A method and system for personal thought mimicry

Abstract

The application discloses a kind of personal thinking simulation method and system, comprising: first, target user is carried out multi-source data acquisition and privacy preprocessing in public, private and dynamic questionnaire, and obtains desensitization behavior data.Secondly, features are extracted by three-level segmentation strategy, and multidimensional joint semantic vector is obtained by cross-modal association of graph attention network.Accordingly, graph database and vector storage database containing five-dimensional knowledge graph of cognition are constructed.Combined with five-dimensional knowledge graph, preset template and real-time environmental data, dynamic prompt words are constructed, and after multi-level retrieval enhancement, the final decision suggestion for target user is generated by combining large model, to realize accurate personal thinking simulation and efficient decision assistance, and privacy protection is considered.

Description

Technical Field

[0001] This invention relates to the field of personal thought imitation, and more specifically, to a method and system for imitating personal thought. Background Technology

[0002] With the development of artificial intelligence technology, the demand for simulating and understanding user behavior and thought patterns is growing. Traditional user profiling has limitations such as being single-dimensional and static, making it difficult to comprehensively and accurately depict user characteristics. Meanwhile, privacy protection is also a major concern during data collection and processing. Existing methods are insufficient in integrating multi-dimensional information, dynamic modeling, and privacy enhancement processing, failing to adequately meet the requirements for deep imitation of individual thinking and providing effective decision support. Therefore, a more advanced method for imitating individual thinking is needed to overcome these technological bottlenecks. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for imitating personal thinking.

[0004] In a first aspect, embodiments of the present invention provide a method for imitating personal thought, comprising:

[0005] Multi-source data collection and privacy preprocessing were performed on target users from public domains, private domains, and dynamic questionnaires to obtain de-identified behavioral data.

[0006] A three-level segmentation strategy is used to extract features from the desensitized behavior data, and cross-modal association is performed through a graph attention network to obtain a multi-dimensional joint semantic vector;

[0007] Based on the multi-dimensional joint semantic vectors, a graph database and a vector storage database are constructed. The graph database is used to store a five-dimensional knowledge graph, which includes cognitive, behavioral, emotional, social, and evolutionary dimensions.

[0008] Dynamic prompt words are constructed based on the aforementioned five-dimensional knowledge graph, the preset five-dimensional weight injection template, and real-time environmental data;

[0009] Multi-level retrieval enhancement is performed based on the dynamic prompts, the graph database, and the vector storage database, and a final decision suggestion is generated by combining it with a pre-trained large model as a personalized thinking imitation result for the target user.

[0010] Secondly, embodiments of the present invention provide a personal thought imitation system, comprising:

[0011] The acquisition module is used to collect multi-source data from target users in public domains, private domains, and dynamic questionnaires, and perform privacy preprocessing to obtain anonymized behavioral data. A three-level segmentation strategy is used to extract features from the anonymized behavioral data, and cross-modal association is performed through a graph attention network to obtain multi-dimensional joint semantic vectors. Based on the multi-dimensional joint semantic vectors, a graph database and a vector storage database are constructed. The graph database is used to store a five-dimensional knowledge graph, which includes cognitive, behavioral, emotional, social, and evolutionary dimensions. Dynamic prompt words are constructed based on the five-dimensional knowledge graph, a preset five-dimensional weight injection template, and real-time environmental data.

[0012] The execution module is used to perform multi-level retrieval enhancement based on the dynamic prompts, the graph database, and the vector storage database, and combine them with a pre-trained large model to generate a final decision suggestion as a personalized thinking imitation result for the target user.

[0013] Compared to existing technologies, the beneficial effects provided by this invention include: The method and system for imitating personal thinking, as disclosed in this invention, include: firstly, collecting multi-source data from public, private, and dynamic questionnaires of the target user and performing privacy preprocessing to obtain de-identified behavioral data. Then, extracting features using a three-level segmentation strategy, and using a graph attention network to cross-modal associate multi-dimensional joint semantic vectors. Based on this, constructing a graph database containing a five-dimensional knowledge graph including cognition and a vector storage database. Combining the five-dimensional knowledge graph, preset templates, and real-time environmental data, constructing dynamic prompt words, and after multi-level retrieval enhancement, jointly generating final decision suggestions for the target user using a large model, achieving accurate personal thinking imitation and efficient decision assistance while protecting privacy. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart illustrating the steps of the personal thought imitation method provided in an embodiment of the present invention;

[0016] Figure 2 A hierarchical framework diagram of a personal thought imitation system provided in an embodiment of the present invention;

[0017] Figure 3 This is an interactive schematic diagram of a personal thought imitation system provided in an embodiment of the present invention;

[0018] Figure 4A schematic block diagram of the structure of the personal thought imitation system provided in the embodiments of the present invention;

[0019] Figure 5 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0022] In order to solve the technical problems mentioned in the background art Figure 1 This is a flowchart illustrating a personal thought imitation method provided in an embodiment of the present disclosure. The personal thought imitation method will be described in detail below.

[0023] Step S201: Collect multi-source data from the target user in the public domain, private domain and dynamic questionnaire and perform privacy preprocessing to obtain de-identified behavioral data;

[0024] Step S202: A three-level segmentation strategy is used to extract features from the desensitized behavior data, and cross-modal association is performed through a graph attention network to obtain a multi-dimensional joint semantic vector;

[0025] Step S203: Construct a graph database and a vector storage database based on the multi-dimensional joint semantic vector. The graph database is used to store a five-dimensional knowledge graph, which includes cognitive dimension, behavioral dimension, emotional dimension, social dimension and evolutionary dimension.

[0026] Step S204: Construct dynamic prompt words based on the five-dimensional knowledge graph, the preset five-dimensional weight injection template, and real-time environmental data;

[0027] Step S205: Perform multi-level retrieval enhancement based on the dynamic prompt words, the graph database, and the vector storage database, and combine with a pre-trained large model to generate a final decision suggestion as a personalized thinking imitation result for the target user.

[0028] In this embodiment of the invention, for example, the local terminal serves as one of the starting points for data collection, undertaking the task of accessing user private domain data and interaction data. Assume the target user is a researcher named Xiao Li.

[0029] Private Data Collection: Xiao Li uses Obsidian for daily research note-taking. The local machine connects to the Obsidian note-taking tool, acquiring Xiao Li's notes in real time through relevant technologies. For example, Xiao Li recorded his thoughts on an interdisciplinary research project in medicine and engineering in Obsidian, including inductive reasoning about different research methods and deductive reasoning about research directions. These notes contain rich cognitive information, and the local machine collects them as private data. Simultaneously, Xiao Li also records his project progress and daily thoughts in Notion; the local machine also supports importing Notion data, incorporating it into the category of private data.

[0030] Interactive Data Collection: The system provided Xiao Li with an embedded dynamic questionnaire system (customized by Typeform), supporting voice and gesture input. During an AI training session, the system displayed a dynamic scenario-based multiple-choice question in real time: "When conducting scientific research experiments, would you prefer the system to prioritize logical rigor or emotional resonance?" Xiao Li selected logical rigor via voice, and the local device collected this selection as interactive data. Furthermore, Xiao Li can also interact with the system through voice and gestures in other ways, such as using gestures to emphasize key points when explaining his research ideas; the local device collects all these interactive information.

[0031] The edge device plays a certain role in relaying and preliminary processing during the data collection process, and is also responsible for some of the collection of public domain data.

[0032] Public Domain Data Collection: Taking Weibo as an example, the edge client obtains Xiao Li's public interaction data through the Weibo Open Platform API. Xiao Li frequently forwards public welfare-related scientific research content on Weibo, such as public welfare projects related to rare disease research, reflecting his inclination towards social responsibility. He also publishes original science and technology-related answers, explaining his rational decision-making preferences on certain scientific research issues. The edge client collects data such as forwards, comments, likes, and original content. In addition, the edge client also integrates public domain data such as GitHub code submission patterns and Zhihu answer styles. Suppose Xiao Li frequently submits code for scientific algorithm optimization on GitHub at night, reflecting his characteristic of "deep focus" in the cognitive dimension; the edge client will integrate and collect this cross-platform public domain data.

[0033] Whether it's private domain data collected locally or public domain data collected at the edge, it all needs to go through a privacy preprocessing module.

[0034] Local preprocessing: On the local machine, for Xiao Li's private data, such as sensitive information in his notes (e.g., specific research experiment details that may involve the confidentiality of personal research results), gradient anonymization (ε = 0.4) will be implemented. This means that differential privacy noise will be added to the data, making individual data entries untraceable. For sensitive information such as geographical location, generalization processing will be performed. For example, if Xiao Li occasionally records the location of an academic conference he attended in his notes, the system will generalize the specific street address to a city level.

[0035] Cloud synchronization preprocessing: Before synchronizing data to the cloud, k-anonymization (k=50) is performed to obfuscate user group characteristics. For example, the relevant data of Xiao Li and 49 other users are mixed, making it impossible to accurately identify Xiao Li's individual characteristics from the cloud data, thereby protecting Xiao Li's privacy.

[0036] After the above multi-source data collection and privacy preprocessing, de-identified behavioral data is obtained, laying the foundation for subsequent feature extraction and other steps.

[0037] A three-level segmentation strategy is adopted to extract features from the de-identified behavioral data. This process is mainly completed by the fractal parsing engine in the feature processing layer.

[0038] Core Concept Extraction: Using a draft research paper by Xiao Li as a data example, the LayoutLMv3 model was used to analyze the document. The LayoutLMv3 model can identify information such as titles and paragraph structure in a document and extract core concepts from them. For example, the title "Research on Novel Algorithms for Medical Image Processing" and related chapter themes in the draft paper were identified as core concepts. These concepts reflect Xiao Li's core focus in the research and belong to important information in the cognitive dimension.

[0039] Supporting Argument Identification: SpaCy_NER was used to process the document and identify supporting arguments. In Xiao Li's draft paper, SpaCy_NER was able to identify the explanations of the principles of the new algorithm and the comparison of experimental data as supporting arguments for the main concepts. These arguments further enriched the understanding of Xiao Li's research thinking and helped to grasp his cognitive characteristics more comprehensively.

[0040] Meta-knowledge atom decomposition: By processing the document using Tesseract_OCR, its content is broken down into meta-knowledge atoms. For example, technical terms and formulas in a draft paper are broken down into the most basic knowledge units. These meta-knowledge atoms are crucial for constructing knowledge graphs and for in-depth analysis of Xiao Li's knowledge system and cognitive structure.

[0041] The graph attention network (GAT) is used to perform cross-modal association on the features extracted by the three-level segmentation strategy, generating a 768-dimensional joint semantic vector.

[0042] Suppose that Xiao Li not only has text data such as drafts of scientific research papers, but also video of his speeches at academic conferences and related presentation slides.

