechanism.

[0009] A domain-rule-based symbolic reasoning engine is constructed to perform interpretable logical reasoning and obtain symbolic reasoning conclusions. A heterogeneous graph representation learning model for materials domains is used for distributed representation learning and multi-task optimization to obtain learned reasoning conclusions. Finally, an attention-driven hybrid reasoning decision fusion mechanism is used to integrate the symbolic reasoning conclusions and the learned reasoning conclusions to generate final decision recommendations. The symbolic reasoning conclusions include a set of candidate materials that meet the requirements and corresponding supporting evidence. The learned reasoning conclusions are used to discover and recommend potential structure-performance correspondences and novel candidate materials that are not explicitly recorded in the heterogeneous knowledge base.

[0010] In one implementation, multi-source heterogeneous data from the field of organic semiconductor luminescent materials are integrated. Through structured knowledge extraction and format standardization, unstructured knowledge parsing and semantic extraction, and domain ontology-driven semantic alignment and quality verification, a high-quality heterogeneous knowledge base is constructed, including:

[0011] This paper integrates structured data sources in the field of organic semiconductor light-emitting materials, including material performance parameter databases, molecular structure databases, and device characteristic table data. The structured data sources are preprocessed and integrated into a unified structured knowledge set through data format conversion and field alignment.

[0012] Using natural language processing technology adapted to the field of organic semiconductors, unstructured data sources are extracted from scientific literature, patent documents and experimental reports to construct a BERT adapted to the material domain. Based on a semantic extraction algorithm that combines rules and statistics, semantic triples are extracted from the output of the material domain-adapted BERT to form an unstructured knowledge set. The semantic triples include two types: material-process-performance and structure-characteristics-application.

[0013] A domain ontology dictionary for organic semiconductor luminescent materials is constructed. Based on the domain ontology dictionary, a semantic alignment function is designed to map terms from different sources in structured and unstructured knowledge sets to standard concepts in the domain ontology dictionary, achieving semantic consistency across data sources. An expert verification team for the organic semiconductor luminescent materials domain is established, a knowledge quality assessment index system is developed, and the semantically aligned knowledge units are reviewed one by one. Knowledge confidence is calculated through expert scoring, and a high-quality heterogeneous knowledge base is constructed based on the knowledge confidence.

[0014] In one implementation, a domain-adaptive BERT-based semantic parsing model is used to transform user natural language queries into structured queries. An adaptive knowledge scheduling engine based on a demand-aware graph neural network is designed to perform multi-hop retrieval in a heterogeneous knowledge base based on the structured queries. Furthermore, a priority scheduling mechanism is used to dynamically construct task-related knowledge subgraphs, including:

[0015] Receive user natural language queries, perform deep processing on user natural language queries through a domain-adaptive BERT semantic parsing model, calculate the intent vector of user natural language queries, and generate structured queries based on the intent vector;

[0016] Based on the heterogeneous knowledge base, a knowledge graph of organic semiconductor luminescent materials is constructed. An adaptive knowledge scheduling engine based on a demand-aware graph neural network is designed to transform structured queries into dynamic attention weights, perform weighted processing on entities and relationships in the knowledge graph, generate multi-hop retrieval paths, and calculate the weights of knowledge fragments in the heterogeneous knowledge base.

[0017] Based on the priority scheduling mechanism, target knowledge fragments whose weights meet preset conditions are selected, and the knowledge subgraph is constructed. The knowledge subgraph includes: the selected entity set and the set of relationships between entities.

[0018] In one implementation, a domain-rule-based symbolic reasoning engine is constructed to perform interpretable logical reasoning and obtain symbolic reasoning conclusions. A heterogeneous graph representation learning model from the materials domain is then used for distributed representation learning and multi-task optimization to obtain learned reasoning conclusions, including:

[0019] An expert rule base in the field of organic semiconductor luminescent materials is constructed, and a symbolic reasoning engine based on the domain rules is built based on the expert rule base. Knowledge subgraphs and structured queries are input into the symbolic reasoning engine, and the matching score of each rule is calculated through a rule matching algorithm. Based on the matching score, target rules that meet the requirements are selected, logical reasoning is performed and an interpretable reasoning path is generated, and the symbolic reasoning conclusion is output.

[0020] Distributed representation learning is performed using a heterogeneous graph representation learning model in the materials domain. This model introduces a performance correlation enhancement module to strengthen the feature propagation of two key relationships: material-performance and structure-performance.

[0021] A multi-task learning framework is designed to simultaneously optimize the material performance prediction task and the structure-performance relationship mining task. Through multi-task learning, the heterogeneous graph representation learning model in the material domain simultaneously captures the performance patterns and potential correlations of materials and outputs learning inference conclusions.

[0022] In one implementation, an attention-driven hybrid reasoning decision fusion mechanism integrates symbolic reasoning conclusions and learned reasoning conclusions to generate a final decision recommendation, including:

[0023] The symbolic reasoning conclusion and the learned reasoning conclusion are respectively transformed into feature vectors of a unified dimension;

[0024] The fusion weights are dynamically allocated through the attention calculation module. Based on the attention-driven hybrid reasoning decision fusion mechanism, the fused decision vector is obtained, and preliminary decision suggestions are generated through linear mapping.

[0025] A conflict detection and resolution algorithm is developed to calculate the semantic similarity between the symbolic reasoning conclusion and the feature vector corresponding to the learned reasoning conclusion. When the semantic similarity is lower than the conflict threshold, the conflict resolution process is triggered, and the preliminary decision suggestions are corrected based on the domain ontology dictionary and expert rule base to output the final decision suggestions.

[0026] In one implementation, the method further includes:

[0027] Design a meta-learning adaptation framework for materials science to adapt to new material categories and decision-making scenarios from limited expert feedback;

[0028] Implement a knowledge complexity-aware learning strategy to train relevant models in the field of organic semiconductor light-emitting materials in ascending order of complexity.

[0029] In one implementation, the method further includes:

[0030] This web-based interactive decision support platform provides a natural language query interface, a visual decision dashboard, and real-time decision generation capabilities. It also continuously collects and uses data to optimize the performance of relevant models in the field of organic semiconductor luminescent materials.

[0031] Secondly, embodiments of the present invention also provide a decision-making system for organic semiconductor light-emitting materials, wherein the system is used to implement the steps of the decision-making method for organic semiconductor light-emitting materials described in any of the above solutions, and the system includes:

[0032] The heterogeneous knowledge base construction module is used to integrate multi-source heterogeneous data in the field of organic semiconductor luminescent materials. It constructs a high-quality heterogeneous knowledge base through structured knowledge extraction and format standardization, unstructured knowledge parsing and semantic extraction, and domain ontology-driven semantic alignment and quality verification.

[0033] The knowledge subgraph construction module is used to transform user natural language queries into structured queries using a semantic parsing model based on BERT adapted to the materials domain. It designs an adaptive knowledge scheduling engine based on a demand-aware graph neural network, performs multi-hop retrieval in a heterogeneous knowledge base according to the structured query, and dynamically constructs task-related knowledge subgraphs through a priority scheduling mechanism.

