Cross-domain vocabulary logical reasoning and expansion method

Abstract

The invention relates to the technical field of natural language processing, and discloses a cross-domain vocabulary logical reasoning and extension method. According to the method, a multi-source cross-domain text data stream is received, cross-language alignment and semantic fusion are performed through a pre-trained semantic decoupling model, an adaptive semantic extension threshold model and a causal reasoning framework are constructed, logic interruption nodes are identified, a compensation strategy is configured, and collaborative parameters are iteratively optimized by using a distributed semantic collaborative learning framework. Multi-source data can be effectively integrated, semantic barriers can be broken, the vocabulary expansion range and mode can be accurately controlled, logic transmission coherence and accuracy are ensured, efficient collaboration of multi-language agents is achieved, algorithm efficiency and semantic accuracy are improved, the problem of semantic instability is monitored in real time, the problem of semantic instability is solved, and powerful support is provided for cross-domain natural language processing.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a cross-domain vocabulary logic reasoning and expansion method. Background Art

[0002] In today's era of digital information explosion, natural language processing technology has broad and in-depth application needs in many fields. With the acceleration of globalization and the increasing trend of cross-disciplinary integration, the accurate understanding, reasoning, and expansion of cross-domain vocabulary have become key issues that need to be addressed in the field of natural language processing.

[0003] Traditional natural language processing methods have many limitations when dealing with cross-domain vocabulary. On the one hand, most existing technologies find it difficult to effectively integrate multi-source cross-domain text data. These data come from a wide range of sources, including multilingual corpora, domain terminology libraries, etc. The data formats and semantic structures are complex and diverse, and traditional methods are often unable to process and analyze them comprehensively and accurately. For example, when processing multilingual texts, it is difficult to achieve efficient and accurate cross-language alignment and semantic fusion, and it is impossible to fully explore the semantic associations between words in different languages, resulting in deviations in cross-language information transmission and understanding.

[0004] On the other hand, traditional methods lack effective strategies for the logical reasoning and expansion of vocabulary across different domains. When constructing semantic models, they fail to adequately account for domain specificity, making it difficult to rationally expand vocabulary based on the semantic characteristics and logical rules of different domains. Furthermore, when dealing with semantic conflicts and uncertainties, traditional methods are less adaptable and prone to problems such as irrational vocabulary expansion and interrupted logical transmission. For example, in text processing in some professional fields, inaccurate semantic understanding of domain terminology can lead to incorrect semantic associations during vocabulary expansion, affecting the overall semantic expression and information transfer of the text.

[0005] Existing vocabulary expansion algorithms also have shortcomings in balancing computational efficiency and accuracy. While some algorithms can achieve a certain degree of vocabulary expansion, they consume excessive computing resources and suffer from low processing efficiency when processing large-scale multi-source data. Meanwhile, some algorithms that pursue high efficiency struggle to guarantee the accuracy of vocabulary expansion and semantic consistency. Furthermore, in multilingual environments, the lack of effective coordination mechanisms between different language agents makes it impossible to achieve optimal resource allocation and semantic collaborative optimization, making it difficult to meet the high-quality requirements for cross-domain vocabulary processing in practical applications.

[0006] Therefore, there is an urgent need for a new technical solution that can overcome the shortcomings of existing technologies and achieve efficient and accurate logical reasoning and expansion of cross-domain vocabulary in a multi-source data environment to meet the growing demand for natural language processing applications. Summary of the Invention

[0007] The purpose of the present invention is to provide a cross-domain vocabulary logic reasoning and expansion method to solve the problems raised in the above background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a cross-domain vocabulary logic reasoning and expansion method, the method comprising:

[0009] Receiving a multi-source cross-domain text data stream, wherein the data stream includes a multi-language corpus, a domain term library, and contextual metadata;

[0010] Based on the pre-trained semantic decoupling model, the text data is cross-language aligned and semantically fused to generate a cross-domain semantic network graph and a vocabulary expansion probability matrix;

[0011] Based on the semantic network graph, an adaptive semantic extension threshold model is constructed to generate a multi-granularity vocabulary inference rule set, wherein the rule set includes a cross-language mapping topology structure and a semantic path dynamic generation strategy;

[0012] Based on a predefined causal reasoning framework, simulate the logical transfer path of vocabulary between different domains and optimize the semantic coordination parameters in the inference rule set;

[0013] The collaborative parameters are iteratively optimized through a distributed semantic collaborative learning framework, and cross-domain vocabulary extension sequences are output to a multilingual knowledge graph platform.