[0043] Text-Image Association: In his presentation video, Xiao Li showcased sample medical images related to medical image processing algorithms mentioned in his draft paper. Graph Attention Networks can identify the association between descriptions of algorithms in text and image features in images, such as the correspondence between a certain image feature extraction method mentioned in the text and a specific image representation in the image, thereby associating the text and image data.

[0044] Text-Speech Association: During his presentation, Xiao Li's explanation of the algorithm also contained important information in his spoken words. Graph Attention Networks can align the key content conveyed in the speech with the information in the draft text of the paper. For example, the algorithm's advantages emphasized in the speech can be matched with the semantics of relevant paragraphs in the text, thus achieving an association between the text and speech modalities.

[0045] This cross-modal association yields a multi-dimensional joint semantic vector, which more comprehensively reflects Xiao Li's thinking and behavioral characteristics, providing a rich data foundation for subsequent steps such as building a knowledge graph.

[0046] The Neo4j graph database is constructed based on multi-dimensional joint semantic vectors to store five-dimensional knowledge graphs.

[0047] Cognitive Dimension Storage: Taking Xiao Li as an example, the thought patterns he demonstrates in his research notes and draft papers, such as inductive reasoning (summarizing various medical image processing methods), deductive reasoning (deriving processing methods for specific images from general algorithm principles), and cross-domain transferability (applying engineering optimization algorithms to medical image processing), are stored as nodes in the cognitive dimension in the graph database. Each node also includes a timestamp recording the time these thought patterns were manifested, and a confidence label reflecting the reliability of the system's judgments on these features. Nodes are connected by edges, which define the relationships between different thought patterns within the cognitive dimension, such as the logical relationship between a certain inductive reasoning method and subsequent deductive reasoning based on it.

[0048] Behavioral dimension storage: Xiao Li's public domain behaviors such as forwarding and commenting on Weibo, as well as behavioral data such as his selection preferences in dynamic questionnaires, are stored as nodes in the behavioral dimension. For example, the behavioral node of his frequent forwarding of public welfare scientific research content is connected to the related nodes of his social responsibility tendency shown in the dynamic questionnaire through an edge, reflecting the intrinsic connection between behaviors. At the same time, the behavioral dimension also stores information related to innovative measurement indicators such as value choice entropy and social influence radius.

[0049] Emotional Dimension Storage: Xiao Li's emotional expressions during academic exchanges, such as anxiety when discussing research challenges and joy when achieving research progress, are stored as nodes in the emotional dimension through emotional polarity information obtained from multimodal semantic parsing. Emotional information obtained by combining indirect inference from biosignals (e.g., inferring a potential stress-relieving emotional state from his frequent late-night likes on entertainment content in public data) is also stored in this dimension to construct a three-dimensional mapping model of "emotion-value-physiological response." Measurement indicators such as emotional network entropy are also recorded in the emotional dimension nodes.

[0050] Social dimension storage: This includes information about Xiao Li's role transitions on social networks (such as WeChat Moments and LinkedIn). For example, his role shift from being an information receiver in academic exchange groups to becoming a disseminator of valuable research information is stored as an important indicator in the nodes. Information such as the coupling strength between his work collaboration network and professional network is also stored in this dimension. For instance, his collaborative relationships with colleagues and connections with industry experts on LinkedIn are extracted across platforms to construct a composite social graph, which is then stored in the graph database.

[0051] Evolutionary Dimension Storage: Xiao Li's long-term digital footprint, such as the evolution of his blogging style over five years (from simply recording the research process to in-depth analysis of research problems and sharing research insights) and his choice of paid knowledge courses (from basic research methodology courses to cutting-edge interdisciplinary research courses), serves as important data storage for this evolutionary dimension. Key event markers, such as career turning points in his research career (promotion from ordinary researcher to project leader) and learning breakthroughs (mastering a key research technology), are also stored in this dimension. Information related to breakthrough indicators such as knowledge phase transition thresholds and fitness surfaces is also recorded in the evolutionary dimension nodes.

[0052] Multimodal feature vectors are stored using the Milvus vector database, supporting millisecond-level similarity retrieval (optimized by the Faiss engine).

[0053] Xiao Li's multimodal feature vectors, including text feature vectors from his research papers and image and speech feature vectors from his presentation videos, are stored in the Milvus vector database. When the system needs to find similar cases related to Xiao Li's current research problem, it searches the vector database using the Faiss engine. For example, when Xiao Li is researching a new medical image denoising algorithm, the system can quickly retrieve other cases in the database related to image processing algorithms with high feature vector similarity, providing him with references. When the similarity is greater than 0.7, association suggestions are triggered to help Xiao Li make better research decisions.

[0054] Dynamic prompts are constructed based on a five-dimensional knowledge graph, a pre-defined five-dimensional weight injection template, and real-time environmental data. Taking Xiao Li as an example, the system, based on his information in the five-dimensional knowledge graph, understands that he has strong logical reasoning and cross-domain transfer abilities in the cognitive dimension, a high sense of social responsibility in the behavioral dimension, relative rationality in the emotional dimension, a certain social influence in the social dimension, and is in the process of transitioning from single-discipline research to interdisciplinary research in the evolutionary dimension.

[0055] Suppose the current preset five-dimensional weight injection template is "cognitive weight 0.8 & emotional weight 0.6". Based on this template, the system will place more emphasis on Xiao Li's cognitive and emotional dimensions when generating prompts.

[0056] Real-time environmental data is used as an example of calendar schedules. Xiao Li's calendar shows that he has an important research project presentation in the next week, and his schedule is highly demanding (schedule_tension>0.8). Based on this real-time environmental data, the system will make adjustments when constructing dynamic prompts.

[0057] Based on the above information, the generated dynamic prompts are as follows:

[0058] ```markdown

[0059] #Cognitive weight 0.8 & Emotional weight 0.6

[0060] Generation rules:

[0061] Four logical comparison schemes are provided to highlight the cost-effectiveness analysis of different methods in medical image noise reduction algorithms. By leveraging Xiao Li's logical reasoning ability in the cognitive dimension, it helps him clearly explain the advantages of the algorithm in his project presentation.

[0062] Three similar historical cases were incorporated, and cases that were similar to the current research problem and reflected rational decision-making in the emotional dimension were retrieved from the vector storage database to provide a reference for Xiao Li's presentation.

[0063] The emotional intensity is limited to Level 2. Considering Xiao Li's relatively rational emotional nature and the formality of the project presentation, excessive emotional expression should be avoided.

[0064] Such dynamic prompts can provide targeted guidance for generating decision-making suggestions based on Xiao Li's personal characteristics and the real-time environment.

[0065] Multi-level retrieval enhancement and decision suggestion generation: Level 1: Faiss-based vector similarity retrieval: Based on the requirements in the dynamic prompts, the system first performs a Faiss-based vector similarity retrieval in the Milvus vector database. Taking Xiao Li's medical image denoising algorithm research as an example, the system retrieves the Top 50 candidate cases with high similarity in feature vectors to the current research problem. These cases may include other similar image processing algorithm research, related technical applications, etc., providing an initial case pool for subsequent decision suggestions. Level 2: Graph neural network path credibility calculation: For the candidate cases obtained in the first level retrieval, the system calculates the path credibility in the Neo4j graph database using a graph neural network (optimized with the HNSW algorithm). For example, whether the association path between the candidate case and Xiao Li in the five-dimensional knowledge graph dimensions such as cognition and behavior is reasonable, and whether it can truly provide valuable reference for Xiao Li's current research. By calculating path credibility, more relevant and reliable cases are selected. Level 3: User profile relevance weighting: Based on Xiao Li's user profile, the relevance of the cases selected in the first two levels is weighted. Xiao Li has a high weight in the cognitive dimension (based on the five-dimensional weighting template and his own characteristics), and dimensions with a historical adoption rate >60% are prioritized. The system will prioritize cases closely related to Xiao Li's cognitive dimension and with a high past adoption rate to further optimize the search results and provide a more accurate basis for generating decision-making suggestions.

[0066] After multi-level retrieval enhancement, the system combines a pre-trained large model (such as GPT-4) to generate the final decision recommendation.

[0067] The system inputs filtered and weighted case information, along with dynamic prompts and other relevant content, into the large model. Based on this information, and combined with its language understanding and generation capabilities, the large model generates decision-making suggestions for Xiao Li regarding his research on medical image denoising algorithms. For example, the model might suggest that Xiao Li use a specific algorithm comparison presentation method in his project report, combining the successful experience of similar historical cases to highlight the algorithm's innovation and practicality; or, based on Xiao Li's emotional dimension characteristics, it might provide suggestions on how to appropriately express emotions in his report to enhance persuasiveness. These decision-making suggestions, as a result of imitating Xiao Li's personal thinking, can assist him in research decision-making and project reporting.

[0068] Based on the above description, this personal thinking imitation method can start from multi-source data collection, and after a series of complex and meticulous processing and analysis, ultimately generate decision-making suggestions for target users that fit their personal characteristics and actual needs, thus achieving accurate imitation and effective assistance of personal thinking.

[0069] In this embodiment of the invention, the step of collecting multi-source data from the target user in the public domain, private domain, and dynamic questionnaire and performing privacy preprocessing to obtain de-identified behavioral data can be implemented through the following example.

[0070] Public interaction data of the target user is obtained through the open platform API in the public domain;

[0071] By using a mind map-based structure parsing strategy in the private domain, the private data of the target user can be obtained;

[0072] By outputting a dynamic questionnaire with embedded dynamic situational multiple-choice questions, the questionnaire feedback of the target users is obtained;

[0073] The publicly available interactive data, the private data, and the questionnaire feedback are used as the multi-source data, and gradient desensitization and anonymization are performed to obtain the desensitized behavioral data.

[0074] In this embodiment of the invention, for example, it is assumed that the target user is a working professional named Xiao Zhang.

[0075] Regarding public domain data collection, the system obtained Xiao Zhang's public interaction data through Weibo's open platform API. Xiao Zhang frequently participates in industry topic discussions on Weibo, forwards articles about improving workplace skills, and comments on trending events. For example, he forwarded an article about "how to improve team collaboration efficiency" and added his own insights, believing that clear division of labor and effective communication are key; he also commented on a major industry collaboration, expressing optimism about the prospects of the partnership. This public interaction data, including forwards and comments, was acquired by the system and became part of the multi-source data.