[0034] The decision suggestion generation module is used to construct a domain-rule-based symbolic reasoning engine, perform interpretable logical reasoning, obtain symbolic reasoning conclusions, use a material domain heterogeneous graph representation learning model for distributed representation learning and multi-task optimization to obtain learned reasoning conclusions, and integrate the symbolic reasoning conclusions and learned reasoning conclusions through an attention-driven hybrid reasoning decision fusion mechanism to generate final decision suggestions. The symbolic reasoning conclusions include a set of material candidate materials that meet the requirements and corresponding supporting evidence, while the learned reasoning conclusions are used to discover and recommend potential structure-performance correspondences and novel candidate materials that are not explicitly recorded in the heterogeneous knowledge base.

[0035] Thirdly, embodiments of the present invention also provide a terminal, wherein the terminal includes a memory, a processor, and a decision program for organic semiconductor light-emitting materials stored in the memory and executable on the processor. When the processor executes the decision program for organic semiconductor light-emitting materials, it implements the steps of the decision method for organic semiconductor light-emitting materials of any of the above-described schemes.

[0036] Fourthly, embodiments of the present invention also provide a computer-readable storage medium, wherein the computer-readable storage medium stores a decision program for an organic semiconductor light-emitting material, and the decision program for the organic semiconductor light-emitting material implements the steps of the decision method for the organic semiconductor light-emitting material described in any of the above-described schemes on the computer-readable storage medium.

[0037] Beneficial Effects: Compared with existing technologies, this invention provides a decision-making method for organic semiconductor luminescent materials. First, it integrates multi-source heterogeneous data in the field of organic semiconductor luminescent materials. Through structured knowledge extraction and format standardization, unstructured knowledge parsing and semantic extraction, and domain ontology-driven semantic alignment and quality verification, a high-quality heterogeneous knowledge base is constructed. Next, a semantic parsing model based on a materials domain-adapted BERT is used to transform user natural language queries into structured queries. An adaptive knowledge scheduling engine based on a demand-aware graph neural network is designed to perform multi-hop retrieval in the heterogeneous knowledge base according to the structured queries, and a priority scheduling mechanism dynamically constructs task-related knowledge subgraphs. Then, a domain-rule-based symbolic reasoning engine is constructed to perform interpretable logical reasoning and obtain symbolic reasoning conclusions. A heterogeneous graph representation learning model for materials domains is then used for distributed representation learning and multi-task optimization to obtain learned reasoning conclusions. Finally, an attention-driven hybrid reasoning decision fusion mechanism is used to integrate the symbolic reasoning conclusions and the learned reasoning conclusions to generate final decision recommendations. The symbolic reasoning conclusions include a set of candidate materials that meet the requirements and corresponding supporting evidence. The learned reasoning conclusions are used to discover and recommend potential structure-performance correspondences and novel candidate materials that are not explicitly recorded in the heterogeneous knowledge base. This invention, based on heterogeneous knowledge scheduling and logical reasoning, overcomes the shortcomings of existing technologies and improves the efficiency and accuracy of materials research and development by constructing a multi-source heterogeneous knowledge base, designing an adaptive scheduling engine, and realizing multi-level reasoning fusion. Attached Figure Description

[0038] Figure 1 This is a flowchart of a preferred embodiment of the decision-making method for organic semiconductor light-emitting materials according to an embodiment of the present invention.

[0039] Figure 2 This is a schematic diagram of the overall technical route of the decision-making method for organic semiconductor light-emitting materials according to an embodiment of the present invention.

[0040] Figure 3 This is a schematic diagram of the processing flow and data flow of the core model component in the decision-making method for organic semiconductor light-emitting materials according to an embodiment of the present invention.

[0041] Figure 4 A framework diagram of a decision system for organic semiconductor light-emitting materials provided in an embodiment of the present invention.

[0042] Figure 5 This is a schematic diagram of the architecture of a decision system for organic semiconductor light-emitting materials provided in an embodiment of the present invention.

[0043] Figure 6 A schematic diagram of a terminal provided in an embodiment of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0045] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content, operations, or steps, nor does it require execution in the described order. For example, some operations or steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0046] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0047] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. For example, the first control information and the second control information are only used to distinguish different control information and do not limit their order.

[0048] Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or the order of execution, and that the words "first" and "second" do not necessarily imply that they are different.

[0049] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0050] Existing technologies often exhibit significant shortcomings when processing multi-source heterogeneous data on organic semiconductor luminescent materials. First, they struggle to achieve semantic alignment and unified representation across data sources, leading to fragmented knowledge and low reliability. Second, in terms of knowledge scheduling, existing methods lack adaptive mechanisms, failing to dynamically retrieve and weight relevant knowledge fragments based on user queries, thus limiting the accuracy and timeliness of decision-making. Furthermore, at the reasoning level, existing technologies largely rely on single symbolic rules or neural network models, making it difficult to balance interpretability and generalization ability, and failing to effectively handle conflicting rules and potential patterns in material design. In summary, existing technologies still have significant deficiencies in data integration, dynamic knowledge scheduling, hybrid reasoning, and automated decision support.

[0051] With the development of Heterogeneous Graph Neural Networks (HGNNs) and logical reasoning techniques, new solutions have been provided for processing heterogeneous knowledge of organic semiconductor luminescent materials. HGNNs can learn distributed representations of multiple types of entities and relationships, effectively capturing complex patterns between material structure and performance, while symbolic reasoning ensures the transparency and reliability of the decision-making process. Based on this, this embodiment proposes a decision-making method for organic semiconductor luminescent materials based on heterogeneous knowledge scheduling and logical reasoning. In specific applications, this embodiment first integrates multi-source heterogeneous data in the field of organic semiconductor luminescent materials. Through structured knowledge extraction and format standardization, unstructured knowledge parsing and semantic extraction, and domain ontology-driven semantic alignment and quality verification, a high-quality heterogeneous knowledge base is constructed. Next, a semantic parsing model based on material domain-adapted BERT is adopted to transform user natural language queries into structured queries. An adaptive knowledge scheduling engine based on demand-aware graph neural networks is designed to perform multi-hop retrieval in the heterogeneous knowledge base according to the structured queries, and a priority scheduling mechanism is used to dynamically construct task-related knowledge subgraphs. Then, a domain-rule-based symbolic reasoning engine is constructed to perform interpretable logical reasoning and obtain symbolic reasoning conclusions. A heterogeneous graph representation learning model for materials domains is used for distributed representation learning and multi-task optimization to obtain learned reasoning conclusions. Finally, through an attention-driven hybrid reasoning decision fusion mechanism, the symbolic reasoning conclusions and the learned reasoning conclusions are integrated to generate final decision recommendations. The symbolic reasoning conclusions include a set of candidate materials that meet the requirements and corresponding supporting evidence. The learned reasoning conclusions are used to discover and recommend potential structure-performance correspondences and novel candidate materials that are not explicitly recorded in the heterogeneous knowledge base.