[0014] Preferably, the steps of constructing the semantic decoupling model include:

[0015] Collect parallel text corpora from multiple fields and construct a multi-dimensional training set including semantic role annotation, syntactic dependency tree and cross-language alignment relations;

[0016] Performing latent semantic decomposition on the multidimensional training set through quantum semantic embedding to extract independent representations of the main semantics of the vocabulary and contextual modification factors;

[0017] Combined with the lexical logic transfer equation, the topological constraints of the dynamic association between semantic factors are constructed;

[0018] The topological constraints are embedded in a hypergraph attention network to generate the semantic decoupling model that supports incremental learning.

[0019] Preferably, the adaptive semantic extension threshold model includes:

[0020] Dynamically dividing the vocabulary semantic density level according to the node connectivity of the semantic network graph;

[0021] Calculate the vocabulary expansion risk score based on domain term coverage and semantic ambiguity index;

[0022] Performing nonlinear mapping of the risk score and the semantic density level using a hyperbolic function to generate a domain-specific expansion threshold;

[0023] The activation condition of the multi-level lexical inference rule is triggered according to the threshold.

[0024] Preferably, the steps of constructing the causal reasoning framework include:

[0025] Collect historical cross-domain vocabulary migration data and construct a semantic causal graph dataset;

[0026] The probability and temporal delay parameters of the vocabulary transfer path are extracted through the counterfactual reasoning algorithm;

[0027] Combining semantic complex network theory to construct a lexical transitive directed hypergraph and quantify the logical dependency strength between nodes;

[0028] The dependency strength and the real-time context disturbance factor are input into a spatiotemporal gating network to generate the causal reasoning framework.

[0029] Preferably, the method further comprises:

[0030] Identifying key logic transfer interruption nodes based on simulation results of the causal reasoning framework;

[0031] configuring a semantic compensation strategy for the node in the lexical inference rule set;

[0032] Based on the compensation strategy, a cross-language collaborative mapping solution is automatically generated, including root alignment instructions and semantic bridging paths.

[0033] Preferably, the calculation of the vocabulary expansion risk score includes:

[0034] Obtain real-time semantic conflict matrices and domain term coverage heat maps to build a vocabulary expansion risk assessment cube;

[0035] Calculate the low-rank association weights of multi-dimensional semantic features through tensor decomposition network;

[0036] Performing a hypertensor contraction operation on the association weight and the assessment cube to obtain a comprehensive risk score;

[0037] The calculation formula for the comprehensive risk score is:

[0038]

[0039] In the formula, Γ represents the comprehensive risk score value, α i represents the semantic conflict intensity of the i-th category, β i represents the adjustment coefficient of the i-th type of conflict, γ crepresents the domain-based semantic compatibility constant, ⊙ represents the Hadamard product, and n represents the total number of semantic conflict classifications.

[0040] Preferably, the embedding of the topological constraint condition includes:

[0041] Performing semantic stability analysis on the latent semantic decomposition results to screen logically transitive association patterns;

[0042] Generate vocabulary evolution trajectories that conform to semantic constraints through random field sampling;

[0043] The trajectory data is used to perform sparse training on the hyperedge weights of the hypergraph attention network to ensure that the model output conforms to the cross-domain logical transmission rules.

[0044] Preferably, the execution of the distributed semantic collaborative learning framework includes:

[0045] Define the reward function for multi-agent semantic collaboration, including semantic consistency goals and extended efficiency goals;

[0046] Train the policy network of each language agent through a staged adversarial training strategy;

[0047] In each round of training, the gradient distribution weights between agents are dynamically adjusted according to the degree of goal conflict;

[0048] outputting the vocabulary expansion sequence that satisfies the Pareto optimality condition;

[0049] Wherein, the reward function is:

[0050] R=λ·S c +(1-λ)·θ

[0051] In the formula, R represents the reward value, S c represents the semantic consistency score, θ represents the expansion efficiency score, and λ is the dynamic adjustment factor.

[0052] Preferably, the method further comprises:

[0053] After configuring the semantic compensation strategy, the semantic transfer entropy of the logic transfer node is monitored in real time;

[0054] If the transfer entropy exceeds a preset threshold, a simulated annealing optimization mechanism is triggered to replan the spatiotemporal collaborative path of the cross-language mapping.

