A multi-dimensional data fusion-based human-machine collaboration intention evaluation method and system

Abstract

The application discloses a kind of man-machine cooperation intention evaluation method and system based on multidimensional data fusion, it is related to man-machine interaction and artificial intelligence data processing field.The present collaborative system lacks the problem of implicit flow perception and original contribution quantification of user, the present application proposes: collect the result snapshot data and process log data of user in content arrangement space;Perform content pedigree tracing and ownership evolution algorithm, solve text evolution source and output attribute tracing quantitative index;Comprehensive operation behavior weight and instruction semantic feature, generate explicit canvas feature evaluation index and implicit collaborative state feature data;Based on evaluation index rendering generation space cooperation heat map, combined with implicit feature data output mapping current collaboration mode behavior clustering portrait label, and adaptive interaction intervention mechanism including cascade early warning and permission release can be triggered accordingly.The present application realizes closed loop from bottom interaction silent perception to high-order cognitive state adaptive guidance, provides objective structured base for accurately describing and archiving man-machine collaborative efficiency.

Description

Technical Field

[0001] This invention belongs to the fields of computer human-computer interaction, artificial intelligence applications and data processing technology, and specifically relates to a method, system and computer-readable storage medium for evaluating human-computer collaborative intent based on multi-dimensional data fusion. Background Technology

[0002] With the deepening application of generative artificial intelligence (AIGC) and large language model (LLM), human-computer collaborative creation has become the core mode for high-level knowledge workers to produce content and reconstruct logic. However, existing AI-based assisted creation systems and platforms generally remain at the stage of one-way passive response of "prompt-generation" or simple text completion.

[0003] In real-world scenarios involving complex knowledge construction and long-text creation, traditional collaborative systems exhibit the following significant technical shortcomings: First, existing systems can only passively receive explicit user interaction commands, lacking the ability to silently perceive and process implicit user flow states. Traditional systems often rely on coarse time monitoring (such as reading time or input pauses) and cannot accurately capture and analyze purely procedural operational behaviors such as node topology reorganization, text morphological modification, and high-frequency command interactions. This results in the system's inability to accurately and efficiently determine the user's current cognitive load level and mental activity. Second, existing content editors lack underlying data lineage and evolutionary genealogy tracking mechanisms, making it difficult to accurately quantify the proportion of original human input in the generated corpus after multiple rounds of user deletion, morphological modification, and spatial reorganization. Consequently, it is impossible to objectively define the actual mental contribution of both humans and machines in the collaborative creation process.

[0004] Furthermore, due to the lack of fine-grained data monitoring based on the dual-track integration of "process operation - result status" and a scientific behavioral clustering evaluation system, existing systems struggle to dynamically profile users' collaborative intentions and work patterns (such as highly dominant or dependent / stagnant types). They also cannot provide intuitive visual feedback (such as heatmaps) to the front end regarding the current collaborative health and algorithm node performance. This superficial understanding of user operational status and the absence of an evaluation mechanism significantly limit the system's potential to provide personalized assistance and severely hinder further improvements in human-computer collaborative creation efficiency and the quality of cognitive accumulation. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a method and system for evaluating human-computer collaboration intent based on multi-dimensional data fusion, addressing the problems of existing collaborative assistance systems lacking dual-track silent monitoring of process operations and result topology status, difficulty in quantifying the original ownership of node content, and inability to objectively quantify and visualize the implicit cognitive state of users.

[0006] To address the aforementioned technical problems, this invention provides a method for evaluating human-computer collaborative intent based on multi-dimensional data fusion, applied to a human-computer collaborative interaction system. The method includes: Real-time collection of multi-dimensional collaborative data of users in the content arrangement and interaction environment; the multi-dimensional collaborative data is divided into result snapshot data containing node physical attributes and topology structure, and process log data recording the time sequence logic of user interaction operations; Based on the process log data and result snapshot data, the content genealogy tracing and ownership evolution algorithm is executed to dynamically calculate the text evolution source flow of each content node in the content canvas, and output the source tracing quantitative index that represents the relative proportion weight of human native input, artificial intelligence fully generated and human-machine hybrid enhanced attributes. Throughout the entire human-computer collaborative interaction process, the multi-dimensional collaborative data and traceability quantitative indicators are continuously input into the collaborative efficiency evaluation model. By integrating the operational behavior weight logic, the semantic parsing features of human-computer interaction instructions, and the cumulative ratio of human-computer text output, the model calculates and generates explicit canvas feature evaluation indicators visible to the user, as well as implicit collaborative state feature data used to characterize the user's current implicit cognitive state. Visual color mapping is performed based on the explicit canvas feature evaluation index to generate a spatial collaboration heatmap; at the same time, the implicit collaboration state feature data is input into a preset template matching and behavior clustering algorithm to output behavior clustering profile labels that map the current collaboration mode, and drive the visualization state update and data persistence storage of the human-computer collaborative interaction interface.