[0052] Specifically, the decision-making method for organic semiconductor light-emitting materials in this embodiment can be applied to terminals, including intelligent product terminals such as computers. Specifically, as... Figure 1 As shown in the figure, the decision-making method for organic semiconductor light-emitting materials in this embodiment includes the following steps:

[0053] Step S100: Integrate multi-source heterogeneous data in the field of organic semiconductor luminescent materials, and construct a high-quality heterogeneous knowledge base through structured knowledge extraction and format standardization, unstructured knowledge parsing and semantic extraction, and domain ontology-driven semantic alignment and quality verification.

[0054] Combination Figure 2 As shown in the overall technical roadmap diagram, the technical roadmap of this embodiment includes: construction and semantic alignment of multi-source heterogeneous knowledge base, knowledge scheduling and demand parsing based on deep learning and graph neural networks, construction of multi-level reasoning and knowledge discovery models, model optimization and iterative learning strategies, and deployment and application of decision support systems.

[0055] In practical applications, the multi-source heterogeneous data in this embodiment includes structured data sources and unstructured data sources, combined with... Figure 3 As shown, the construction and semantic alignment of a multi-source heterogeneous knowledge base includes: structured knowledge extraction and format standardization, unstructured knowledge parsing and semantic extraction, and domain ontology-driven semantic alignment and quality verification, ultimately constructing a high-quality heterogeneous knowledge base. Specifically, in the process of structured knowledge extraction and format standardization, this embodiment first integrates structured data sources in the field of organic semiconductor luminescent materials. These structured data sources include material performance parameter databases, molecular structure databases, and device characteristic table data, thereby constructing a unified knowledge storage framework. Next, the structured data source is preprocessed, specifically including removing invalid records lacking key information. Then, this embodiment integrates the scattered structured data sources into a unified structured knowledge set through data format conversion and field alignment operations. In this context, each knowledge unit in the structured knowledge set is represented as... , A unique identifier for organic semiconductor light-emitting materials. This represents the standardized set of attribute values. In practice, data format conversion involves transforming structured data sources with different storage formats into a unified storage format or directly writing them into a unified knowledge storage framework. This ensures a consistent storage format. Field alignment, based on data format conversion, addresses the heterogeneity issue of fields with different names but the same meaning or the same name but different meanings. This embodiment establishes domain-unified data field mapping rules, and then maps heterogeneous fields in the structured data source to standard fields based on these rules, resulting in a set of standard fields. ,in This represents a standardized material attribute field, which ensures semantic consistency of the data.

[0056] In performing unstructured knowledge parsing and semantic extraction, this embodiment employs natural language processing technology adapted to the organic semiconductor field. Unstructured data sources are extracted from scientific literature, patent documents, and experimental reports to construct a materials-adapted BERT (Material-BERT). BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained language representation method based on the Transformer architecture. This embodiment's materials-adapted BERT innovatively embeds a domain-specific terminology enhancement module based on the multi-head attention mechanism and Transformer encoder of the original BERT architecture. Specifically, it constructs a professional vocabulary knowledge graph of organic semiconductor materials, extracting key terms such as conjugate structure naming rules and energy level regulation parameters. Then, a dynamic masking strategy is used to perform secondary pre-training on the vocabulary of the original BERT architecture, ultimately obtaining the materials-adapted BERT. When processing texts in fields such as molecular structure description and spectral characteristic analysis, the materials-adapted BERT constructed in this embodiment can accurately capture the semantic relationships between chemical groups and performance parameters, effectively improving the semantic understanding depth and representation accuracy of professional texts. Furthermore, this embodiment can extract semantic triples from the output results of BERT adapted for the materials domain based on a semantic extraction algorithm that combines rules and statistics. Then, all semantic triples are integrated to form an unstructured knowledge set. The semantic triples extracted in this embodiment include two types: material-process-performance and structure-characteristics-application. The semantic triples in this embodiment are represented as follows: ,in and These respectively represent entities in the field of organic semiconductor light-emitting materials. It indicates the relationship between entities.

[0057] Furthermore, in performing domain ontology-driven semantic alignment and quality verification, this embodiment constructs a domain ontology dictionary for organic semiconductor light-emitting materials. The domain ontology dictionary covers core concepts such as material categories, molecular structural units, performance parameters, preparation processes, and application scenarios, and clearly defines the hierarchical, synonymous, and associative relationships among these core concepts. Next, this embodiment designs a semantic alignment function based on the aforementioned domain ontology dictionary. The core operation object of semantic alignment functions is semantic triples. In and as well as The included terms (such as heterogeneous terms like "driving voltage" and "operating voltage" expressed in different texts) are compared with the domain ontology dictionary. Standard concept matching, thereby eliminating semantic triples. Internal terminological ambiguity, therefore, is addressed by using this semantic alignment function to structure the knowledge set. Unstructured knowledge set Terms from different sources are mapped to standard concepts in the domain ontology dictionary, thereby semantically aligning knowledge units in structured and unstructured knowledge sets to achieve semantic consistency across data sources. Then, this embodiment establishes an expert verification team in the field of organic semiconductor luminescent materials, establishes a knowledge quality assessment index system, reviews each semantically aligned knowledge unit, and calculates knowledge confidence through expert scoring. The calculation method for knowledge confidence is as follows: ,in, This indicates the number of experts involved in the verification. Indicates the first The quality scores, out of 10, were given by several experts. Finally, a high-quality heterogeneous knowledge base was constructed based on the knowledge confidence score. Specifically, after calculating the knowledge confidence score, this embodiment extracts knowledge units from the structured and unstructured knowledge sets whose confidence scores are below a preset threshold. Then, the remaining knowledge units from the structured and unstructured knowledge sets are integrated to construct the heterogeneous knowledge base. .

[0058] As can be seen, this embodiment achieves unified semantic representation of structured and unstructured data by constructing a domain ontology dictionary and semantic alignment function. Combined with an expert-verified knowledge confidence evaluation system, it overcomes the problems of terminological ambiguity and inconsistent data quality in traditional methods and establishes a high-quality heterogeneous knowledge base for organic semiconductor light-emitting materials.

[0059] Step S200: Using a semantic parsing model based on BERT adapted to the materials domain, the user's natural language query is transformed into a structured query. An adaptive knowledge scheduling engine based on a demand-aware graph neural network is designed to perform multi-hop retrieval in a heterogeneous knowledge base according to the structured query, and to dynamically construct a knowledge subgraph related to the task through a priority scheduling mechanism.

[0060] Combination Figure 2 As can be seen from this, the knowledge scheduling and demand parsing based on deep learning and graph neural networks in this embodiment specifically includes two aspects: semantic parsing of decision demands based on domain-adaptive BERT and an adaptive knowledge scheduling engine based on demand-aware GAN (graph neural network). Combined with... Figure 3 As shown in the figure, in the process of knowledge scheduling and requirement parsing, this embodiment first performs semantic parsing of decision requirements, then designs an adaptive knowledge scheduling engine, then constructs a knowledge graph, and finally constructs a knowledge subgraph related to the task.