[0055] Preferably, the quantification of the expansion efficiency target includes:

[0056] Establish a semantic utility decay model for multi-granularity lexical resources and define a resource redundancy penalty function;

[0057] According to the semantic network evolution phase diagram, the marginal utility value of resource scheduling in each stage is calculated;

[0058] Using the marginal utility value as a dynamic gain coefficient of the reward function;

[0059] The calculation formula of the attenuation model is:

[0060]

[0061] Where K j represents the initial utility value of the j-th category of vocabulary resources, η j represents the semantic time decay coefficient, δ j (t) represents the resource activation status indicator function at time t, and m represents the total number of resource types.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] In the processing of multi-source cross-domain text data, the present invention can accurately receive and process multi-source cross-domain text data streams containing multilingual corpora, domain terminology libraries and context-related metadata. By performing cross-language alignment and semantic fusion based on a pre-trained semantic decoupling model, the semantic barriers between different languages and domains are broken. The model can deeply match and integrate the vocabulary and semantics in texts of different languages, generate a cross-domain semantic network map and a vocabulary expansion probability matrix, and provide a solid data foundation and semantic association basis for subsequent vocabulary reasoning and expansion. This makes it possible to more accurately understand the semantic connotation of vocabulary when processing multilingual and multi-domain texts, avoid semantic misunderstandings caused by language differences and domain barriers, and greatly improve the accuracy and effectiveness of cross-language and cross-domain information transmission.

[0064] In terms of vocabulary logical reasoning and expansion strategies, the adaptive semantic expansion threshold model and causal reasoning framework constructed by the present invention play a key role. The adaptive semantic expansion threshold model dynamically divides the vocabulary semantic density level according to the semantic network graph, and generates a domain-specific expansion threshold by accurately calculating the vocabulary expansion risk score. It can effectively control the scope and method of vocabulary expansion and avoid semantic conflicts and errors caused by blind expansion. The causal reasoning framework collects historical cross-domain vocabulary migration data, constructs a semantic causal graph data set, uses counterfactual reasoning algorithms and semantic complex network theory, accurately identifies key logical transmission interruption nodes, and configures semantic compensation strategies to ensure that the logical transmission of vocabulary between different fields is coherent and accurate, thereby realizing the reasonable and effective expansion of vocabulary in cross-domain scenarios.

[0065] In terms of algorithm efficiency and collaborative optimization, the distributed semantic collaborative learning framework of the present invention has outstanding advantages. By defining a reward function that includes semantic consistency goals and expansion efficiency goals, and adopting a phased adversarial training strategy to train the policy network of each language agent, the gradient distribution weights between agents can be dynamically adjusted to achieve efficient collaborative parameter optimization. At the same time, by establishing a semantic utility decay model for multi-granularity vocabulary resources and defining a resource redundancy penalty function, the efficiency of vocabulary expansion and semantic accuracy are effectively balanced. This enables the present invention to ensure the speed of vocabulary expansion while ensuring semantic consistency and accuracy when processing large-scale multi-source data, greatly improving the overall performance and practical application value of the algorithm.

[0066] Furthermore, the present invention excels in real-time monitoring and dynamic adjustment. By monitoring the semantic transfer entropy of logical transfer nodes in real time and promptly triggering a simulated annealing optimization mechanism when problems arise, it replans the spatiotemporal collaborative paths of cross-language mappings. This allows for rapid response and resolution of semantic instability issues that arise during vocabulary expansion, further enhancing the stability and reliability of the present invention in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A diagram showing the working principle of the cross-domain vocabulary logic reasoning and expansion method of the present invention;

[0068] Figure 2 Diagram showing the working principle of the adaptive semantic extended threshold model;

[0069] Figure 3 A flowchart constructed for the causal inference framework;

[0070] Figure 4 Flowchart for training a hypergraph attention network. DETAILED DESCRIPTION

[0071] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0072] See also Figures 1-4 The present invention provides a cross-domain vocabulary logic reasoning and expansion method, and its specific implementation steps are as follows:

[0073] Acquire a multi-source, cross-domain text data stream consisting of a multilingual corpus, a domain terminology library, and contextual metadata. Multilingual corpora encompass text information in multiple languages, providing rich material for cross-language processing; domain terminology libraries store domain-specific vocabulary and their definitions, facilitating accurate understanding and processing of domain-related content; and contextual metadata, containing the contextual information of vocabulary within the text, plays a key role in semantic understanding and reasoning.