[0076] In terms of private domain data collection, Xiao Zhang used Obsidian to record work notes, which had a mind map structure. The system analyzed Xiao Zhang's Obsidian notes based on a mind map structure parsing strategy. In his notes, Xiao Zhang outlined a project planning process in the form of a mind map, including goal setting, task allocation, and timeline scheduling. By analyzing the nodes and connections in the mind map, the system obtained private data such as Xiao Zhang's thinking patterns and decision-making logic during the project planning process. For example, the mind map showed that Xiao Zhang considered the strengths and expertise of team members when allocating tasks, reflecting his cognitive and behavioral tendencies in personnel management.

[0077] To obtain feedback from the questionnaire, the system outputs a dynamic questionnaire with embedded dynamic scenario-based multiple-choice questions to Xiao Zhang. In a training session on work decision-making, the dynamic questionnaire presented the question: "When faced with an urgent and important task, and at the same time, the team members are experiencing resource shortages, would you choose to prioritize allocating external resources or try internal coordination?" After consideration, Xiao Zhang chose to try internal coordination, and the system recorded this feedback.

[0078] The system uses public Weibo interaction data obtained from Xiao Zhang, private Obsidian notes data, and dynamic questionnaire feedback as multi-source data. This data then undergoes privacy preprocessing. In the gradient anonymization process, differential privacy noise is added to content in Xiao Zhang's Weibo comments that may contain sensitive personal opinions, making it difficult to trace individual comments back to Xiao Zhang. During anonymization, Xiao Zhang's data is mixed with data from a certain number of other users, implementing k-anonymization (assuming k=50) to obfuscate user group characteristics, ultimately obtaining anonymized behavioral data. This anonymized behavioral data will be used for subsequent feature extraction and other operations to simulate and analyze Xiao Zhang's personal thinking.

[0079] In this embodiment of the invention, the three-level segmentation strategy is used to extract features from the desensitized behavior data, and cross-modal association is performed through a graph attention network to obtain a multi-dimensional joint semantic vector. This can be implemented through the following example.

[0080] The de-identified behavior data is sequentially processed by extracting the main concepts based on LayoutLMv3, identifying supporting arguments based on SpaCy_NER, and performing meta-knowledge atomic decomposition based on Tesseract_OCR to obtain multiple features to be processed based on the three-level segmentation strategy.

[0081] The multiple features to be processed are adjusted and aligned using a cross-modal attention mechanism;

[0082] The graph attention network is used to perform cross-modal association on multiple aligned features to obtain the multi-dimensional joint semantic vector. The cross-modal association includes lexical association based on FastText cross-training, entity association based on the TransEdge algorithm, attribute association based on graph attention network, instance association based on adversarial training, and rule association based on SWRL inference.

[0083] In this embodiment of the invention, we will continue to take Zhang, a working professional, as an example.

[0084] When performing feature extraction for the three-level segmentation strategy, the first step is to extract the core concepts from Xiao Zhang's anonymized behavior data based on LayoutLMv3. Xiao Zhang's anonymized behavior data includes a project summary report document he wrote. The LayoutLMv3 model analyzes this document, identifying the title "Quarterly Project Summary and Future Planning" and the main themes of each chapter, such as "Overview of Project Achievements," "Problems Encountered and Solutions," and "Future Work Plan," which are determined to be the core concepts.

[0085] Next, SpaCy_NER was used to identify supporting arguments for Xiao Zhang's anonymized behavioral data. In the "Project Outcomes Overview" section, Xiao Zhang mentioned that "by optimizing the process, the project delivery time was shortened by 20% and the cost was reduced by 15%." SpaCy_NER was able to identify "optimized process," "20% reduction in delivery time," and "15% reduction in cost" as supporting arguments for the core concept of "significant project results."

[0086] Then, based on Tesseract_OCR, the anonymized behavioral data is decomposed into meta-knowledge atoms. Xiao Zhang's project summary report contains some technical terms, such as "Agile development" and "KPI assessment." Tesseract_OCR decomposes these terms, along with formulas and specific abbreviations in the report, into meta-knowledge atoms, thus obtaining multiple features to be processed based on a three-level segmentation strategy.

[0087] Subsequently, these multiple features to be processed were adjusted and aligned using a cross-modal attention mechanism. In addition to the text format, Xiao Zhang's project summary report also included a presentation with images and an oral presentation at a team meeting. The cross-modal attention mechanism aligns the core concepts and supporting arguments in the text with the charts and graphs in the images and the key points emphasized in the audio, ensuring semantic consistency across different modalities.

[0088] Finally, graph attention networks are used for cross-modal association. At the lexical level, based on FastText cross-training, technical terms in the text are associated with the same words mentioned in the audio; for example, "agile development" appears in both text and audio, establishing a connection between them. At the entity level, based on the TransEdge algorithm, the representations of specific entities in the project, such as team member names and partner company names, are associated across different modalities. At the attribute level, graph attention networks are used to analyze the relationship between project attributes described in the text (such as delivery time and cost) and attributes displayed in charts in images. At the instance level, through adversarial training, information about specific project instances in different modalities is associated, such as the correspondence between the execution status of a specific task in text descriptions and image displays. At the rule level, based on SWRL inference, the manifestation of some project management rules in different modal data is derived, such as the consistency of performance appraisal rules in textual KPI assessment descriptions and audio explanations. Through this series of operations, multi-dimensional joint semantic vectors are obtained, providing a rich and closely related data foundation for subsequent knowledge graph construction.

[0089] In addition to the embodiments provided in this invention, the following implementation methods are also provided.

[0090] When the data deviation in at least three of the five dimensions of the knowledge graph—cognitive, behavioral, emotional, social, and evolutionary dimensions—exceeds a preset deviation threshold, a multimodal verification process is triggered, and the five-dimensional knowledge graph is adjusted based on the verification results.

[0091] The cognitive dimension employs an LSTM-GNN hybrid model to capture long-term cognitive transition trajectories; the behavioral dimension develops a cognitive-behavioral consistency verification algorithm, triggering a dual verification process when the deviation between the questionnaire statement and public domain behavior exceeds a preset deviation threshold; the emotional dimension develops an emotional state transition matrix, combined with a seasonal decomposition algorithm to identify periodic patterns; the social dimension introduces a dynamic community discovery algorithm and applies the preference attachment model from complex network theory to predict the trend of social influence changes within a preset time range; the evolutionary dimension introduces complex systems theory, establishes a cognitive attractor model to explain nonlinear growth patterns, and simulates cognitive evolution paths under different environments using the Markov chain Monte Carlo method.

[0092] In this embodiment of the invention, we will continue to use Zhang, a working professional, as an example to illustrate the details.

[0093] When monitoring Xiao Zhang's five-dimensional knowledge graph, if the data deviation in at least three of the four dimensions—cognitive, behavioral, emotional, social, and evolutionary—exceeds a preset deviation threshold (assuming a preset deviation threshold of 20%), a multimodal verification process will be triggered. For example, if the calculated deviations in Xiao Zhang's recent thinking pattern changes in the cognitive dimension, his choice preference data in the behavioral dimension, and his social relationship changes in the social dimension all exceed 20%, the multimodal verification process will be immediately initiated. This involves re-verifying and analyzing Xiao Zhang's text data, voice interaction data, and other multimodal information, and then adjusting the five-dimensional knowledge graph based on the verification results. In the cognitive dimension, a hybrid LTSTM / GNN model is used to capture Xiao Zhang's long-term cognitive transition trajectory. When Xiao Zhang first entered the workforce, he relied more on experience-driven approaches in project planning. As his work experience increased and he participated in various training programs and projects, he gradually shifted to a data-driven mindset, and later began to develop systems thinking. The LSTMGNN hybrid model analyzes Xiao Zhang's work documents, decision-making records, and other data from different stages, capturing his phased leaps from experience-driven to data-driven to systems thinking, and recording them in the cognitive dimension of the five-dimensional knowledge graph. For the behavioral dimension, a cognitive-behavioral consistency verification algorithm was developed. Xiao Zhang stated in the dynamic questionnaire that he would prioritize efficiency when time conflicts arise. However, analysis of his public domain behavioral data (such as his daily work shared on Weibo, where he repeatedly sacrificed efficiency to ensure work quality) revealed that the questionnaire statement deviated from his public domain behavior by more than a preset deviation threshold (assumed to be 25%). This triggered a double verification process, further combining Xiao Zhang's other behavioral data and interaction records for re-judgment and verification to ensure the accuracy of the behavioral dimension data. In the emotional dimension, an emotional state transition matrix was developed, and a seasonal decomposition algorithm was used to identify periodic patterns. At the end of each quarter, Xiao Zhang's behavior patterns in public domain data (such as frequently posting anxiety-related content on social media and frequently liking entertainment content late at night) and multimodal semantic analysis (analyzing the dynamic sentiment polarity of his posts) were used. Emotional state transition matrices and seasonal decomposition algorithms were employed to identify the cyclical pattern of his rising anxiety index at the end of each quarter, and this was recorded in the sentiment dimension. In the social dimension, a dynamic community discovery algorithm was introduced, and a preference attachment model from complex network theory was applied to predict the trend of social influence changes over a preset time frame (e.g., the next 6 months). On professional social networks like LinkedIn, the dynamic community discovery algorithm analyzed the changes in Xiao Zhang's social circles, revealing that he gradually integrated from a smaller technical exchange community into a larger community of industry elites. Simultaneously, using a preference attachment model combined with his social behavior data (such as the number of contacts added and the interaction volume of his posts), it was predicted that his social influence might gradually increase over the next 6 months, and this prediction information was recorded in the social dimension. In the evolutionary dimension, complex systems theory was introduced to establish a cognitive attractor model to explain Xiao Zhang's nonlinear growth pattern.After encountering a major industry transformation (such as the emergence of new technologies), Xiao Zhang's thinking patterns and behavioral habits underwent significant changes. By analyzing indicators such as the rate of change of his behavioral standard deviation before and after the transformation, and combining this with a cognitive attractor model, we can explain this non-linear growth. Simultaneously, we simulated Xiao Zhang's cognitive evolution path under different industry environments using the Markov chain Monte Carlo method, providing a basis for more accurately depicting his growth trajectory and recording it in the evolutionary dimension.

[0094] In addition to the embodiments provided in this invention, the following implementation methods are also provided.

[0095] The five-dimensional knowledge graph dynamically adjusts the weights of the cognitive, behavioral, emotional, social, and evolutionary dimensions through a five-dimensional weight matrix, and dynamically adjusts the emphasis of suggested dimensions in conjunction with real-time environmental data.

[0096] The five-dimensional weight matrix is ​​obtained through the formula: W t+1 =αW t +β(ΔF+γΔC), to achieve dynamic adjustment, where W t+1 For the adjusted weights, W t The weights before adjustment are α, the historical decay coefficient is β, the feedback gain is β, the cross-dimensional coupling factor is γ, the feedback weight adjustment factor is ΔF, and the dimension weight adjustment factor is ΔC.