[0061] In practical applications, the first step is to receive natural language queries from users. This natural language query Having expressed the decision-making needs, this embodiment then performs deep processing on the user's natural language query using a domain-adaptive BERT semantic parsing model to calculate the user's natural language query intent vector. Based on this intent vector, a structured query is generated. Specifically, this embodiment introduces a demand scenario encoding module into the input layer of the domain-adaptive BERT semantic parsing model. This module transforms typical application scenarios of organic semiconductor light-emitting materials into scenario feature vectors and fuses them with the text vector. Therefore, this embodiment can calculate the user's natural language query intent vector using this domain-adaptive BERT semantic parsing model. The intent vector is represented as: ,in This represents the learnable parameters of the Material-BERT model. This represents the feature vector of the application scenario. Then, based on the intent vector... Key constraints in decision-making requirements are identified through multilayer perceptrons. Performance goals and application scenarios And according to the preset structured template, the natural language query expressing the decision-making needs is processed. Transform into a structured query, which is represented as: This provides clear target guidance for subsequent knowledge scheduling.

[0062] Furthermore, this embodiment is based on the aforementioned heterogeneous knowledge base. Constructing a knowledge graph of organic semiconductor luminescent materials ,in, Represents a set of knowledge entities. This represents the set of relationships between entities. Next, to overcome the problem of traditional GNNs struggling to focus on key information in complex retrieval scenarios, this embodiment embeds a demand attention module into the traditional GNN's message passing and aggregation mechanism, designing an adaptive knowledge scheduling engine based on a demand-aware graph neural network. This demand attention module can construct an interaction matrix between demand feature vectors and graph entity features, and integrate structured queries... This is transformed into dynamic attention weights, which are then used to weight entities and relationships in the knowledge graph. This allows the model to dynamically adjust the information propagation path according to needs during message passing, achieving efficient multi-hop retrieval and generating multi-hop retrieval paths, represented as follows: ,in This refers to the intermediate related entities from the starting point to the ending point. Indicates the number of jumps.

[0063] In the process of generating multi-hop retrieval paths, this embodiment can calculate the relevance between entities and requirements using cosine similarity. ,in The output of the demand-aware graph neural network is the first Entity embedding vector, intent vector By using structured queries Features are extracted from the input Transformer encoder. Simultaneously, a multi-head attention mechanism is used to process requirements and entity relationships from different semantic perspectives in parallel, thereby enhancing the understanding of complex semantics. Next, this embodiment develops a priority scheduling mechanism, which can be used to prioritize the scheduling of knowledge fragments that meet the requirements, i.e., knowledge units extracted from the heterogeneous knowledge base, to construct a task-related knowledge subgraph. Specifically, the weights of the knowledge fragments in the heterogeneous knowledge base are first calculated, expressed as: ,in 、 、 The weighting coefficients are satisfied. , This represents the timeliness coefficient corresponding to the publication time of knowledge. To represent knowledge confidence, a priority scheduling mechanism is used to select target knowledge fragments whose weights meet preset conditions, and a knowledge subgraph is constructed. The knowledge subgraph includes: a set of selected entities and a set of relationships between entities. The knowledge subgraph is represented as follows: ,in For the filtered set of entities, It is a set of relationships between entities.

[0064] As can be seen, this embodiment designs a knowledge scheduling algorithm based on a demand-aware graph neural network. By dynamically adjusting the information propagation path through the demand attention module and combining a priority scheduling mechanism that weights multiple factors such as timeliness, confidence, and relevance, it achieves the ability to accurately extract task-related knowledge subgraphs from massive heterogeneous knowledge.

[0065] Step S300: Construct a domain-rule-based symbolic reasoning engine, perform interpretable logical reasoning to obtain symbolic reasoning conclusions, adopt a material domain heterogeneous graph representation learning model for distributed representation learning and multi-task optimization to obtain learning reasoning conclusions, and integrate symbolic reasoning conclusions and learning reasoning conclusions through an attention-driven hybrid reasoning decision fusion mechanism to generate final decision recommendations. The symbolic reasoning conclusions include a set of material candidate materials that meet the requirements and corresponding supporting evidence, while the learning reasoning conclusions are used to discover and recommend potential structure-performance correspondences and novel candidate materials that are not explicitly recorded in the heterogeneous knowledge base.

[0066] In the R&D decision-making of organic semiconductor materials, traditional methods face a contradiction: symbolic reasoning offers strong interpretability but weak generalization ability, while neural networks offer strong generalization ability but poor interpretability. This embodiment aims to integrate the determinism of symbolic reasoning with the generalization ability of neural representation learning to achieve decision support that combines reliable interpretation with innovative discovery. Figure 2 and Figure 3 As shown, the process of constructing the multi-level reasoning and knowledge discovery model in this embodiment includes three aspects: a domain-rule-based symbolic reasoning engine, a material domain heterogeneous graph representation model, and an attention-driven hybrid reasoning decision fusion mechanism, so as to finally generate reliable decision suggestions.

[0067] Specifically, this embodiment first constructs an expert rule base in the field of organic semiconductor light-emitting materials. Expert rule base It includes material design criteria, performance correlation rules, and failure mechanism judgment logic. All rules are formally represented using the Datalog language, a declarative logic programming language primarily used for database queries and other information retrieval tasks. The rule structure in this embodiment is as follows: ,in This indicates several rule premises. Indicates the number of prerequisites for the rule. It indicates the conclusion of a reasoning.

[0068] This embodiment can build a domain-rule-based symbolic reasoning engine based on the expert rule base, and integrate the knowledge subgraph. and structured queries The input is fed into the symbolic inference engine, where a matching score is calculated for each rule using a rule-matching algorithm. The matching score is calculated as follows:

[0069] .

[0070] The algorithm logic compares each condition in the rule premises one by one to see if a corresponding fact or highly fitting expression can be found in the current knowledge subgraph and structured query, accumulating a score for each successfully matched condition. This indicates the degree of matching between the preconditions and the knowledge subgraph and structured query. This embodiment can filter target rules that meet the requirements based on the matching score, perform logical reasoning, generate interpretable reasoning paths, and output symbolic reasoning conclusions. For example, it can filter rules with matching scores higher than a set threshold. The rules are used to obtain the target rules, and then logical reasoning is performed to generate interpretable reasoning paths. This leads to a deterministic inference conclusion. This conclusion includes a set of candidate materials that meet the requirements and corresponding supporting evidence.

[0071] Furthermore, this embodiment employs a materials domain heterogeneous graph representation learning model (using... This embodiment of the heterogeneous graph representation learning model in the materials domain introduces a performance correlation enhancement module based on the message passing mechanism of traditional heterogeneous graph neural networks. This module strengthens the feature propagation of two key relationships—material-performance and structure-performance—through an attention mechanism. The attention mechanism designed for this performance correlation enhancement module is represented as follows:

[0072] .