[0074] Based on a pre-trained semantic decoupling model, the received text data undergoes cross-language alignment and semantic fusion. This model matches and integrates vocabulary and semantics across texts in different languages, generating a cross-domain semantic network graph and a vocabulary expansion probability matrix. The cross-domain semantic network graph graphically displays the semantic connections between words, while the vocabulary expansion probability matrix provides a probabilistic basis for subsequent vocabulary expansion.

[0075] Based on the generated semantic network graph, an adaptive semantic extension threshold model is constructed. This model dynamically divides lexical semantic density levels, calculates lexical extension risk scores, generates domain-specific extension thresholds, and generates a multi-granularity lexical inference rule set based on these thresholds. The rule set includes a cross-language mapping topology and a dynamic semantic path generation strategy. These rules provide specific guidance for lexical inference and extension across different domains and languages.

[0076] Based on a predefined causal reasoning framework, we simulate the logical transfer paths of vocabulary between different fields. By collecting relevant data and analyzing the transfer patterns of vocabulary between different fields, we optimize the semantic coordination parameters in the inference rule set, making the inference rules more accurate and effective.

[0077] Leveraging a distributed semantic collaborative learning framework, collaborative parameters are iteratively optimized. This framework continuously adjusts parameters by defining a reward function and employing a phased adversarial training strategy. Ultimately, it outputs a cross-domain vocabulary expansion sequence, which is then transferred to a multilingual knowledge graph platform, effectively expanding the cross-domain vocabulary and enriching the knowledge graph.

[0078] The implementation of the present invention will be further described below with reference to Examples 1 to 6.

[0079] Example 1:

[0080] In this embodiment, the process of constructing a semantic decoupling model is described. Specifically, it includes:

[0081] We collect parallel text corpora from multiple fields. These corpora come from a wide range of sources, such as academic literature databases collecting bilingual Chinese-English literature from various disciplines, and professional texts in multiple languages from professional translation websites. Using specialized annotation tools and methods, we construct a multidimensional training set that includes semantic role annotation, syntactic dependency trees, and cross-language alignment relationships. Semantic role annotation clarifies the semantic role of each word in a sentence, such as agent or patient; syntactic dependency trees demonstrate the syntactic structural relationships between words in a sentence; and cross-language alignment relationships establish correspondences between words in different languages.

[0082] Quantum semantic embedding technology is used to perform latent semantic decomposition on a multidimensional training set. Based on the principles of quantum mechanics, quantum semantic embedding can map the semantic information of a word into a low-dimensional vector space, thereby extracting independent representations of the word's primary meaning and contextual modifiers. In this process, the primary meaning represents the core semantic connotation of the word itself, while the contextual modifiers reflect the semantic changes of the word in a specific context.

[0083] Combining the lexical logic transfer equation, we construct topological constraints for the dynamic associations between semantic factors. This mathematical model, based on the semantic transfer patterns of words in different contexts and domains, describes how lexical semantics change under the influence of different factors. By analyzing these relationships, we determine the topological constraints between semantic factors, ensuring the rationality and accuracy of semantic transfer.

[0084] Topological constraints are embedded in the hypergraph attention network. Semantic stability analysis is performed on the results of latent semantic decomposition to identify logically transferable association patterns. Random field sampling is used to generate vocabulary evolution trajectories that conform to semantic constraints. These trajectories are then used to perform sparse training on the hyperedge weights of the hypergraph attention network. The hypergraph attention network automatically assigns attention weights based on the associations between words, highlighting important semantic connections. After training, the model output conforms to cross-domain logical transfer rules, resulting in a semantically decoupled model that supports incremental learning. Incremental learning enables the model to continuously optimize and improve performance as new data is continuously acquired, adapting to different application scenarios.

[0085] Throughout the entire process, each step is closely interconnected and mutually influential. For example, the quality of the multidimensional training set directly affects the effectiveness of latent semantic decomposition, while the topological constraints between semantic factors determine the accuracy of hypergraph attention network training. Through precise construction and optimization, the semantic decoupling model can provide strong support for subsequent cross-language alignment and semantic fusion.

[0086] Example 2:

[0087] The adaptive semantic expansion threshold model first dynamically divides lexical semantic density levels based on the node connectivity of the semantic network graph. Nodes in the semantic network graph represent words, and node connectivity reflects the closeness of the word's relationship with other words. Words with high connectivity have relatively high semantic density levels, indicating that they occupy a relatively central position in the semantic network and have a greater impact on semantic understanding and expansion. Words with low connectivity have low semantic density levels.