[0007] Furthermore, the multidimensional collaboration data includes multiple feature subsets representing the explicit and implicit collaboration states of users, specifically including: The resulting snapshot data includes: the absolute coordinates of each node in the spatial canvas, the number of characters contained in the node, the identity attribution attribute label, and the directed graph topology connection state matrix composed of nodes and connection lines. The procedural log data includes the following low-level silent records of operation categories: topology-level spatial operation events for nodes, text-level morphological modification events for nodes, and instruction interaction frequency and instruction semantic evolution paths for artificial intelligence computing power or interactive systems.

[0008] Furthermore, the evaluation process of executing content genealogy tracing, solving evaluation indicators, and outputting profile tags specifically includes: Capture the lifecycle start trigger source of any content node in the content canvas space to assign an initial attribution attribute, dynamically monitor node reorganization and text modification behavior during the interaction process; based on the accumulated action type weight and text editing distance difference, drive the node attribute to undergo a relative magnitude smooth drift between pure human native input, fully generated by artificial intelligence and human-machine hybrid enhanced state, and output the source tracing quantitative index. For unstructured operation flow, calculate nonlinear operation integrals and instruction interaction semantic analysis weights, and solve for explicit canvas feature evaluation indicators that include at least the information clipping ratio and the proportion of native canvas content, as well as implicit collaborative state calculation matrices that characterize user cognitive load level, divergent thinking activity, algorithm dependence and cognitive overload dissipation state. The implicit collaborative state calculation matrix is ​​matched with a preset multi-type typical behavior clustering model. When the matching degree exceeds a preset threshold, the corresponding behavior clustering profile label is extracted and output for explicit display and archiving.

[0009] Furthermore, the method also includes a closed-loop intervention mechanism based on the implicit collaborative state: the process log data synchronously monitors the intervention event stream initiated by the system, and the system dynamically triggers the interface's early warning prompts, interaction restriction locks, or permission releases based on the real-time integral dissipation state.

[0010] To address the aforementioned technical problems, this invention also provides a human-computer collaboration intent evaluation system based on multi-dimensional data fusion, the system comprising: The multidimensional data monitoring module is used to collect result-oriented snapshot data based on the spatial topology environment in parallel, and process log data that records various user interaction events; The content genealogy tracing module is used to calculate the dynamic evolution and attribution of content node attributes in the human, artificial intelligence and human-machine augmented states based on behavioral weights and editing distance differences. A comprehensive evaluation engine is used to integrate behavioral operations, semantic parsing, and text output to calculate explicit canvas evaluation metrics and implicit collaborative feature data. The visualization and profiling clustering module is used to generate collaborative heatmaps representing the density of logical interactions based on evaluation data, and output corresponding behavioral clustering profile labels.

[0011] To address the aforementioned technical problems, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0012] The beneficial effects achieved by this invention are mainly reflected in: It achieves real-time interactive evaluation throughout the entire process: breaking the technical deadlock of traditional evaluation software that can only perform static post-audit after creation, and deeply integrating the evaluation mechanism into the dynamic full lifecycle of human-computer interaction. Through real-time streaming input of process and result collaborative data, it realizes the instant capture and perception of the creator's high-dimensional cognitive flow and workload dynamics; A frictionless lineage genetic evolutionary tree was constructed: without requiring active human declaration, the system drives the smooth evolution of node ownership across multiple levels by discretely accumulating the edit distance and action weights of text morphological and topological modification events. This not only accurately defines the true proportion of original intellectual contribution but also effectively intercepts data forgery and evasion attempts to fake originality through mechanical manipulation and local fine-tuning. A probe feedback network with behavioral feedback capability was designed: it innovatively and seamlessly integrates the system's cascading early warning, state locking, and user-initiated interaction-triggered dissipation and permission unlocking logic. The evaluation model can not only output static behavioral clustering profile labels, but also dynamically drive the interface control layer to unlock or restrict based on the rise and fall of the post-calculated matrix, completing a high-order closed-loop technology evolution from "pure monitoring" to "adaptive interactive guidance". Attached Figure Description

[0013] Figure 1 is a flowchart of the human-computer collaboration intent evaluation method based on multi-dimensional data fusion provided in an embodiment of the present invention.

[0014] Figure 2 is a block diagram of the logical architecture of the human-computer collaboration intent evaluation system provided in an embodiment of the present invention.