[0073] in, Indicates the first In a layered network, entities Its neighboring entities (through association) Attention weights for connections. Representing entities In the Layer embedding vectors, Representing neighboring entities In the Layer embedding vectors, and For all, a learnable weight matrix, For relation-specific weight matrices, For entity-specific weight matrices, For attention weight vectors, This is the transpose of the attention weight vector. This represents a vector concatenation operation. Representing entities Neighboring entities gather, Representing entities with neighboring entities The set of relationships between them Refers to neighboring entities Any neighboring entity in the set, It refers to any one of the associations in the set of associations. For any association A dedicated weight matrix, Represents any neighboring entity In the Layer embedding vectors, This is the activation function. Based on this performance correlation enhancement module, it can be an entity. Each neighbor entity and connection relationship Calculate an attention weight vector The attention weight vector depending on The current representation, The current representation and connection relationships The joint transformation. By normalizing the weights of all neighbor entities and connections, it can adaptively focus on those neighbor entities most relevant to performance prediction, thereby strengthening key semantic associations during message aggregation. This design enables heterogeneous graph representation learning models in the materials domain to adaptively focus on relationships closely related to performance, improving the targeting of representation learning. For knowledge subgraphs Each entity in Its embedding vector The calculation formula is:

[0074] .

[0075] in, This represents the model parameters of OSL-HGNN.

[0076] This embodiment designs a multi-task learning framework that simultaneously optimizes the material property prediction task and the structure-property relationship mining task. The total loss function is:

[0077] .

[0078] in and This represents the task weighting coefficient. This represents the mean square error loss in the prediction of material properties. This represents the cross-entropy loss used in structure-performance relationship mining. Through multi-task learning in this embodiment, the heterogeneous graph representation learning model in the materials domain can simultaneously capture the performance characteristics and potential correlations of materials, outputting learning inference conclusions. Learning to reason and draw conclusions Based on the distributed representation learned by the model, it is possible to discover and recommend potential structure-performance correspondences and novel material candidates that are not explicitly recorded in the knowledge base but are implicit in the data distribution.

[0079] Furthermore, this embodiment integrates the determinism of symbolic reasoning and the heterogeneous graph representation learning model in the materials domain ( To enhance the generalization ability of reasoning, an attention-driven hybrid reasoning fusion mechanism is designed. Specifically, firstly, the symbolic reasoning conclusions are... and the learning reasoning conclusions Transform them into feature vectors of a uniform dimension. and ,in Feature encoding from symbolic reasoning path From The entities are embedded and stitched together. Then, the fusion weights are dynamically allocated through the attention calculation module. The calculation formula is:

[0080] .

[0081] in This represents the Sigmoid activation function. and These represent the weight matrix and bias vector of the attention module, respectively. This embodiment uses an attention-driven hybrid reasoning decision fusion mechanism to obtain a fused decision vector and generate preliminary decision suggestions through linear mapping. Specifically, this embodiment concatenates the features of the two reasoning conclusions and then uses a linear transformation plus nonlinear activation to generate an attention weight between 0 and 1. . The magnitude of the dynamics reflects the system's dependence on the deterministic conclusions of symbolic reasoning in the current decision-making scenario. The fused decision vector is represented as... This embodiment can generate preliminary decision suggestions through linear mapping. In practical applications, after obtaining the fused feature vector... Subsequently, this embodiment can fuse high-dimensional, abstract feature vectors. The results are transformed into low-dimensional, interpretable structured decision indicators through linear mapping, such as the fit score of material candidates and the probability of performance compliance, thereby obtaining preliminary decision suggestions. .

[0082] Furthermore, this embodiment develops a conflict detection and resolution algorithm, which calculates the semantic similarity between the symbolic reasoning conclusion and the feature vector corresponding to the learned reasoning conclusion. The semantic similarity is calculated as follows: .when and When the semantic similarity is below the conflict threshold, a conflict resolution process is triggered, based on the domain ontology dictionary. and expert rule base Regarding the aforementioned preliminary decision recommendations Make corrections and output final decision recommendations. Final decision recommendations in this embodiment This is a structured, implementable, and highly reliable R&D decision-making guide for organic semiconductor light-emitting materials, output to users after a full process of "requirements analysis - knowledge scheduling - multi-level reasoning - conflict resolution." It is the core output of this embodiment and can directly serve users' actual R&D needs, such as material selection, process optimization, and new material design. In practical applications, the process of revising the preliminary decision suggestions based on the domain ontology dictionary and expert rule base follows a closed-loop logic of "conflict identification → unified semantics → rule matching verification → consistency verification." Conflict identification is based on... and The semantic similarity function accurately identifies conflicting or unreasonable points in the initial decision-making suggestions. For example, if the semantic similarity is below a conflict threshold, a conflict is identified. Specific conflict types include entity ambiguity conflicts, rule conflicts, and conceptual contradictions. Unified semantics is based on a domain ontology dictionary. To address issues such as semantic inconsistency, conceptual ambiguity, and abnormal attribute range in preliminary decision-making recommendations, for example, by using a domain ontology dictionary. The rules define "hierarchical relationships" and "associations" to verify the rationality of relationships between entities in the preliminary decision-making recommendations. Rule matching verification is based on an expert rule base. To address issues such as "rule conflicts, performance prediction deviations, and unreasonable process adaptation" in preliminary decision-making recommendations, for example, by integrating the core conclusions of the preliminary decision-making recommendations with the expert rule base. The rules in the dictionary are matched one by one to verify whether the conclusion conforms to the domain criteria. Consistency verification is a secondary verification of the final decision recommendation obtained after being corrected by the domain ontology dictionary and expert rule base (including semantic consistency verification, performance parameter consistency verification, etc.) to ensure that the final decision recommendation has no new conflicts or logical loopholes.

[0083] In summary, this embodiment innovatively integrates a Datalog-based symbolic reasoning engine and a performance-enhanced heterogeneous graph neural network. Through an attention-driven decision fusion mechanism and a conflict resolution algorithm, it combines the interpretability of the reasoning process with the generalization ability to discover potential patterns.

[0084] In other implementations, this embodiment further provides model optimization and iterative learning strategies. Specifically, it designs a meta-learning adaptation framework for the materials domain to adapt to new material categories and decision-making scenarios from limited expert feedback, and implements a knowledge complexity-aware curriculum learning strategy to train relevant models in the field of organic semiconductor luminescent materials in ascending order of complexity. Combined with... Figure 2 and Figure 3 As shown, the model optimization and iterative learning strategy in this embodiment includes two aspects: a meta-learning adaptation framework for the materials domain and a course learning strategy that is only complexity-aware, thereby achieving model parameter optimization.

[0085] Specifically, this embodiment designs an Organic Semiconductor Light-Emitting Material Meta-Learning Adaptation Framework (OSL-Meta) to enable the relevant models involved in the above embodiments (i.e., the core model cluster supporting the entire organic semiconductor light-emitting material decision-making process) to quickly adapt to new material categories and decision-making scenarios from limited expert feedback. This framework includes a task generation module, an adaptive learning module, and a parameter update module. The task generation module draws from a heterogeneous knowledge base. By sampling decision-making needs from different material categories and application scenarios, a meta-learning task set is constructed. Each task ,in, Indicates task query. This represents a subset of task-related knowledge. This indicates expert feedback. The adaptive learning module employs an improved architecture of Model-Agnostic Meta-Learning (MAML), introducing a material category adaptation layer on top of the original MAML's gradient-based optimization strategy. For a given material category The adaptation layer uses a learnable weight matrix. Perform a linear transformation on the input features and use an activation function. Enhancing nonlinear expressive power, i.e. ,in This serves as the input feature vector. This adaptation layer, combined with the model's base network structure, enables the relevant models to capture the differences in properties and decision-making needs across different material categories.