[0088] Next, a vocabulary expansion risk score is calculated based on the domain term coverage and semantic ambiguity index. A real-time semantic conflict matrix and domain term coverage heat map are obtained to construct a vocabulary expansion risk assessment cube. The real-time semantic conflict matrix records the potential semantic conflicts that may arise with other terms during the vocabulary expansion process. The greater the semantic conflict, the higher the risk. The domain term coverage heat map shows the coverage of domain terms in different regions. Regions with low coverage may have more semantic uncertainty.

[0089] The low-rank association weights of multi-dimensional semantic features are calculated using a tensor decomposition network. The tensor decomposition network can reduce the dimensionality of high-dimensional semantic feature data and extract key association weights, which reflect the relative importance of different semantic features. The association weights are then combined with the assessment cube using a super-tensor contraction operation to obtain a comprehensive risk score. The formula for calculating the comprehensive risk score is:

[0090]

[0091] Where, Γ represents the comprehensive risk score value, which is a quantitative indicator to measure the risk of vocabulary expansion; α i represents the intensity of semantic conflict of type i, which is used to describe the severity of different types of semantic conflicts; β i represents the adjustment coefficient of the i-th type of conflict, which adjusts its impact on the comprehensive score according to the characteristics of different semantic conflicts; γ c It represents the domain-based semantic compatibility constant, reflecting the semantic compatibility of the domain itself. The value of this constant varies in different domains. ⊙ represents the Hadamard product, which is used to multiply corresponding elements. n represents the total number of semantic conflict classifications. By classifying and counting semantic conflicts, various possible conflict situations are comprehensively considered.

[0092] Risk scores and semantic density levels are nonlinearly mapped using a hyperbolic function to generate domain-specific expansion thresholds. This function allows for a reasonable transformation of risk scores and semantic density levels, making the generated thresholds more consistent with practical application requirements. This threshold triggers the activation conditions of multi-level lexical inference rules, thereby controlling the scope and method of lexical expansion and ensuring its accuracy and effectiveness.

[0093] By accurately calculating the vocabulary expansion risk score and reasonably generating domain-specific expansion thresholds, the adaptive semantic expansion threshold model can provide a reliable basis for vocabulary expansion in different fields and effectively avoid erroneous expansion caused by semantic conflicts and uncertainties.

[0094] Example 3:

[0095] The steps to build a causal inference framework include:

[0096] We collect historical cross-domain vocabulary migration data and construct a semantic causal graph dataset. This data can be extracted from a large amount of cross-domain text, such as multilingual technical documents and academic papers. By analyzing the migration of vocabulary across different domains and languages, we construct a semantic causal graph that records the causal relationships of vocabulary migration.

[0097] The probability and timing delay parameters of lexical migration paths are extracted using a counterfactual reasoning algorithm. Counterfactual reasoning algorithms reason based on hypothetical scenarios, analyzing the probability and timing delay of migration paths by comparing actual lexical migration with a hypothetical scenario where migration did not occur. For example, if a certain word does not migrate across domains at a certain point in time, the impact on subsequent semantic understanding of the text can be observed to determine the importance and probability of the migration path.

[0098] Combining semantic complex network theory with word-transitive directed hypergraphs, this approach quantifies the strength of logical dependencies between nodes. Semantic complex network theory considers words as nodes in a network and the semantic relationships between words as edges. By constructing directed hypergraphs, we can more accurately describe the complex logical dependencies between words. Quantifying the strength of logical dependencies between nodes provides precise data support for subsequent reasoning.

[0099] The dependency strength and real-time context perturbation factor are input into the spatiotemporal gating network to generate a causal inference framework. The real-time context perturbation factor reflects the dynamic influence of the current text context on the semantic meaning of the word. Based on this information, the spatiotemporal gating network can dynamically adjust the focus on the word transmission path, thereby generating an accurate causal inference framework.

[0100] Based on the simulation results of the causal reasoning framework, key logical transmission interruption nodes are identified. These nodes may cause lexical logic transmission interruption due to semantic conflicts, domain differences, and other reasons. Semantic compensation strategies are configured for these nodes in the lexical reasoning rule set, such as by finding synonyms or adjusting the semantic scope of the lexical words.

[0101] Based on the compensation strategy, a cross-language collaborative mapping solution is automatically generated, including root alignment instructions and semantic bridging paths. Root alignment instructions align words with the same or similar roots in different languages, while semantic bridging paths establish semantic connections between words in different domains, enabling smooth transfer and expansion of vocabulary across different domains and languages.