[0015] Figure 3 is a schematic diagram of the evolution of content genealogy tracing and ownership status provided in the embodiments of the present invention.

[0016] Figure 4 is a schematic diagram of the logic for outputting latent feature multidimensional mapping and behavior clustering profile labels provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments and accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Example 1: A dual-track collaborative data monitoring mechanism that separates process and results In a preferred embodiment of the present invention, the system deploys an independent and parallel dual-track data monitoring architecture in a content orchestration and interaction environment based on a multi-level spatial topology. This monitoring architecture does not depend on a specific front-end UI presentation but extracts data from two dimensions: underlying behavior and state snapshots. Specifically, it includes: Result-oriented snapshot data acquisition: The result-oriented snapshot data represents the explicit structural features of the current canvas state at a specific time-series cross-section, and is data that users can intuitively observe in the interface. The system does not need to perform continuous high-frequency monitoring; instead, a single global scan is triggered by specific interactive events (such as edit submission, module closure, or report generation). Specifically, the result-oriented snapshot data includes: the absolute physical coordinates of each content node within the spatial canvas, the current text count of each node, identity attribution attribute tags, and a directed graph topology connection state matrix composed of nodes and directional connections. In subsequent processes, the result-oriented snapshot data is primarily used to drive visual rendering and rendering enhancements such as spatial collaboration heatmaps. Process Log Data Collection: The process log data is an implicit dynamic record of various fine-grained operations performed by the user throughout the entire interactive creation lifecycle. It is silently output by the system's underlying layer and stored on disk as a structured log, remaining implicitly invisible in the user interface. To prevent competitors from circumventing monitoring by altering peripheral interaction methods, the system highly abstracts the process log entries into three major feature domains: event type identifier domain (used to identify the nature of the action), operation target attribute domain (used to identify the attributes of the target object), and content difference weight domain (used to record text fluctuations). The log entries specifically cover the following process behaviors: topological spatial operation events targeting nodes (including node splicing, connection, combination, detachment, disconnection, reorganization, and iterative category label changes), text-level morphological modification events targeting nodes (including cursor splitting, text swallowing and merging, micro-addition, deletion, and modification of content, and text hierarchy adjustment), and instruction interaction events during human-computer interaction (including the frequency of asking questions to artificial intelligence and the semantic structure of the questions). In particular, the process log data also tracks and records the adaptive reverse interaction intervention event flow initiated by the system at the user level throughout the entire process and in real time. The intervention event flow includes cascading early warning prompts triggered by the system, operation status locking events executed based on feature critical conditions, and dynamic interaction unblocking logs after the user is unblocked, which are used to collaboratively participate in the continuous solution of the entire process.

[0019] Example 2: Content Genealogy Tracing and Algorithm Solution Model Based on Multidimensional Fusion While the dual-track collaborative data collection continues, the system continuously inputs the collected dual-track data into the evaluation model for feature fusion analysis throughout the entire human-machine collaborative interaction process. This evaluation algorithm is calibrated based on a large amount of sample data from real user tests, and upgraded and converged into a parallel solution architecture targeting multiple subsets of explicit and implicit collaborative state features. In the underlying algorithm solution process, the system does not employ traditional static linear integral logic, but instead constructs a comprehensive solution pipeline encompassing the following three core features: Behavioral operation scoring logic and topology reorganization effort evaluation: The system assigns different topology deformation integrals based on the graph theory contribution of different operations in the process log data. For example, behaviors such as actively establishing logical connections between nodes, performing seamless spatial adsorption and linear reorganization of nodes, or actively changing classification labels to reduce visual entropy are all defined in the algorithm as structured reorganization efforts with high cognitive investment. The system then assigns behavioral operation scores accordingly and calculates the user's logical reorganization feature value. Regular expression and semantic analysis of AI communication content: The system performs text regular expression extraction and semantic feature density analysis on the instruction interaction behavior in the process log. The algorithm focuses on monitoring the density of task assignment vocabulary, the depth of unstructured questions, and the iterative drift of long sentence semantics in the instructions to determine whether the user is conducting high-order thinking inquiry or low-order tool-like command assignment; Human-computer text output accumulation and content lineage tracing: This is the core discrete state machine algorithm for determining content ownership. The system captures the lifecycle origin of each independent node and assigns it an initial ownership attribute (such as pure human input or AI-generated content). When a node undergoes splitting, merging, or micro-deletion based on cursor state, the system dynamically calculates the editing distance between the current text and the initial baseline text using Levenshtein distance. The algorithm constructs a state transition probability matrix, driving the node's attributes to undergo a smooth drift of relative magnitude between three states: human-native input, AI-generated content, and human-computer hybrid enhancement. For example, when two nodes with different attributes are merged by the user's topology, the system calculates the evolutionary attributes of the merged new node based on a preset attribute coverage strength weight table (such as the coverage strength of co-created enhanced attributes being higher than that of human-original attributes), and forcibly performs a "one-time state reset (Rebasing)" at the moment of merging, locking it as the new comparison origin, ultimately achieving content lineage tracing that cannot be evaded by simple data forgery behaviors such as copying and pasting.