[0086] During parameter updates, the adaptive loss function of meta-learning is minimized based on the meta-gradient descent strategy. Initialize model parameters Optimization will be carried out. Among them, This refers to the total number of independent meta-learning tasks contained in the meta-learning task set. Indicates learning tasks in the new yuan The adaptive loss, achieved by averaging the losses from multiple tasks, guides the model to learn more generalizable initialization parameters. Specifically, in each meta-learning task... First, based on the current initialization parameters... Perform a small number of gradient updates to obtain task-specific parameters. :

[0087] ;

[0088] .

[0089] in For learning rate, Indicates the first In the individual learning task, after The task-specific parameters obtained after gradient update. Indicates the first In the individual learning task, the task-specific parameters obtained after one step of gradient update are... Indicates about parameters The gradient. Finally, by minimizing the meta-learning loss. right The model is updated so that it can achieve good performance with only a few gradient updates when facing new tasks. Through the OSL-Meta framework in this embodiment, the model can quickly adapt to the decision-making requirements of novel organic semiconductor light-emitting materials, improving its cross-scenario transfer capabilities.

[0090] Furthermore, this embodiment implements a knowledge complexity-based learning strategy, training relevant models in ascending order of knowledge complexity to improve learning efficiency and stability. This strategy mimics the human learning process from easy to difficult, aiming to solve the problems of training instability, slow convergence, or getting stuck in local optima that easily occur when models directly learn complex, sparse, high-order knowledge. First, a knowledge complexity evaluation function is defined. In the knowledge complexity evaluation function, This represents a knowledge unit in the knowledge base. This represents the parameters of the complexity evaluation model. This function comprehensively considers the conceptual abstraction level of knowledge units, the density of relational connections, and the scarcity of the domain to calculate the complexity value. Based on the complexity value, heterogeneous knowledge bases are then evaluated. The knowledge units in the curriculum are divided into basic courses. Advanced courses and advanced courses ,in It includes low-complexity knowledge such as basic material properties and simple preparation processes. It includes knowledge of structure-performance relationships of moderate complexity and typical application cases. It includes knowledge of novel material design with high complexity and performance optimization involving multiple coupled factors. During training, this embodiment employs a dynamic learning rate adjustment strategy, with the learning rate update formula being: In the learning rate update formula, This represents the initial learning rate. Indicates the attenuation coefficient. Indicates the current training round. This represents the total number of training epochs. A relatively large learning rate is used in the early stages of training to achieve rapid convergence; as training progresses into the mid-to-late stages, the learning rate is gradually decreased for fine-tuning, ensuring stable model convergence. By learning knowledge of varying complexity in stages, the model can gradually accumulate domain knowledge, avoiding training instability caused by leaps in knowledge difficulty.

[0091] As can be seen, the meta-learning adaptation framework of the material category adaptation layer designed in this embodiment enables the relevant model in this invention to quickly adapt to new material categories from limited expert feedback; and by implementing knowledge complexity-based course learning and training the model in order of increasing complexity, the cross-scene transfer capability and learning stability of the model can be significantly improved.

[0092] Furthermore, in other implementations, this embodiment also provides the deployment and application of a decision support system. Specifically, it can be a web-based interactive decision support platform that provides a natural language query interface, a visual decision dashboard, and real-time decision generation functions, and continuously collects and uses data to optimize the performance of relevant models in the field of organic semiconductor luminescent materials. Combined with... Figure 2 As shown, the deployment and application of the decision support system in this embodiment includes three aspects: the development of a web-based interactive decision support platform, the implementation of real-time decision support and multi-dimensional analysis functions, and the expert feedback loop and system iterative optimization.

[0093] Specifically, this embodiment develops a web-based interactive decision support platform, namely an organic semiconductor luminescent material decision support platform, which adopts a front-end and back-end separation architecture. The front-end constructs the user interface and visualizes the decision results; the back-end develops core services and enables data interaction between the front-end and back-end through API interfaces. The platform provides a natural language query interface, allowing users to input their decision needs in everyday language. The interface uses Material-BERT semantic parsing service built with Flask (Flask is a lightweight web application framework written in Python) to convert natural language requests into structured queries in real time. The platform has a built-in user permission management module that assigns different knowledge base access permissions and function usage permissions according to user roles, ensuring knowledge security and standardized usage.

[0094] The decision support platform for organic semiconductor light-emitting materials in this embodiment can generate design suggestions for organic semiconductor light-emitting materials in real time. After the user inputs the decision requirements, a knowledge subgraph is quickly constructed through the above step S200. After processing in step S300, targeted decision-making suggestions are output within 10 seconds, including a list of candidate materials, performance expectations, preparation process schemes, and application suitability analysis. The platform also supports multi-scheme comparison, displaying performance parameter comparisons, cost assessments, and technology maturity analyses of different candidate materials through a visual dashboard, helping users intuitively select the optimal solution. Furthermore, the platform can integrate a sensitivity analysis module to calculate feature importance based on the OSL-HGNN model, identify key factors affecting material performance, and simulate the impact trends of changes in key factors on material performance, thereby providing users with directions for parameter optimization.

[0095] The decision support platform for organic semiconductor light-emitting materials in this embodiment can also integrate an expert feedback loop. Users can evaluate the accuracy and completeness of decision suggestions during use, and submit modification opinions and supplementary knowledge. Furthermore, the feedback collection module can store expert feedback information in real time. The feedback data is regularly organized and analyzed, and effective feedback is categorized into knowledge supplementation, rule correction, and model optimization. For knowledge supplementation feedback, the heterogeneous knowledge base is automatically updated. This triggers the semantic alignment process in step S100; for rule correction feedback, it is submitted to domain experts for review, and the expert rule base is updated. For model optimization feedback, it serves as training data for the meta-learning adaptation framework and is used to update model parameters. The organic semiconductor luminescent material decision support platform of this embodiment can be systematically verified and applied in typical organic semiconductor luminescent material R&D scenarios such as flexible displays, solid-state lighting, and biosensing. Furthermore, it collects real-world application data to continuously optimize the knowledge base quality and model inference accuracy, forming a closed-loop iterative mechanism to continuously improve the system's decision support capabilities.