[0102] Throughout the entire process, the construction of a causal reasoning framework provides a basis for identifying logical transmission interruption nodes, while the semantic compensation strategy and cross-language collaborative mapping scheme effectively solve the problem of logical interruption in the process of cross-domain vocabulary expansion, ensuring the coherence and accuracy of vocabulary expansion.

[0103] Embodiment 4:

[0104] In this embodiment, the specific process of embedding topological constraints into the hypergraph attention network is described in detail.

[0105] After completing latent semantic decomposition, the decomposition results are subjected to semantic stability analysis. This analysis assesses the stability of a word by calculating the degree of semantic variation across different contexts. For example, if a word's semantics remain relatively stable across multiple similar contexts, it is considered highly stable. Conversely, if its semantics vary significantly, its stability is low. This method screens out logically transitive association patterns, ensuring the rationality and reliability of the semantic associations in subsequent processing.

[0106] Next, random field sampling is used to generate a vocabulary evolution trajectory that conforms to semantic constraints. Random field sampling is a sampling method based on probability distribution. It samples words within a certain probability space according to their semantic features and established semantic constraints. For example, in a semantic network, the next possible word is randomly selected based on the strength of the semantic associations between words and the semantic constraints, thus generating a vocabulary evolution trajectory. During the trajectory generation process, semantic constraints are strictly adhered to, ensuring that each word in the trajectory has a reasonable semantic connection with the previous word.

[0107] This trajectory data is used to perform sparse training on the hyperedge weights of the Hypergraph Attention Network. Hyperedges in the Hypergraph Attention Network connect multiple nodes and represent complex semantic relationships between multiple words. During training, the network uses trajectory data to learn which hyperedges are more important for lexical semantic transmission and which can be weakened or even removed. Sparse training allows the Hypergraph Attention Network to focus more on key semantic associations, reducing unnecessary computation while improving the accuracy and efficiency of the model. During training, the hyperedge weights are continuously adjusted to ensure that the model output conforms to the laws of cross-domain logical transmission. For example, for words with similar semantic functions in different domains, the model can, through training, strengthen the hyperedge weights between them, thereby better capturing and transmitting semantic information.

[0108] Through the above steps, the topological constraints are effectively embedded in the hypergraph attention network, so that the semantic decoupling model can better handle the semantic relations of cross-domain vocabulary and provide stronger support for cross-domain vocabulary logical reasoning and expansion.

[0109] Example 5:

[0110] This embodiment details the execution process of the distributed semantic collaborative learning framework and the quantification method of the expansion efficiency target.

[0111] First, we define the reward function for multi-agent semantic collaboration, which includes the semantic consistency goal and the expansion efficiency goal. The reward function is:

[0112] R=λ·S c +(1-λ)·θ

[0113] In the formula, R represents the reward value, which is used to measure the comprehensive performance of multiple agents in the semantic collaboration process; S c represents the semantic consistency score, which reflects the degree of match between the expanded vocabulary and the original semantics. A higher score indicates better semantic consistency. θ represents the expansion efficiency score, which reflects the speed of vocabulary expansion and resource utilization efficiency. λ is a dynamic adjustment factor, which adjusts the weights of the semantic consistency goal and the expansion efficiency goal in the reward function according to the actual application requirements and the task focus at different stages.

[0114] In order to achieve the quantitative expansion efficiency goal, it is necessary to establish a semantic utility decay model for multi-granularity vocabulary resources and define a resource redundancy penalty function. The calculation formula of the semantic utility decay model is:

[0115]

[0116] Where K j represents the initial utility value of the jth type of lexical resource. Different types of lexical resources, such as professional terms and common words, have different initial utilities. j represents the semantic time decay coefficient, which reflects the decay rate of the semantic utility of lexical resources over time. Different types of lexical resources have different decay rates; δ j (t) represents the resource activation status indicator function at time t. When the j-th type of lexical resource is activated at time t, its value is 1, otherwise it is 0; m represents the total number of resource types, covering all different types of lexical resources involved in lexical expansion.

[0117] Based on the semantic network evolution phase diagram, the marginal utility of resource scheduling at each stage is calculated. This diagram shows the development status and changing trends of the semantic network at different stages. By analyzing the phase diagram, the marginal utility of resource scheduling on vocabulary expansion at each stage is determined. Using this marginal utility value as a dynamic gain coefficient in the reward function allows the reward function to be dynamically adjusted based on the actual effectiveness of resource scheduling, incentivizing multi-agents to more rationally allocate resources and expand vocabulary.