[0097] In this embodiment of the invention, we will take Zhang, a working professional, as an example.

[0098] Xiao Zhang interacts with the system in various ways during his daily work. Based on this interaction data, the system constructs a five-dimensional knowledge graph about Xiao Zhang and dynamically adjusts the weights of the five dimensions—cognition, behavior, emotion, society, and evolution—using a five-dimensional weight matrix. Simultaneously, it dynamically adjusts the suggested dimension emphasis based on real-time environmental data. Firstly, the five-dimensional weight matrix is ​​based on the formula W... t+1 =αW t Dynamic adjustment is achieved using β(ΔF+γΔC). Assume that at a certain time t, the weight Wt (cognition) of the cognitive dimension in Xiao Zhang's five-dimensional weight matrix is ​​0.7, the historical decay coefficient α is set to 0.85, the feedback gain β (explicit feedback) is 1.5, β (implicit feedback) is 0.8, and the cross-dimensional coupling factor γ is dynamically determined based on the correlation between dimensions. Regarding explicit feedback, when Xiao Zhang uses the system to generate decision suggestions, he finds that the system's proposed solutions lack logical reasoning. He proactively adjusts the AI-generated content, rewriting the prompts to emphasize greater logical rigor; this is explicit feedback. At this point, the feedback weight adjustment factor ΔF (explicit) is 1. Since the cognitive dimension is closely related to logical reasoning, the system calculates the adjusted cognitive dimension weight Wt according to the formula. t+1 (Cognition): W t+1(Cognition) = 0.85 × 0.7 + 1.5 × 1 = 0.595 + 1.5 = 2.095, which is then constrained to between 0.1 and 1.0 using the np.clip function to obtain a suitable adjusted weight. Regarding implicit feedback, the system previously predicted that Xiao Zhang would consult a friend when handling a work project (prediction based on the social dimension), but Xiao Zhang actually chose independent decision-making, continuously deviating from the predicted path, which triggered implicit feedback. At this time, the feedback weight adjustment factor ΔF (implicit) is 0.6, and the system adjusts the weight of the social dimension according to the above formula. Simultaneously, the dimension weight adjustment factor ΔC considers the coupling relationship between dimensions. For example, the server analysis reveals a certain antagonistic effect between Xiao Zhang's cognitive and emotional dimensions (correlation coefficient of 0.32). When the cognitive dimension weight is adjusted, ΔC will take this cross-dimensional influence into account, thus comprehensively adjusting the weights of each dimension. When dynamically adjusting the emphasis of suggestion dimensions based on real-time environmental data, suppose Xiao Zhang's calendar shows that he will be giving an important project presentation in a week. This real-time environmental data indicates that he is currently in a relatively intense and important work phase. The server dynamically adjusts the dimensional emphasis of the decision suggestion based on the adjusted weights of each dimension in the five-dimensional weight matrix and the current real-time environmental data. Since Xiao Zhang has a higher weight in the cognitive dimension, and the project presentation requires strong logical thinking and clear expression, the server will place greater emphasis on the cognitive dimension when generating decision suggestions. For example, it will provide more suggestions on building a logical framework for the project presentation and data analysis and verification. At the same time, it will also appropriately consider other dimensions, such as the social dimension, providing suggestions on how to better communicate with team members and superiors during the presentation, to help Xiao Zhang better complete the project presentation.

[0099] In addition to the embodiments provided in this invention, the following implementation methods are also provided.

[0100] The de-identified behavior data is obtained locally using the AES-256 encryption algorithm;

[0101] The graph attention network is deployed locally to perform cross-modal association and combined with differential privacy noise to obtain a multi-dimensional joint semantic vector;

[0102] The multi-dimensional joint semantic vector is synchronized to the cloud based on anonymization processing; wherein, when synchronizing the multi-dimensional joint semantic vector between the local end and the cloud, only the dimension adjustment parameters are uploaded.

[0103] The parameters of the multi-dimensional joint semantic vector are aggregated and updated in the cloud using the FedAvg algorithm;

[0104] Deploy a parameter mapping gateway in the cloud to convert the multi-dimensional joint semantic vector into a large model adaptation format and connect it to the pre-trained large model.

[0105] In this embodiment of the invention, we will continue to take Zhang, a working professional, as an example.

[0106] Xiao Zhang uses the system on his personal device (local end). The system obtains his public Weibo interaction data through the public domain's open platform API, his private Obsidian notes data through the private domain's mind map structure parsing strategy, and his questionnaire feedback through the output of a dynamic questionnaire with embedded dynamic contextual multiple-choice questions. After obtaining this multi-source data, the system encrypts this anonymized data locally using the AES256 encryption algorithm. For example, Xiao Zhang's Weibo interaction data includes some of his work-related discussions, and his private notes record project planning details. This data is encrypted and stored locally to ensure data security and prevent unauthorized access and reading during local storage.

[0107] A graph attention network was deployed locally to perform cross-modal association on anonymized behavioral data. Xiao Zhang's anonymized behavioral data included multimodal data such as text-based work reports, image-based project charts, and audio-based meeting transcripts. The graph attention network processed this multimodal data, performing association analysis on key information in the text, data features in the images, and key content in the audio. Simultaneously, differential privacy noise (e.g., setting ε = 0.4) was incorporated to further protect data privacy, resulting in a multi-dimensional joint semantic vector. For example, when analyzing Xiao Zhang's project reports, the graph attention network associated the descriptions of project goals in the documents, the project progress data shown in the charts, and the key tasks emphasized in the meeting transcripts, generating a multi-dimensional joint semantic vector that comprehensively reflects Xiao Zhang's thinking and behavioral characteristics in the project.

[0108] Based on anonymization processing (e.g., implementing k-anonymization, k=50), the multi-dimensional joint semantic vector is synchronized to the cloud. When synchronizing the multi-dimensional joint semantic vector locally and in the cloud, only the dimension adjustment parameters are uploaded. After anonymization processing locally, Xiao Zhang's multi-dimensional joint semantic vector removes information that could directly identify him. Then, the system only uploads the dimension adjustment parameters to the cloud, such as the adjustment values ​​of the weights of each dimension in the five-dimensional weight matrix, without transmitting the original multi-dimensional joint semantic vector, thus reducing data transmission volume and further protecting privacy.

[0109] In the cloud, the FedAvg algorithm is used to aggregate and update parameters for multi-dimensional joint semantic vectors. The cloud receives dimension adjustment parameters from numerous users like Xiao Zhang, and the FedAvg algorithm aggregates and calculates these parameters. For example, the cloud collects dimension adjustment parameters from multiple professionals in the same industry, reflecting their different characteristics and trends in their work. By integrating these parameters using the FedAvg algorithm, the model parameters in the cloud are updated, enabling the model to better adapt to different users' situations and improving its versatility and accuracy.

[0110] A parameter mapping gateway is deployed in the cloud to convert multi-dimensional joint semantic vectors into a format suitable for large models and then connect them to a pre-trained large model. After processing in the cloud, Xiao Zhang's multi-dimensional joint semantic vectors are converted into a format that large models, such as GPT4, can understand and process. For example, multi-dimensional joint semantic vectors related to Xiao Zhang's thinking and behavioral characteristics at work are converted into vector representations that meet the input requirements of large models and then connected to them. Based on this input information, and combined with its own knowledge and algorithms, the large model generates more accurate and tailored decision-making suggestions for Xiao Zhang, helping him make better decisions in the workplace.

[0111] In this embodiment of the invention, the multi-level retrieval enhancement based on the dynamic prompt words, the graph database, and the vector storage database, combined with the pre-trained large model to generate a final decision suggestion as a personalized thinking imitation result for the target user, can be implemented through the following example.

[0112] Based on the dynamic prompt words, a preset number of first-level multi-dimensional joint vectors are obtained from the vector storage database through Faiss vector retrieval.

[0113] HNSW path credibility is calculated based on the five-dimensional knowledge graph stored in the graph database, and second-level multi-dimensional joint vectors are selected from the first-level multi-dimensional joint vectors.

[0114] The second-level multi-dimensional joint vector is weighted based on the user profile correlation determined by the historical adoption rate to determine the third-level multi-dimensional joint vector;

[0115] Based on the dynamic prompts and the three-level multi-dimensional joint vectors, the pre-trained large model is executed to generate a final decision suggestion as a personalized thinking imitation result for the target user.

[0116] In this embodiment of the invention, the example is still Xiao Zhang, a working professional.

[0117] The server receives dynamically generated prompts for Xiao Zhang, such as "cognitive weight 0.8 & emotional weight 0.6, generation rules: providing 4 logical comparison schemes, incorporating 3 historical similar cases, and limiting emotional intensity to Level 2." Based on these prompts, the server uses Faiss vector retrieval from a vector storage database (such as the Milvus vector database) to find multi-dimensional joint vectors related to Xiao Zhang's current work scenario or problem. Assuming a preset quantity of 50, the server retrieves 50 first-level multi-dimensional joint vectors with high similarity to the cognitive, emotional, and other dimensional features involved in the dynamic prompts. For example, if Xiao Zhang is planning a marketing campaign for a new product, these first-level multi-dimensional joint vectors might include multimodal feature vectors from similar product promotion cases in the past, such as text feature vectors from relevant documents, image feature vectors from the planning process, and voice feature vectors from discussion meetings.

[0118] Next, the server performs HNSW path credibility calculation based on Xiao Zhang's five-dimensional knowledge graph stored in the graph database (Neo4j). The five-dimensional knowledge graph records Xiao Zhang's various characteristics and interrelationships in the cognitive, behavioral, emotional, social, and evolutionary dimensions. For the 50 first-level multi-dimensional joint vectors obtained above, the server analyzes their association paths with nodes and edges in Xiao Zhang's five-dimensional knowledge graph. For example, in the case corresponding to a certain first-level multi-dimensional joint vector, is the thinking pattern involved in its decision-making process similar to Xiao Zhang's thinking chain in the cognitive dimension, and does the behavior conform to Xiao Zhang's value choice entropy and social influence radius characteristics in the behavioral dimension? Through HNSW path credibility calculation, second-level multi-dimensional joint vectors with higher credibility are selected, assuming that 20 remain after selection.

[0119] The server weights the second-level multi-dimensional joint vector based on the user profile relevance determined by historical adoption rates. In the past, when Xiao Zhang used the system-generated decision suggestions, the adoption rate varied across different dimensions. The system constructs user profile relevance based on these historical adoption rates. For example, Xiao Zhang had a high adoption rate (70%) for logical analysis suggestions based on the cognitive dimension, but a low adoption rate (30%) for emotional expression suggestions based on the emotional dimension. Therefore, when weighting the second-level multi-dimensional joint vector, vectors closely related to the cognitive dimension are given higher weights, and vice versa. After weighting, the third-level multi-dimensional joint vector is determined.