[0096] In summary, the present invention has at least the following technical effects:

[0097] (1) Efficient integration and semantic unification of multi-source heterogeneous knowledge: Through domain ontology and semantic alignment technology, seamless integration of structured and unstructured data is achieved, eliminating terminological ambiguity and data fragmentation, improving the integrity and quality of the knowledge base, and providing a reliable data foundation for decision-making. (2) Precise demand analysis and dynamic knowledge scheduling: Using domain-adaptive graph neural networks, it can accurately analyze user query intent and adaptively retrieve and weight relevant knowledge fragments, ensuring that the decision-making process is highly targeted, responsive, and takes into account both the timeliness and confidence of knowledge. (3) Hybrid reasoning ability with both interpretability and generalization: Combining the determinism of symbolic reasoning and the generalization ability of neural network reasoning, the reasoning results are fused through the attention mechanism, providing a transparent and reliable decision-making path, while discovering potential material correlation patterns, enhancing the interpretability and accuracy of decision-making. (4) Adaptive learning and rapid transfer ability: Through meta-learning and curriculum learning strategies, the model can quickly adapt to new material categories and complex scenarios from limited expert feedback, and gradually learn highly complex knowledge, improving cross-domain generalization performance and stability, and reducing dependence on a large amount of labeled data. (5) Practical decision support and closed-loop optimization: It provides real-time decision generation, multi-scheme comparison and sensitivity analysis functions, and continuously iterates and optimizes through expert feedback loop to form a closed-loop improvement mechanism to ensure the practicality, reliability and evolution capability of the system in actual material research and development scenarios.

[0098] Based on the above embodiments, the present invention also provides a decision system for organic semiconductor light-emitting materials, which is used to implement the steps in the above method embodiments, such as... Figure 4 As shown, the decision-making system of this embodiment mainly includes: a heterogeneous knowledge base construction module 10, a knowledge subgraph construction module 20, and a decision suggestion generation module 30. Specifically, the heterogeneous knowledge base construction module 10 of this embodiment is used to integrate multi-source heterogeneous data in the field of organic semiconductor luminescent materials. Through structured knowledge extraction and format standardization, unstructured knowledge parsing and semantic extraction, and domain ontology-driven semantic alignment and quality verification, a high-quality heterogeneous knowledge base is constructed. The knowledge subgraph construction module 20 is used to adopt a semantic parsing model based on BERT adapted to the materials domain to transform user natural language queries into structured queries. An adaptive knowledge scheduling engine based on a demand-aware graph neural network is designed to perform multi-hop retrieval in the heterogeneous knowledge base according to the structured query, and to dynamically construct task-related knowledge subgraphs through a priority scheduling mechanism. The decision suggestion generation module 30 is used to construct a domain-rule-based symbolic reasoning engine, perform interpretable logical reasoning, obtain symbolic reasoning conclusions, use a material domain heterogeneous graph representation learning model for distributed representation learning and multi-task optimization to obtain learned reasoning conclusions, and integrate the symbolic reasoning conclusions and learned reasoning conclusions through an attention-driven hybrid reasoning decision fusion mechanism to generate final decision suggestions. The symbolic reasoning conclusions include a set of material candidate materials that meet the requirements and corresponding supporting evidence, while the learned reasoning conclusions are used to discover and recommend potential structure-performance correspondences and novel candidate materials that are not explicitly recorded in the heterogeneous knowledge base.

[0099] In practical applications, the complete architecture of the decision-making system for the organic semiconductor light-emitting material in this embodiment is as follows: Figure 5 As shown, the architecture specifically includes: a data layer, a knowledge construction layer, a core processing layer, an application layer, and an optimization and iteration layer. The data layer collects structured and unstructured data sources (such as scientific literature and patents, experimental reports, etc.). The knowledge construction layer constructs a multi-source heterogeneous knowledge base, performs semantic alignment and quality verification, resulting in a heterogeneous knowledge base. The core processing layer performs semantic parsing and requirement identification on natural language requirements, constructs a knowledge graph using a knowledge scheduling engine, and generates decision suggestions using a multi-level reasoning model. The application layer provides real-time decision support, including visualization, multi-solution comparison, and sensitivity analysis. The optimization and iteration layer collects expert feedback and updates model parameters using a meta-learning framework and course learning strategies.

[0100] The principle of each module in the decision system of the organic semiconductor light-emitting material in this embodiment is the same as that of each step in the above method embodiment, and will not be elaborated further here.

[0101] Based on the above embodiments, the present invention also provides a terminal, the principle block diagram of which can be as follows: Figure 6 As shown. The terminal may include one or more processors 100 ( Figure 6 (Only one is shown in the diagram), memory 101, and computer program 102 stored in memory 101 and executable on one or more processors 100. For example, a decision-making program for organic semiconductor light-emitting materials. When one or more processors 100 execute computer program 102, they can implement the various steps in the decision-making method embodiment for organic semiconductor light-emitting materials. Alternatively, when one or more processors 100 execute computer program 102, they can implement the functions of various modules / units in the decision-making system embodiment for organic semiconductor light-emitting materials, without limitation herein.

[0102] In one embodiment, the processor 100 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0103] In one embodiment, memory 101 may be an internal storage unit of an electronic device, such as a hard drive or RAM. Memory 101 may also be an external storage device of the electronic device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital Card (SD), or Flash Card. Furthermore, memory 101 may include both internal and external storage units. Memory 101 is used to store computer programs and other programs and data required by the terminal. Memory 101 can also be used to temporarily store data that has been output or will be output.

[0104] Those skilled in the art will understand that Figure 6 The block diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the terminal to which the present invention is applied. A specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0105] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), direct memory bus RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A decision-making method for organic semiconductor light-emitting materials, characterized in that, The method includes: By integrating multi-source heterogeneous data in the field of organic semiconductor luminescent materials, and through structured knowledge extraction and format standardization, unstructured knowledge parsing and semantic extraction, as well as domain ontology-driven semantic alignment and quality verification, a high-quality heterogeneous knowledge base is constructed. A semantic parsing model based on BERT adapted to the materials domain is adopted to transform user natural language queries into structured queries. An adaptive knowledge scheduling engine based on demand-aware graph neural network is designed to perform multi-hop retrieval in heterogeneous knowledge bases according to the structured queries, and to dynamically construct task-related knowledge subgraphs through a priority scheduling mechanism. A domain-rule-based symbolic reasoning engine is constructed to perform interpretable logical reasoning and obtain symbolic reasoning conclusions. A heterogeneous graph representation learning model in the materials domain is used for distributed representation learning and multi-task optimization to obtain learned reasoning conclusions. Then, through an attention-driven hybrid reasoning decision fusion mechanism, the symbolic reasoning conclusions and the learned reasoning conclusions are integrated to generate final decision recommendations. The symbolic reasoning conclusions include a set of candidate materials that meet the requirements and corresponding supporting evidence. The learned reasoning conclusions are used to discover and recommend potential structure-performance correspondences and novel candidate materials that are not explicitly recorded in the heterogeneous knowledge base. A domain-rule-based symbolic reasoning engine is constructed to perform interpretable logical reasoning and obtain symbolic reasoning conclusions. A heterogeneous graph representation learning model from the materials domain is used for distributed representation learning and multi-task optimization to obtain learned reasoning conclusions, including: An expert rule base in the field of organic semiconductor luminescent materials is constructed, and a symbolic reasoning engine based on the domain rules is built based on the expert rule base. Knowledge subgraphs and structured queries are input into the symbolic reasoning engine, and the matching score of each rule is calculated through a rule matching algorithm. Based on the matching score, target rules that meet the requirements are selected, logical reasoning is performed and an interpretable reasoning path is generated, and the symbolic reasoning conclusion is output. Distributed representation learning is performed using a heterogeneous graph representation learning model in the materials domain. This model introduces a performance correlation enhancement module to strengthen the feature propagation of two key relationships: material-performance and structure-performance. A multi-task learning framework is designed to simultaneously optimize the material performance prediction task and the structure-performance relationship mining task. Through multi-task learning, the heterogeneous graph representation learning model in the material domain simultaneously captures the performance patterns and potential correlations of materials and outputs learning inference conclusions.