[0118] The policy network of each language agent is trained using a phased adversarial training strategy. During training, different language agents compete and collaborate with each other to continuously optimize their strategies. In each round of training, the gradient allocation weights between agents are dynamically adjusted based on the degree of goal conflict. For example, when there is a significant conflict between semantic consistency and expansion efficiency, the gradient allocation weight for the key goal is appropriately increased to balance the achievement of both goals. The final output is a vocabulary expansion sequence that satisfies Pareto optimality, maximizing the performance of one goal without compromising the performance of other goals.

[0119] Through the above process, the distributed semantic collaborative learning framework can effectively optimize the collaborative parameters, achieve efficient cross-domain vocabulary expansion, and at the same time ensure the semantic accuracy of the expansion results.

[0120] Example 6:

[0121] After configuring semantic compensation strategies for key logical transfer interruption nodes, the semantic transfer entropy of these nodes is monitored in real time. Semantic transfer entropy is a metric that measures the uncertainty in the transfer of semantic information between nodes. By calculating semantic transfer entropy, we can gain real-time insights into the stability and reliability of vocabulary during the logical transfer process. For example, when a word is transferred from one domain to another, a low semantic transfer entropy indicates stable semantic transfer and minimal information loss. Conversely, a high semantic transfer entropy indicates greater uncertainty and information loss.

[0122] If the semantic transfer entropy exceeds a preset threshold, it indicates a significant problem in the logical transfer process, potentially affecting the accuracy and effectiveness of vocabulary expansion. At this point, the simulated annealing optimization mechanism is triggered. This is an optimization algorithm based on a physical annealing process that can accept poorer solutions with a certain probability, thus avoiding falling into local optimal solutions. In this invention, the simulated annealing optimization mechanism is used to replan the spatiotemporal collaborative paths for cross-language mapping.

[0123] During the replanning process, we first analyze the causes of the current logical transfer interruption and the factors that contribute to the excessively high semantic transfer entropy. This may be due to an incomplete semantic compensation strategy or other unaccounted-for semantic conflicts in the cross-language mapping process. Based on the analysis results, we adjust the cross-language mapping strategy, such as reselecting root alignment methods and optimizing semantic bridging paths. During this adjustment process, we leverage the characteristics of the simulated annealing algorithm to gradually identify optimal solutions. After each adjustment, we recalculate the semantic transfer entropy to determine whether the requirements are met. If, after multiple adjustments, the semantic transfer entropy drops below the preset threshold, the replanned spatiotemporal collaborative path for cross-language mapping is effective and can ensure the smooth transfer and expansion of vocabulary across different domains and languages.

[0124] By real-time monitoring of semantic transfer entropy and timely triggering of the simulated annealing optimization mechanism, the spatiotemporal collaborative path of cross-language mapping can be dynamically adjusted and optimized to ensure the stability and accuracy of the vocabulary expansion process, further improving the performance of the cross-domain vocabulary logic reasoning and expansion method of the present invention.

[0125] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0126] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A cross-domain vocabulary logic reasoning and expansion method, characterized in that: include: Receiving a multi-source cross-domain text data stream, wherein the data stream includes a multi-language corpus, a domain term library, and contextual metadata; Based on the pre-trained semantic decoupling model, the text data is cross-language aligned and semantically fused to generate a cross-domain semantic network graph and a vocabulary expansion probability matrix; Based on the semantic network graph, an adaptive semantic extension threshold model is constructed to generate a multi-granularity vocabulary inference rule set, wherein the rule set includes a cross-language mapping topology structure and a semantic path dynamic generation strategy; Based on a predefined causal reasoning framework, simulate the logical transfer path of vocabulary between different domains and optimize the semantic coordination parameters in the inference rule set; The collaborative parameters are iteratively optimized through a distributed semantic collaborative learning framework, and cross-domain vocabulary extension sequences are output to a multilingual knowledge graph platform.

2. The cross-domain vocabulary logical reasoning and expansion method according to claim 1 is characterized in that: The steps of constructing the semantic decoupling model include: Collect parallel text corpora from multiple fields and construct a multi-dimensional training set including semantic role annotation, syntactic dependency tree and cross-language alignment relations; Performing latent semantic decomposition on the multidimensional training set through quantum semantic embedding to extract independent representations of the main semantics of the vocabulary and contextual modification factors; Combined with the lexical logic transfer equation, the topological constraints of the dynamic association between semantic factors are constructed; The topological constraints are embedded in a hypergraph attention network to generate the semantic decoupling model that supports incremental learning.