[0120] Finally, the server executes a pre-trained large model (such as GPT4) based on dynamic prompts and a three-level multi-dimensional joint vector. The large model combines the specific requirements of the dynamic prompts, such as providing logical comparison schemes and incorporating historical similar cases, with the relevant feature information and similar case information of Xiao Zhang contained in the three-level multi-dimensional joint vector to generate final decision suggestions. For example, the large model might provide Xiao Zhang with a detailed logical comparison scheme for new product marketing planning, including cost-benefit analysis of different promotion channels, precise positioning strategies for the target customer group, etc., and incorporate the experience of three historically similar and successful promotion cases selected from the three-level multi-dimensional joint vector, while controlling the sentiment intensity at Level 2 to meet the requirements of the dynamic prompts. These decision suggestions, as the result of imitating Xiao Zhang's personal thinking, assist Xiao Zhang in making more scientific and reasonable marketing decisions.

[0121] In this embodiment of the invention, the construction of the five-dimensional knowledge graph adopts an incremental learning mechanism, which initiates subgraph updates at preset periodic nodes, processes data changes within a preset effective data range, and triggers full graph rebalancing when the impact of the change exceeds a preset impact threshold.

[0122] The five-dimensional knowledge graph performs exponential decay on inactive nodes for a preset duration, and archives them to the historical database when the confidence level is lower than the preset decay confidence level.

[0123] In this embodiment of the invention, let's take Zhang, a working professional, as an example. The five-dimensional knowledge graph constructed for Zhang by the system adopts an incremental learning mechanism, with the preset cycle node set at midnight every day. Every day, Zhang generates new work-related data, such as his chat history in work groups, newly completed project documents, and records of online training courses he participated in. This data is reflected in different dimensions of the five-dimensional knowledge graph. At midnight every day, the server starts the subgraph update program to process data changes within the preset data validity range. Assuming the preset data validity range is the most recent 24 hours, the server will filter out the new data generated by Zhang in the past 24 hours. For example, Zhang discussed a new project plan with team members yesterday, and his remarks and opinions involved the cognitive dimension (demonstrating thinking patterns and decision-making logic) and the social dimension (interactions with team members). The server adds the nodes and edges corresponding to this new data to the corresponding subgraphs of the five-dimensional knowledge graph. At the same time, the server calculates the impact of the data changes. If Xiao Zhang proposed a completely new project execution approach in yesterday's discussion, significantly different from his previous thinking patterns, this might have a substantial impact on the relevant nodes and edges in the cognitive dimension. The system will assess the impact of this change. When the impact exceeds a preset threshold (assumed to be 0.3), a full graph rebalancing is triggered. At this time, the server will recalculate the relationships between nodes in each dimension of the five-dimensional knowledge graph, adjust edge weights, etc., to ensure that the knowledge graph accurately reflects Xiao Zhang's latest state and characteristics. The five-dimensional knowledge graph also handles inactive nodes for a preset duration. The preset duration is set to 6 months, meaning that if a node has not been accessed or updated within 6 months, it is considered an inactive node. For example, Xiao Zhang participated in a small project six months ago, and at that time, some of his behavioral characteristic nodes in the behavioral dimension of the five-dimensional knowledge graph were recorded. However, due to a shift in his work focus, he no longer involved in related content, and the node is in an inactive state. The system will exponentially decay such inactive nodes, calculated using the formula: confidence = initial value × 0.9t (t is time, in months). As time progresses, the confidence level of this node continuously decreases. When the confidence level falls below a preset decay confidence level (assumed to be 0.2), the system archives the node to the historical database. This avoids accumulating too much invalid or outdated information in the five-dimensional knowledge graph, improves the storage and retrieval efficiency of the knowledge graph, and also ensures that the current knowledge graph focuses on Xiao Zhang's recent main behaviors and characteristics, so as to provide him with more accurate decision-making suggestions and personal thinking imitation.

[0124] In this embodiment of the invention, the graph database uses Neo4j to store the five-dimensional knowledge graph. The nodes include decision cases, psychological characteristics, environmental context, timestamps and confidence labels, and the edges define cross-dimensional influence coefficients.

[0125] The vector storage database uses Milvus to store the multi-dimensional joint semantic vectors and is optimized by the Faiss engine. When the node similarity exceeds a preset similarity threshold, association suggestions are triggered.

[0126] In this embodiment of the invention, an example is taken as a working professional named Xiao Zhang.

[0127] The system uses Neo4j to store Xiao Zhang's five-dimensional knowledge graph. Regarding nodes: Decision Cases: Xiao Zhang has participated in many project decisions at work. For example, in a product optimization project, he proposed a decision to understand user needs through market research and then determine the direction of product feature improvements. This decision case will be stored as a node in the knowledge graph, recording detailed information such as the background, process, and results of the decision. Psychological Characteristics: Through analysis of Xiao Zhang's dynamic questionnaire feedback, daily communication, and other multi-source data, the system determines that Xiao Zhang has strong resilience under pressure and is relatively rational in decision-making. This psychological characteristic information will also be stored as a node, reflecting Xiao Zhang's intrinsic traits. Environmental Context: Xiao Zhang's company is currently in a phase of business expansion, with intense market competition. This company environment and market environment context information will be recorded as nodes in the knowledge graph, because environmental factors influence his decisions and behaviors. Timestamps and confidence labels: Each of the above nodes has a timestamp. For example, the timestamp of the product optimization project decision case records the specific time when the decision occurred. At the same time, the system will assign confidence labels to nodes based on the reliability of the data source and analysis. For example, a high confidence label will be assigned to the psychological characteristic node of Xiao Zhang's strong stress resistance, which is derived from a large amount of reliable data.

[0128] In the definition of an edge, it is used to define a cross-dimensional influence coefficient. For example, Xiao Zhang's rational thinking mode in the cognitive dimension (cognitive dimension node) will influence his decision-making behavior in the behavioral dimension (behavioral dimension node). The two are connected by an edge, and the edge defines a cross-dimensional influence coefficient to represent the degree and direction of this influence. If Xiao Zhang's rational thinking makes him more inclined to analyze data and risks when making decisions, then this influence coefficient will reflect the strength of this connection.

[0129] Milvus is used to store Xiao Zhang's multi-dimensional joint semantic vectors. Xiao Zhang's work-related data, such as project report text, meeting speech audio, and presentation images, are processed through feature extraction and cross-modal association to generate multi-dimensional joint semantic vectors, which are then stored in Milvus. The Faiss engine optimizes Milvus to improve the efficiency and accuracy of vector retrieval. Association suggestions are triggered when the node similarity exceeds a preset similarity threshold (assumed to be 0.7). For example, when Xiao Zhang is preparing a new marketing plan, the system calculates the similarity between the marketing-related multi-dimensional joint semantic vectors and vectors in the database. If a stored vector (corresponding to a previous successful marketing case) has a similarity of 0.75 with the current vector, exceeding the preset threshold, the system triggers association suggestions, displaying relevant information from the previous successful case, such as marketing strategies and market feedback, for Xiao Zhang's reference. This helps Xiao Zhang learn from past experience and make more rational decisions, while also demonstrating the important role of vector storage databases in assisting individual thinking and decision support.

[0130] To more clearly describe the solutions provided in the embodiments of the present invention, a more complete implementation method is provided below.

[0131] 1. Dimensional Reconstruction and Theoretical Deepening of Five-Dimensional Knowledge Graph Analysis Method

[0132] Background: The limitations of traditional user profiling (single-dimensional, static analysis) are addressed by combining and innovating with psychology, behavioral science, and knowledge graph technologies.

[0133] 1.1 Cognitive Dimension: Dynamic Modeling of Thinking Patterns

[0134] 1.1.1 Core Definition:

[0135] This dimension reveals users' core decision-making logic and information processing paradigms, including inductive reasoning, deductive reasoning, and cross-domain transfer capabilities. Unlike traditional cognitive models, this dimension emphasizes capturing the depth and temporal evolution of thought chains.

[0136] 1.1.2 Measurement System:

[0137] Logical chain density: calculated by the number of consecutive follow-up questions and the number of decision path branches in the log (e.g., consecutive follow-up questions ≥ 3 times and path branches ≥ 2 times are considered deep cognition);

[0138] Cross-domain association strength: based on the co-occurrence frequency and semantic similarity of cross-disciplinary nodes in the knowledge graph (e.g., a co-occurrence rate of >30% between medical and engineering concepts is considered a strong association).

[0139] 1.1.3 Data Source Innovation:

[0140] Private domain data: Mind map structure analysis (such as bidirectional link density in Obsidian Notes), e-book annotation patterns (highlighted paragraph topic clustering);

[0141] Public domain data: Academic paper citation network analysis (obtaining interdisciplinary features through SemanticScholarAPI).

[0142] 1.1.4 Dynamic Evolution Mechanism:

[0143] We employ an LSTM-GNN hybrid model to capture long-term cognitive leaps (such as the phased leaps from experience-driven to data-driven to systems thinking).

[0144] 1.2 Behavioral Dimension: Privacy-Friendly Behavioral Modeling

[0145] To address privacy protection needs, a dual-track analysis system based on selection preferences and public domain footprints is established, breaking through the reliance on traditional device sensors.

[0146] 1.2.1 Innovative Measurement Indicators:

[0147] Value choice entropy: Quantifying behavioral inertia through virtual scenario questionnaires (such as "choosing efficiency priority / quality priority when time conflicts occur");

[0148] Social influence radius: calculated based on the forwarding level of social networks (such as WeChat Moments) and the connection strength of professional networks (such as LinkedIn) (three levels of forwarding depth + 50% strong connection ratio is considered as high influence).

[0149] 1.2.2 Data Acquisition Strategy:

[0150] Enhanced questionnaire design: embedding dynamic contextual multiple-choice questions (such as popping up in real time during AI training, "Do you want the system to focus more on logical rigor / emotional resonance?");

[0151] Public domain footprint analysis: Obtain the topic participation matrix through public domain APIs (such as Weibo, Xiaohongshu, Twitter) (original content ratio > 40% is considered as active output behavior).

[0152] 1.2.3 Conflict Resolution Mechanism:

[0153] Develop a cognitive-behavioral consistency verification algorithm that triggers a dual verification process when the deviation between the questionnaire statement and public domain behavior exceeds 25%.

[0154] 1.3 Emotional Dimension: Multimodal Emotional Map

[0155] 1.3.1 Theoretical Breakthrough:

[0156] By integrating knowledge graph sentiment computing technology, a three-dimensional mapping model of "emotion-value-physiological response" is constructed.