2. The decision-making method for organic semiconductor light-emitting materials according to claim 1, characterized in that, This study integrates multi-source heterogeneous data in the field of organic semiconductor luminescent materials. Through structured knowledge extraction and format standardization, unstructured knowledge parsing and semantic extraction, and domain ontology-driven semantic alignment and quality verification, a high-quality heterogeneous knowledge base is constructed, including: This paper integrates structured data sources in the field of organic semiconductor light-emitting materials, including material performance parameter databases, molecular structure databases, and device characteristic table data. The structured data sources are preprocessed and integrated into a unified structured knowledge set through data format conversion and field alignment. Using natural language processing technology adapted to the field of organic semiconductors, unstructured data sources are extracted from scientific literature, patent documents and experimental reports to construct a BERT adapted to the material domain. Based on a semantic extraction algorithm that combines rules and statistics, semantic triples are extracted from the output of the material domain-adapted BERT to form an unstructured knowledge set. The semantic triples include two types: material-process-performance and structure-characteristics-application. A domain ontology dictionary for organic semiconductor luminescent materials is constructed. Based on the domain ontology dictionary, a semantic alignment function is designed to map terms from different sources in structured and unstructured knowledge sets to standard concepts in the domain ontology dictionary, achieving semantic consistency across data sources. An expert verification team for the organic semiconductor luminescent materials domain is established, a knowledge quality assessment index system is developed, and the semantically aligned knowledge units are reviewed one by one. Knowledge confidence is calculated through expert scoring, and a high-quality heterogeneous knowledge base is constructed based on the knowledge confidence.

3. The decision-making method for organic semiconductor light-emitting materials according to claim 1, characterized in that, A semantic parsing model based on BERT adapted to the materials domain is adopted to transform user natural language queries into structured queries. An adaptive knowledge scheduling engine based on a demand-aware graph neural network is designed to perform multi-hop retrieval in a heterogeneous knowledge base according to the structured queries, and to dynamically construct task-related knowledge subgraphs through a priority scheduling mechanism, including: Receive user natural language queries, perform deep processing on user natural language queries through a domain-adaptive BERT semantic parsing model, calculate the intent vector of user natural language queries, and generate structured queries based on the intent vector; Based on the heterogeneous knowledge base, a knowledge graph of organic semiconductor luminescent materials is constructed. An adaptive knowledge scheduling engine based on a demand-aware graph neural network is designed to transform structured queries into dynamic attention weights, perform weighted processing on entities and relationships in the knowledge graph, generate multi-hop retrieval paths, and calculate the weights of knowledge fragments in the heterogeneous knowledge base. Based on the priority scheduling mechanism, target knowledge fragments whose weights meet preset conditions are selected, and the knowledge subgraph is constructed. The knowledge subgraph includes: the selected entity set and the set of relationships between entities.

4. The decision-making method for organic semiconductor light-emitting materials according to claim 1, characterized in that, By integrating symbolic reasoning conclusions and learned reasoning conclusions through an attention-driven hybrid reasoning decision fusion mechanism, a final decision recommendation is generated, including: The symbolic reasoning conclusion and the learned reasoning conclusion are respectively transformed into feature vectors of a unified dimension; The fusion weights are dynamically allocated through the attention calculation module. Based on the attention-driven hybrid reasoning decision fusion mechanism, the fused decision vector is obtained, and preliminary decision suggestions are generated through linear mapping. A conflict detection and resolution algorithm is developed to calculate the semantic similarity between the feature vectors corresponding to the symbolic reasoning conclusion and the learned reasoning conclusion. When the semantic similarity is lower than the conflict threshold, the conflict resolution process is triggered, and the preliminary decision suggestions are corrected based on the domain ontology dictionary and expert rule base to output the final decision suggestions.

5. The decision-making method for organic semiconductor light-emitting materials according to claim 1, characterized in that, The method further includes: Design a meta-learning adaptation framework for materials science to adapt to new material categories and decision-making scenarios from limited expert feedback; Implement a knowledge complexity-aware learning strategy to train relevant models in the field of organic semiconductor light-emitting materials in ascending order of complexity.

6. The decision-making method for organic semiconductor light-emitting materials according to claim 1, characterized in that, The method further includes: This web-based interactive decision support platform provides a natural language query interface, a visual decision dashboard, and real-time decision generation capabilities. It also continuously collects and uses data to optimize the performance of relevant models in the field of organic semiconductor luminescent materials.

7. A decision system for an organic semiconductor light-emitting material, characterized in that, The system is used to implement the steps of the decision-making method for the organic semiconductor light-emitting material according to any one of claims 1-6, the system comprising: The heterogeneous knowledge base construction module is used to integrate multi-source heterogeneous data in the field of organic semiconductor luminescent materials. It constructs a high-quality heterogeneous knowledge base through structured knowledge extraction and format standardization, unstructured knowledge parsing and semantic extraction, and domain ontology-driven semantic alignment and quality verification. The knowledge subgraph construction module is used to transform user natural language queries into structured queries using a semantic parsing model based on BERT adapted to the materials domain. It designs an adaptive knowledge scheduling engine based on a demand-aware graph neural network, performs multi-hop retrieval in a heterogeneous knowledge base according to the structured query, and dynamically constructs task-related knowledge subgraphs through a priority scheduling mechanism. The decision suggestion generation module is used to construct a domain-rule-based symbolic reasoning engine, perform interpretable logical reasoning, obtain symbolic reasoning conclusions, use a material domain heterogeneous graph representation learning model for distributed representation learning and multi-task optimization to obtain learned reasoning conclusions, and integrate the symbolic reasoning conclusions and learned reasoning conclusions through an attention-driven hybrid reasoning decision fusion mechanism to generate final decision suggestions. The symbolic reasoning conclusions include a set of material candidate materials that meet the requirements and corresponding supporting evidence, while the learned reasoning conclusions are used to discover and recommend potential structure-performance correspondences and novel candidate materials that are not explicitly recorded in the heterogeneous knowledge base.

8. A terminal, characterized in that, The terminal includes a memory, a processor, and a decision program for organic semiconductor light-emitting materials stored in the memory and executable on the processor. When the processor executes the decision program for organic semiconductor light-emitting materials, it implements the steps of the decision method for organic semiconductor light-emitting materials as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a decision program for organic semiconductor light-emitting materials, and the decision program for organic semiconductor light-emitting materials implements the steps of the decision method for organic semiconductor light-emitting materials as described in any one of claims 1-6 on the computer-readable storage medium.

Images