3. The cross-domain vocabulary logical reasoning and expansion method according to claim 1 is characterized in that: The adaptive semantic extension threshold model includes: Dynamically dividing the vocabulary semantic density level according to the node connectivity of the semantic network graph; Calculate the vocabulary expansion risk score based on domain term coverage and semantic ambiguity index; Performing nonlinear mapping of the risk score and the semantic density level using a hyperbolic function to generate a domain-specific expansion threshold; The activation condition of the multi-level lexical inference rule is triggered according to the threshold.

4. The cross-domain vocabulary logical reasoning and expansion method according to claim 1 is characterized in that: The steps for constructing the causal reasoning framework include: Collect historical cross-domain vocabulary migration data and construct a semantic causal graph dataset; The probability and temporal delay parameters of the vocabulary transfer path are extracted through the counterfactual reasoning algorithm; Combining semantic complex network theory to construct a lexical transitive directed hypergraph and quantify the logical dependency strength between nodes; The dependency strength and the real-time context disturbance factor are input into a spatiotemporal gating network to generate the causal reasoning framework.

5. The cross-domain vocabulary logical reasoning and expansion method according to claim 4 is characterized in that: Also includes: Identifying key logic transfer interruption nodes based on simulation results of the causal reasoning framework; configuring a semantic compensation strategy for the node in the lexical inference rule set; Based on the compensation strategy, a cross-language collaborative mapping solution is automatically generated, including root alignment instructions and semantic bridging paths.

6. The cross-domain vocabulary logical reasoning and expansion method according to claim 3 is characterized in that: The calculation of the vocabulary expansion risk score includes: Obtain real-time semantic conflict matrices and domain term coverage heat maps to build a vocabulary expansion risk assessment cube; Calculate the low-rank association weights of multi-dimensional semantic features through tensor decomposition network; Performing a hypertensor contraction operation on the association weight and the assessment cube to obtain a comprehensive risk score; The calculation formula for the comprehensive risk score is: In the formula, Γ represents the comprehensive risk score value, α i represents the semantic conflict intensity of the i-th category, β i represents the adjustment coefficient of the i-th type of conflict, γ c represents the domain-based semantic compatibility constant, ⊙ represents the Hadamard product, and n represents the total number of semantic conflict classifications.

7. The cross-domain vocabulary logical reasoning and expansion method according to claim 2 is characterized in that: The embedding of the topological constraint condition includes: Performing semantic stability analysis on the latent semantic decomposition results to screen logically transitive association patterns; Generate vocabulary evolution trajectories that conform to semantic constraints through random field sampling; The trajectory data is used to perform sparse training on the hyperedge weights of the hypergraph attention network to ensure that the model output conforms to the cross-domain logical transmission rules.

8. The cross-domain vocabulary logical reasoning and expansion method according to claim 1 is characterized in that: The execution of the distributed semantic collaborative learning framework includes: Define the reward function for multi-agent semantic collaboration, including semantic consistency goals and extended efficiency goals; Train the policy network of each language agent through a staged adversarial training strategy; In each round of training, the gradient distribution weights between agents are dynamically adjusted according to the degree of goal conflict; outputting the vocabulary expansion sequence that satisfies the Pareto optimality condition; Wherein, the reward function is: R=λ·S c +(1-λ)·θ In the formula, R represents the reward value, S c represents the semantic consistency score, θ represents the expansion efficiency score, and λ is the dynamic adjustment factor.

9. The cross-domain vocabulary logical reasoning and expansion method according to claim 5, characterized in that: Also includes: After configuring the semantic compensation strategy, the semantic transfer entropy of the logic transfer node is monitored in real time; If the transfer entropy exceeds a preset threshold, a simulated annealing optimization mechanism is triggered to replan the spatiotemporal collaborative path of the cross-language mapping.

10. The cross-domain vocabulary logical reasoning and expansion method according to claim 8, characterized in that: The quantification of the extended efficiency targets includes: Establish a semantic utility decay model for multi-granularity lexical resources and define a resource redundancy penalty function; According to the semantic network evolution phase diagram, the marginal utility value of resource scheduling in each stage is calculated; Using the marginal utility value as a dynamic gain coefficient of the reward function; The calculation formula of the attenuation model is: Where K j represents the initial utility value of the j-th category of vocabulary resources, η j represents the semantic time attenuation coefficient, δ j (t) represents the resource activation status indicator function at time t, and m represents the total number of resource types.

Images