[0157] 1.3.2 Measurement Innovation:

[0158] Sentiment network entropy: Calculates the dispersion of the distribution of text sentiment word vectors in the knowledge graph (standard deviation > 0.5 is considered high volatility);

[0159] Value orientation index: Based on the choice experiment method, it quantifies self-interest / altruistic preference (the intensity of conflict between donation behavior and self-interested options is >3).

[0160] 1.3.3 Data Source Expansion:

[0161] Indirect inference from biosignals: Inferring emotional state from behavioral patterns in public domain data (e.g., frequent liking of entertainment content late at night - a tendency to release stress);

[0162] Multimodal semantic parsing: By parsing the dynamics posted by users, we analyze the emotional polarity and construct composite emotion tags.

[0163] 1.3.4 Dynamic Modeling:

[0164] Develop an emotional state transition matrix and combine it with a seasonal decomposition algorithm to identify periodic patterns (such as the upward trend of anxiety index at the end of each quarter).

[0165] 1.4 Social Dimension: Reconstruction Analysis of Relationship Networks

[0166] 1.4.1 Methodological Innovation:

[0167] A dynamic community discovery algorithm is introduced to overcome the limitations of traditional static social graphs.

[0168] 1.4.2 Core Indicators:

[0169] Role migration coefficient: Calculates the rate of change of centrality in a social network over a six-month period (e.g., the speed of the shift from information receiver to disseminator);

[0170] Cross-platform synergy: For example, assess the coupling strength between work collaboration networks and professional networks (a common node ratio > 15% is considered high synergy).

[0171] 1.4.3 Data Fusion Strategy:

[0172] Cross-platform relationship extraction: Constructing a composite social graph using social network data provided by users;

[0173] Implicit Relationship Mining: Identifying undeclared mentor-apprentice relationships, potential collaboration intentions, etc., through knowledge graph reasoning technology.

[0174] 1.4.4 Evolutionary Prediction:

[0175] We use the preference attachment model from complex network theory to predict the trend of social influence changes over the next 6 months.

[0176] 1.5 Evolutionary Dimension: Phase Transition Analysis of Cognitive Growth

[0177] By introducing complex systems theory, a cognitive attractor model is established to explain the nonlinear growth pattern.

[0178] 1.5.1 Breakthrough Indicator:

[0179] Knowledge phase transition threshold: Identifying the intensity of key stimuli that cause a sudden change in thinking patterns (such as the rate of change in the standard deviation of behavior before and after a major event > 40%);

[0180] Fitness surface: The cognitive evolution path under different environments is simulated using the Markov chain Monte Carlo method.

[0181] 1.5.2 Data Source Innovation:

[0182] Long-term digital footprint: Integrating the evolution of blog writing styles and the trajectory of knowledge-paying course selection over 5 years;

[0183] Key event markers: Identify career turning points and learning breakthroughs using calendar data.

[0184] 1.5.3 Dynamic Modeling:

[0185] Construct a cognitive adaptation landscape model to predict the direction of thought pattern evolution over the next 12-18 months.

[0186] 1.6 Five-Dimensional Collaboration Mechanism

[0187] 1.6.1 Cross-dimensional coupling matrix:

[0188] We constructed a five-dimensional correlation strength map to identify synergistic / antagonistic effects between dimensions (e.g., a high social dimension may inhibit the intensity of emotional expression, with a correlation coefficient of -0.32).

[0189] 1.6.2 Conflict Early Warning System:

[0190] When the deviation of three or more dimensions of data is greater than 20%, a multimodal validation process is triggered (such as cross-validation by combining voice sentiment analysis with questionnaire results).

[0191] 1.6.3 Dynamic weight allocation:

[0192] Design a reinforcement learning framework that automatically adjusts dimension weights and updates the formula based on user adoption rates of AI suggestions:

[0193] Python

[0194] W t+1 =αWt +β(ΔF+γΔC)#α=historical attenuation coefficient,β=feedback gain,γ=cross-dimensional coupling factor

[0195] ```

[0196] 1.7 The theoretical innovation value of the five-dimensional knowledge graph analysis method:

[0197] The pioneering "cognitive attractor" model breaks through the traditional static analysis framework of psychology;

[0198] Establish a privacy-friendly behavioral analysis paradigm to address the ethical dilemmas of device data collection;

[0199] To achieve deep coupling between knowledge graph technology and complex systems theory.

[0200] 2. Data-driven five-dimensional dynamic modeling and knowledge base construction

[0201] 2.1 Data Acquisition and Feature Engineering Optimization

[0202] 2.1.1 Multi-source data fusion strategy

[0203] 2.1.1.1 Public Domain Behavior Analysis:

[0204] Public social network data collection: Obtain publicly available user interaction data (forwards, comments, likes) through API interfaces (such as Weibo Open Platform) to construct a "behavior-value" correlation model. For example, frequent forwarding of public welfare content reflects a tendency towards social responsibility, while original science and technology-related answers reflect a preference for rational decision-making.

[0205] Cross-platform data integration: Integrate public domain data such as GitHub code submission patterns and Zhihu answer styles to form composite behavioral profiles (e.g., continuous nighttime code submissions - cognitive dimension "deep focus" feature + weight).

[0206] 2.1.1.2 Enhanced Questionnaire Design: Develop a dynamic situational choice question bank, embedding five-dimensional guided questions (such as "relying more on intuition / logic when making decisions"), and employing contrastive learning techniques to align questionnaire self-reports with public domain behavioral data. A double verification process is triggered when the deviation >25%.

[0207] 2.1.2 Privacy-Friendly Feature Extraction

[0208] Text semantic analysis: The BERT model is used for sentiment word vector analysis, combined with LDA topic model to mine implicit behavioral features in user-generated content (such as the systematic thinking tendency in Zhihu answers).

[0209] Synthetic data generation: For sensitive data (such as medical records), generative adversarial networks (GANs) are used to generate synthetic data to ensure that the distribution characteristics are consistent with the real data.

[0210] 2.2 Five-Dimensional Weight Dynamic Adjustment Algorithm

[0211] 2.2.1 Dual Feedback Driven Mechanism

[0212] Python

[0213] class DimensionOptimizer:

[0214] def__init__(self, dimensions):

[0215] self.weights = dimensions # Initial five-dimensional weights

[0216] self.history_decay = 0.85 # Historical influence decay coefficient

[0217] self.feedback_gain={

[0218] 'explicit':1.5, #User-modified

[0219] 'implicit': 0.8# Behavioral deviation prediction

[0220] }

[0221] defupdate(self,dimension,feedback_type):

[0222] delta=self.feedback_gain[feedback_type]*(1if feedback_type=='explicit'else-0.6)

[0223] new_weight=self.weights[dimension]*self.history_decay+delta

[0224] self.weights[dimension]=np.clip(new_weight,0.1,1.0)

[0225] return self.weights

[0226] ```

[0227] Explicit feedback: User actively adjusts AI-generated content (e.g., rewrites the emotional expression in prompts) - corresponding dimension weight gain of 15%.

[0228] Implicit feedback: Continuous deviation from the predicted path (e.g., the social dimension predicts "consulting friends" but the actual decision is made independently) - triggers weight decay.

[0229] 2.2.2 Cross-dimensional coupling analysis

[0230] A five-dimensional correlation matrix was constructed to identify synergistic / antagonistic effects (e.g., a high affective dimension may inhibit the cognitive dimension, with a correlation coefficient of -0.32).

[0231] Granger causality test is used to verify the direction of influence of time series (e.g., whether changes in behavioral dimensions precede changes in evolutionary dimensions).

[0232] 2.3 Knowledge Base Construction and Evolution Mechanism

[0233] 2.3.1 Multimodal RAG Architecture

[0234] Graph database storage: Neo4j is used to store a five-dimensional relationship network, where nodes contain (decision cases, psychological characteristics, and environmental context), and edges define cross-dimensional influence coefficients.

[0235] Enhanced vectorized retrieval: User's historical decisions are encoded into 768-dimensional vectors, and context-aware matching is achieved through the Faiss engine (association suggestions are triggered when similarity > 0.7).

[0236] 2.3.2 Dynamic Prompt Word Generation Engine

[0237] Five-dimensional label injection template:

[0238] ```markdown

[0239] #Cognitive weight 0.7 & Social weight 0.5

[0240] The generated suggestions must meet the following requirements:

[0241] Provide three rationally compared options (emphasizing cost-effectiveness).

[0242] Integrating two community decision-making reference cases

[0243] Limit the intensity of emotional expression to below Level 3

[0244] ```

[0245] Context-adaptive optimization: Dynamically adjust the emphasis of suggested dimensions by combining real-time environmental data (such as calendar schedule stress).

[0246] 2.3.3 Knowledge Evolution Strategies

[0247] Incremental learning mechanism: Subgraph updates are initiated at midnight every day, and only data changes in the most recent 24 hours are processed (full graph rebalancing is triggered when the impact of the change is >0.3).

[0248] Confidence-based elimination algorithm: Nodes that have not been activated for 6 months undergo exponential decay (confidence = initial value × 0.9^t), and are archived to the historical database when their confidence level falls below 0.2.

[0249] 2.4 Privacy Protection and System Performance

[0250] 2.4.1 Gradient Desensitization Framework

[0251] Differential privacy noise (ε = 0.4) is added during the feature extraction stage to ensure that individual data is not traceable.

[0252] During cloud synchronization, k-anonymization (k=50) is implemented to obfuscate the characteristics of the user group.

[0253] 2.4.2 Edge-Cloud Collaborative Computing

[0254] Deploy a lightweight GNN model locally (parameters < 80MB), only upload the dimensional adjustment parameters.

[0255] A federated learning framework is used to update the population model, avoiding the loss of original data.

[0256] 2.5 Innovative Technology Integration

[0257] 2.5.1 Cross-dimensional coupling analysis: Breaking through the limitations of traditional single-dimensional profiling, it integrates the dynamic correlation of five dimensions: cognition, behavior, emotion, society, and evolution.

[0258] 2.5.2 Synthetic Data and Federated Learning: Synthetic data is generated using GANs to solve the cold start problem, and privacy is protected by combining it with federated learning.

[0259] 2.5.3 Dynamic Knowledge Base Architecture: Supports millisecond-level retrieval of hundreds of millions of decision cases, and achieves knowledge self-evolution by combining incremental learning and confidence-based elimination.

[0260] 3. System Architecture and Data Flow Design

[0261] 3.1 System Overall Architecture

[0262] Core Architecture: Adopting a four-layer, three-ring architecture (data access layer, feature processing layer, knowledge storage layer, and application service layer) with local-cloud collaboration, it integrates edge computing modules and five-dimensional dynamic modeling capabilities to form a closed-loop data flow. Combined with multimodal fusion, dynamic knowledge evolution, and privacy protection mechanisms, it achieves three-dimensional modeling of user behavior and cognition. Please refer to [reference needed]. Figure 2 , Figure 2 This is a system framework diagram of a personal thought imitation system provided in an embodiment of the present invention.

[0263] 3.2 Detailed Explanation of Layered Architecture

[0264] 3.2.1. Data Access Layer

[0265] 3.2.1.1 Multi-source data acquisition module: Supports voice and gesture interaction and dynamic questionnaire systems, capturing five-dimensional guided behavioral data in real time (such as logical rigor preference selection). Integrates multi-scenario interactive interfaces such as chat history import, psychological testing, and document upload.

[0266] Public domain data: Obtain publicly available user interaction data (forwards / comments / original content) through the open platform APIs of social media platforms such as Weibo.

[0267] Private domain data: Connect to note-taking tools such as Obsidian / Notion, and use the BERT model to extract entity relationship triples (such as mind map nodes - cognitive dimension features) from notes.

[0268] Interactive data: An embedded dynamic questionnaire system (customized by Typeform) captures users' preference adjustment behavior in real time during AI training.

[0269] 3.2.1.2 Privacy Preprocessing Module:

[0270] Gradient desensitization (ε = 0.4) and k-anonymization (k = 50) are implemented to generalize sensitive information (such as geographic location).

[0271] 3.2.2 Feature Processing Layer

[0272] 3.2.2.1 Fractal Analysis Engine:

[0273] Three-level segmentation strategy:

[0274] Python

[0275] #Example of document parsing process

[0276] deffractal_parse(doc):

[0277] main_concept = LayoutLMv3(doc).extract_headings() # Extract the main concept

[0278] supporting_args = SpaCy_NER(doc).get_arguments() # Identify supporting arguments

[0279] atomic_knowledge = Tesseract_OCR(doc).segment() # Meta-knowledge atomic decomposition return construct_subgraph(main_concept, supporting_args, atomic_knowledge)

[0280] ```

[0281] Multimodal alignment: a spatiotemporal alignment technique that aligns text, image, and speech features through a cross-modal attention mechanism.

[0282] Dynamic feature fusion: Graph Attention Network (GAT) is used to model cross-modal associations and generate a 768-dimensional joint semantic vector.

[0283] 3.2.3 Knowledge Storage Layer

[0284] 3.2.3.1 Dual-engine storage architecture:

[0285] Graph database (Neo4j): Stores a five-dimensional relationship network (cognitive / behavioral / emotional / social / evolutionary dimensions), with nodes containing timestamps and confidence labels.

[0286] Vector database (Milvus): Stores multimodal feature vectors and supports millisecond-level similarity retrieval (optimized by the Faiss engine).

[0287] 3.2.3.2 Incremental Update Mechanism:

[0288] A temporal graph convolutional network that initiates subgraph updates daily at midnight and triggers full graph rebalancing when the impact of a change is greater than 0.3.

[0289] 3.2.4 Application Service Layer

[0290] 3.2.4.1 Dynamic suggestion generation engine:

[0291] Five-dimensional weight injection template (RAG enhancement strategy):

[0292] ```markdown

[0293] #Cognitive weight 0.8 & Emotional weight 0.6

[0294] Generation rules:

[0295] Four logical comparison schemes are provided;

[0296] Incorporating three similar historical cases;

[0297] The intensity of emotion is limited to Level 2;

[0298] ```

[0299] Federated Learning Service: Deploy a lightweight GNN model (80MB) at the edge, upload only the dimensional parameters, and update it in the cloud through the FedAvg algorithm.

[0300] 3.3 Core Data Flow:

[0301] Input phase: Multi-source data aggregation - de-identification processing - local encrypted storage.

[0302] Processing stages: Feature extraction (temporal pattern analysis, sentiment analysis) - Knowledge fusion (dynamically updating the knowledge graph).

[0303] Output and Application: Generate decision suggestions on demand - user feedback - adjust weights - iterate the knowledge base.

[0304] 3.3.1 Data Input Stage

[0305] 3.3.1.1 Multimodal data flow: Weibo API - text cleaning; user notes - OCR parsing; dynamic questionnaire - behavior coding; text cleaning results, OCR parsing results, and behavior coding results are feature aligned by the feature alignment module.

[0306] 3.3.1.2 Key Processing:

[0307] Entity relations were extracted from text data using BERT+CRF.

[0308] Image / video data were used to extract cross-modal features using the CLIP model.

[0309] 3.3.2 Knowledge Construction Stage

[0310] 3.3.2.1 Cross-dimensional correlation modeling:

[0311] A five-dimensional coupling matrix was constructed to identify synergistic / antagonistic effects between dimensions (e.g., high social dimension inhibits emotional expression, r = -0.32).

[0312] Granger causality tests were used to verify the leading influence of the behavioral dimension on the evolutionary dimension (p<0.01).

[0313] 3.3.3 Service Output Phase

[0314] 3.3.3.1 Enhanced Multi-Level Search:

[0315] Level 1: Faiss-based vector similarity retrieval (Top 50 candidates)

[0316] Level 2: Graph Neural Network Path Reliability Calculation (HNSW Algorithm Optimization)

[0317] Level 3: User profile relevance weighting (dimensions with a historical adoption rate > 60% are given priority).

[0318] 3.4 Key Technology Implementation

[0319] 3.4.1 Fractal Analytical Optimization

[0320] The document layout analysis is implemented using the LayoutLMv3 model (F1-score 0.89), which improves the parsing speed by 3 times compared to the traditional PyMuPDF solution.

[0321] 3.4.2 Dynamic Vector Space Management

[0322] Design a decay mechanism with a forgetting coefficient λ = 0.9:

[0323] Python

[0324] defupdate_vector(old_vec,new_vec,λ=0.9):

[0325] returnλ*old_vec+(1-λ)*new_vec

[0326] ```

[0327] Supports the preservation of the timeliness of historical decision-making cases (half-life of 6 months).

[0328] 3.4.3 Cross-Domain Mapping Gateway

[0329] Five-stage alignment strategy:

[0330] 1. Lexical level: FastText cross-training;

[0331] 2. Entity level: TransEdge algorithm;

[0332] 3. Attribute-level: Graph Attention Network;

[0333] 4. Instance-level: Adversarial training;

[0334] 5. Rule-level: SWRL reasoning;

[0335] 4. User interaction flow design: based on multimodal data input and dynamic feedback mechanism

[0336] User local data - encryption and cleaning - five-dimensional analysis - generation of RAG knowledge base - cloud synchronization - injection of exclusive prompt words when calling large models (such as GPT-4)

[0337] 4.1 Interaction Flow Framework

[0338] This system adopts a "five-stage closed-loop interaction model," covering the entire process from data input to obtaining decision-making suggestions. Please refer to the relevant documentation. Figure 3 , Figure 3 This is an interactive diagram of the personal thinking imitation system provided in an embodiment of the present invention, and the operation path is optimized by combining the characteristics of a dynamic knowledge base.

[0339] 4.2 Detailed Explanation of the Core Interaction Phase

[0340] 4.2.1 Data Authorization and Input

[0341] Multimodal access channel:

[0342] Public domain data: Automatically scraped from Weibo / Zhihu APIs (requires user authorization via OAuth 2.0 protocol);

[0343] Private domain data: Supports importing Notion / Obsidian notes (using document parsing technology);

[0344] Interactive data: Embedded dynamic questionnaire (Customized Typeform, supports voice / gesture input);

[0345] Privacy Control Panel:

[0346] It provides 23 fine-grained permission options (such as "Analyze only forwarded content" and "Do not parse like records");

[0347] The authorization period is 7 days by default, and a renewal reminder will be triggered 3 days before the expiration date.

[0348] 4.2.2 Dynamic Calibration and Training

[0349] 4.2.2.1 Deviation Detection Mechanism:

[0350] Python

[0351] #Alignment algorithm between questionnaire and behavioral data

[0352] defalignment_check(questionnaire,behavior_data):

[0353] cosine_sim=cosine_similarity(

[0354] bert_embed(questionnaire),

[0355] lda_embed(behavior_data) )

[0357] ifcosine_sim<0.75:# Deviation>25%

[0358] trigger_double_validation()

[0359] ```

[0360] 4.2.2.2 AI Training Sandbox:

[0361] Provides a visual parameter adjustment interface (cognitive / emotional / social dimension sliders).

[0362] Support for playback and annotation of historical decision-making cases (feedback enhancement strategy)

[0363] 4.2.3 Decision Generation and Optimization

[0364] 4.2.3.1 Multi-level suggestion generation, please refer to Table 1:

[0365] Table 1

[0366]

[0367] 4.2.3.2 Real-time tuning interface:

[0368] Supports one-click optimization such as "logic enhancement" and "emotional weakening". Provides visual traceability of decision-making paths (Neo4j relationship graph rendering).

[0369] 4.3 Anomaly Handling and Feedback Mechanism

[0370] 4.3.1 Fault-tolerant design

[0371] 4.3.1.1 Input error:

[0372] Unstructured document parsing failure triggers OCR enhancement mode (Tesseract+LayoutLMv3); sensor data conflict initiates multimodal voting mechanism.

[0373] 4.3.1.2 Logical Exceptions:

[0374] Five-dimensional weight contradictions activate Granger causality test (reconstruct the correlation matrix when p<0.01).

[0375] 4.3.2 Progressive Feedback

[0376] 4.3.2.1 Immediate Feedback:

[0377] Operation confirmed (micro-vibration + visual highlight).

[0378] Processing progress (circular progress bar + estimated remaining time).

[0379] 4.3.2.2 Delayed Feedback:

[0380] Daily behavior analysis report (PDF + interactive dashboard).

[0381] Monthly Cognitive Evolution Map (Dynamic GNN Visualization)

[0382] 4.4 Key Technology Innovation

[0383] 4.4.1 Context-aware process jumps.

[0384] Identify decision urgency through calendar events:

[0385] Python

[0386] ifschedule_tension>0.8: # Deadline 3 days prior

[0387] bypass_calibration() # Skip the questionnaire calibration phase

[0388] activate_efficiency_mode() # Enables efficiency priority mode

[0389] ```

[0390] Optimize the interaction path based on the time pressure model.

[0391] 4.4.2 Interactive Synchronization in a Federated Learning Environment

[0392] Edge-cloud collaborative update mechanism:

[0393] Save the last 7 days of interaction records locally (lightweight SQLite)

[0394] Gradient obfuscation (differential privacy scheme) is implemented during cloud synchronization.

[0395] Through the above technical solutions, this invention achieves accurate imitation of individual thinking and efficient decision-making assistance, while simultaneously balancing rapid local response with powerful cloud computing capabilities. By integrating edge computing, blockchain, distributed machine learning, encryption technology, and AI and large AI model technologies, the system not only improves performance and security but also enhances scalability and user-friendliness.

[0396] Please refer to the following: Figure 4 , Figure 4 A schematic block diagram of a personal thought imitation system 110 provided in an embodiment of the present invention includes:

[0397] The acquisition module 1101 is used to collect multi-source data from target users in public domains, private domains, and dynamic questionnaires, and perform privacy preprocessing to obtain de-identified behavioral data; a three-level segmentation strategy is used to extract features from the de-identified behavioral data, and cross-modal association is performed through a graph attention network to obtain multi-dimensional joint semantic vectors; a graph database and a vector storage database are constructed based on the multi-dimensional joint semantic vectors, and the graph database is used to store a five-dimensional knowledge graph, which includes cognitive, behavioral, emotional, social, and evolutionary dimensions; dynamic prompt words are constructed based on the five-dimensional knowledge graph, a preset five-dimensional weight injection template, and real-time environmental data;

[0398] The execution module 1102 is used to perform multi-level retrieval enhancement based on the dynamic prompt words, the graph database and the vector storage database, and combine it with a pre-trained large model to generate a final decision suggestion as a personalized thinking imitation result for the target user.

[0399] It should be noted that the implementation principle of the aforementioned personal thought imitation system 110 can refer to the implementation principle of the aforementioned personal thought imitation method, and will not be repeated here. It should be understood that the division of the various modules of the above device is only a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated.

[0400] This invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned personal thought imitation system 110. For example... Figure 5 As shown, Figure 5 This is a structural block diagram of a computer device 100 provided in an embodiment of the present invention. The computer device 100 includes a personal thought imitation system 110, a memory 111, a processor 112, and a communication unit 113.

[0401] To enable data transmission or interaction, the memory 111, processor 112, and communication unit 113 are electrically connected to each other directly or indirectly. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The personal thought imitation system 110 includes at least one software function module that can be stored in the memory 111 or embedded in the operating system (OS) of the computer device 100 in the form of software or firmware. The processor 112 is used to execute the personal thought imitation system 110 stored in the memory 111, such as the software function modules and computer programs included in the personal thought imitation system 110.

[0402] This invention provides a readable storage medium, which includes a computer program. When the computer program runs, it controls the computer device where the readable storage medium is located to execute the aforementioned personal thought imitation system 110.

[0403] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the foregoing illustrative discussions are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in accordance with the foregoing teachings. These embodiments were chosen and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the disclosure and to employ various embodiments with different modifications to suit a particular intended application.

Claims

1. A method for imitating personal thinking, characterized in that, include: Multi-source data collection and privacy preprocessing were performed on target users from public domains, private domains, and dynamic questionnaires to obtain de-identified behavioral data. A three-level segmentation strategy is used to extract features from the desensitized behavior data, and cross-modal association is performed through a graph attention network to obtain a multi-dimensional joint semantic vector; Based on the multi-dimensional joint semantic vectors, a graph database and a vector storage database are constructed. The graph database is used to store a five-dimensional knowledge graph, which includes cognitive, behavioral, emotional, social, and evolutionary dimensions. Dynamic prompt words are constructed based on the aforementioned five-dimensional knowledge graph, the preset five-dimensional weight injection template, and real-time environmental data; Multi-level retrieval enhancement is performed based on the dynamic prompts, the graph database, and the vector storage database, and a final decision suggestion is generated by combining a pre-trained large model as a personalized thinking imitation result for the target user. The three-level segmentation strategy is used to extract features from the desensitized behavior data, and cross-modal association is performed through a graph attention network to obtain a multi-dimensional joint semantic vector, including: The de-identified behavior data is sequentially processed by extracting the main concepts based on LayoutLMv3, identifying supporting arguments based on SpaCy_NER, and performing meta-knowledge atomic decomposition based on Tesseract_OCR to obtain multiple features to be processed based on the three-level segmentation strategy. The multiple features to be processed are adjusted and aligned using a cross-modal attention mechanism; The graph attention network is used to perform cross-modal association on multiple aligned features to obtain the multi-dimensional joint semantic vector. The cross-modal association includes lexical association based on FastText cross-training, entity association based on the TransEdge algorithm, attribute association based on graph attention network, instance association based on adversarial training, and rule association based on SWRL inference.

2. The method according to claim 1, characterized in that, The process involves collecting multi-source data from target users in public domains, private domains, and dynamic questionnaires, followed by privacy preprocessing to obtain anonymized behavioral data, including: Public interaction data of the target user is obtained through the open platform API in the public domain; By using a mind map-based structure parsing strategy in the private domain, the private data of the target user can be obtained; By outputting a dynamic questionnaire with embedded dynamic situational multiple-choice questions, the questionnaire feedback of the target users is obtained; The publicly available interactive data, the private data, and the questionnaire feedback are used as the multi-source data, and gradient desensitization and anonymization are performed to obtain the desensitized behavioral data.

3. The method according to claim 1, characterized in that, The method further includes: When the data deviation in at least three of the five dimensions of the knowledge graph—cognitive, behavioral, emotional, social, and evolutionary dimensions—exceeds a preset deviation threshold, a multimodal verification process is triggered, and the five-dimensional knowledge graph is adjusted based on the verification results. The cognitive dimension employs an LSTM-GNN hybrid model to capture long-term cognitive transition trajectories; the behavioral dimension develops a cognitive-behavioral consistency verification algorithm, triggering a dual verification process when the deviation between the questionnaire statement and public domain behavior exceeds a preset deviation threshold; the emotional dimension develops an emotional state transition matrix, combined with a seasonal decomposition algorithm to identify periodic patterns; the social dimension introduces a dynamic community discovery algorithm and applies the preference attachment model from complex network theory to predict the trend of social influence changes within a preset time range; the evolutionary dimension introduces complex systems theory, establishes a cognitive attractor model to explain nonlinear growth patterns, and simulates cognitive evolution paths under different environments using the Markov chain Monte Carlo method.

4. The method according to claim 1, characterized in that, The method further includes: The five-dimensional knowledge graph dynamically adjusts the weights of the cognitive, behavioral, emotional, social, and evolutionary dimensions through a five-dimensional weight matrix, and dynamically adjusts the emphasis of suggested dimensions in conjunction with real-time environmental data. The five-dimensional weight matrix is ​​obtained through the formula: W t+1 =αW t +β(ΔF+γΔC), to achieve dynamic adjustment, where W t+1 For the adjusted weights, W t The weights before adjustment are α, the historical decay coefficient is β, the feedback gain is β, the cross-dimensional coupling factor is γ, the feedback weight adjustment factor is ΔF, and the dimension weight adjustment factor is ΔC.

5. The method according to claim 1, characterized in that, The method further includes: The de-identified behavior data is obtained locally using the AES-256 encryption algorithm; The graph attention network is deployed locally to perform cross-modal association and combined with differential privacy noise to obtain a multi-dimensional joint semantic vector; The multi-dimensional joint semantic vector is synchronized to the cloud based on anonymization processing; wherein, when synchronizing the multi-dimensional joint semantic vector between the local end and the cloud, only the dimension adjustment parameters are uploaded. The parameters of the multi-dimensional joint semantic vector are aggregated and updated in the cloud using the FedAvg algorithm; Deploy a parameter mapping gateway in the cloud to convert the multi-dimensional joint semantic vector into a large model adaptation format and connect it to the pre-trained large model.

6. The method according to claim 1, characterized in that, The process of performing multi-level retrieval enhancement based on the dynamic prompts, the graph database, and the vector storage database, and combining this with a pre-trained large model to generate a final decision suggestion as a personalized thought imitation result for the target user, includes: Based on the dynamic prompt words, a preset number of first-level multi-dimensional joint vectors are obtained from the vector storage database through Faiss vector retrieval. HNSW path credibility is calculated based on the five-dimensional knowledge graph stored in the graph database, and second-level multi-dimensional joint vectors are selected from the first-level multi-dimensional joint vectors. The second-level multi-dimensional joint vector is weighted based on the user profile correlation determined by the historical adoption rate to determine the third-level multi-dimensional joint vector; Based on the dynamic prompts and the three-level multi-dimensional joint vectors, the pre-trained large model is executed to generate a final decision suggestion as a personalized thinking imitation result for the target user.

7. The method according to claim 1, characterized in that, The construction of the five-dimensional knowledge graph adopts an incremental learning mechanism, which initiates subgraph updates at preset periodic nodes, processes data changes within a preset effective data range, and triggers full graph rebalancing when the impact of the change exceeds a preset impact threshold. The five-dimensional knowledge graph performs exponential decay on inactive nodes for a preset duration, and archives them to the historical database when the confidence level is lower than the preset decay confidence level.

8. The method according to claim 1, characterized in that, The graph database uses Neo4j to store the five-dimensional knowledge graph. Nodes include decision cases, psychological characteristics, environmental context, timestamps, and confidence labels, while edges define cross-dimensional influence coefficients. The vector storage database uses Milvus to store the multi-dimensional joint semantic vectors and is optimized by the Faiss engine. When the node similarity exceeds a preset similarity threshold, association suggestions are triggered.

9. A personal thought imitation system, characterized in that, include: The acquisition module is used to collect multi-source data from target users in public domains, private domains, and dynamic questionnaires, and to perform privacy preprocessing to obtain de-identified behavioral data. A three-level segmentation strategy is used to extract features from the desensitized behavioral data, and cross-modal association is performed through a graph attention network to obtain multi-dimensional joint semantic vectors. Based on the multi-dimensional joint semantic vectors, a graph database and a vector storage database are constructed. The graph database is used to store a five-dimensional knowledge graph, which includes cognitive, behavioral, emotional, social, and evolutionary dimensions. Dynamic prompt words are constructed based on the five-dimensional knowledge graph, a preset five-dimensional weight injection template, and real-time environmental data. The execution module is used to perform multi-level retrieval enhancement based on the dynamic prompt words, the graph database and the vector storage database, and combine it with a pre-trained large model to generate a final decision suggestion as a personalized thinking imitation result for the target user; The acquisition module is specifically used for: The de-identified behavior data is sequentially processed by extracting the main concepts based on LayoutLMv3, identifying supporting arguments based on SpaCy_NER, and performing meta-knowledge atomic decomposition based on Tesseract_OCR to obtain multiple features to be processed based on the three-level segmentation strategy. The multiple features to be processed are adjusted and aligned using a cross-modal attention mechanism; The graph attention network is used to perform cross-modal association on multiple aligned features to obtain the multi-dimensional joint semantic vector. The cross-modal association includes lexical association based on FastText cross-training, entity association based on the TransEdge algorithm, attribute association based on graph attention network, instance association based on adversarial training, and rule association based on SWRL inference.

Images