[0020] Example 3: Explicit-Hidden Separation Assessment and Adaptive Feedback Closed Loop of Accompanying Behavioral Probe After continuously calculating and solving the aforementioned multidimensional feature matrix and traceability indicators throughout the entire process, the system employs a strategy of separating explicit and implicit metrics for dynamic evaluation and presentation, adaptive interface adjustment, and clustering archiving. Enhanced Explicit Data Presentation and Spatial Heatmap: The system defines explicit evaluation indicators as those that users can intuitively understand from multidimensional data. These indicators include at least the system's redundant information pruning ratio (reflecting the user's efforts to reduce redundant content generated by artificial intelligence) and the percentage of native canvas content. The system extracts the absolute coordinate parameters of nodes from the result snapshot data in real time and maps the values ​​of the explicit evaluation indicators into a two-dimensional visual matrix with color grayscale differences or transparency gradients. This matrix is ​​then rendered on the canvas layer to generate a spatial collaboration heatmap that intuitively reflects the density of logical interactions and the degree to which human cognitive sovereignty is maintained. Implicit Data Extraction and Behavioral Clustering Profile Matching: The system further extracts implicit post-calculated matrix data that is difficult for users to intuitively perceive during interaction. This data includes implicit multi-dimensional post-calculated vectors representing the user's current cognitive load level, divergent thinking activity, algorithm dependence, and cognitive overload dissipation state. The system inputs these implicit multi-dimensional post-calculated vectors into a preset template matching algorithm and compares them with multiple typical behavioral clustering models. When the matching degree exceeds a preset threshold, the system outputs behavioral clustering profile labels (such as healthy collaborative, dominant, or stagnant) mapping the current collaboration mode to the user's front end. Accompanying probe interference mechanism (optional extended closed loop): As a preferred embodiment of a further evolution of the present invention, in order to break the user's cognitive stagnation based on the output profile label, the evaluation system can also selectively output a closed-loop intervention control signal to the upstream control layer to activate the accompanying behavioral probe adaptive feedback closed loop: When the system detects that a user's procedural characteristics have reached a specific stagnation range (e.g., frequently initiating low-level tool-based AI interaction commands, zero effort integral for canvas space topology reorganization, and a persistently low proportion of native content), extreme value mutations occur in the "algorithm dependency" and "cognitive load level" of the implicit post-calculation matrix. At this point, the behavioral clustering model outputs behavioral clustering profile labels representing "command-driven" or "dependency-stagnant" behavior to the front end. Based on this, the system triggers multi-level cascading adaptive intervention feedback in real time: the first stage outputs multi-level cascading reverse warning prompts on the interface to guide users to think independently; if the indicators continue to deteriorate, the second stage is entered, triggering a state restriction lock event to temporarily lock some high-level generation permissions to prevent user cognitive overload; To overcome this limitation, the system's backend maintains a keen awareness of events: when it detects that a user performs higher-order cognitive actions on the canvas, such as establishing directional logical connections, changing category labels, or performing spatial linear reorganization, these higher-order interactive actions are injected into the implicit post-calculation matrix as positive dissipation factors, accelerating the physical-level release of cognitive overload energy (i.e., the cognitive overload release mechanism). When the accumulated cooling energy causes the cognitive load and algorithm dependence in the implicit matrix to fall below a safe threshold, the system spontaneously triggers a dynamic interaction unlocking log, revoking the restriction lock (i.e., the unlocking mechanism), thereby driving a smooth transition of behavioral clustering profile labels to profiles representing a healthy collaborative state. This achieves objective visual feedback, behavioral correction, and structured archiving of higher-order human-machine collaborative features.

Claims

1. A method for evaluating human-computer collaboration intent based on multi-dimensional data fusion, characterized in that, The method, applied to a human-computer collaborative interaction system, includes: Real-time collection of multi-dimensional collaborative data of users in the content arrangement and interaction environment; the multi-dimensional collaborative data is divided into result snapshot data containing node physical attributes and topology structure, and process log data recording the time sequence logic of user interaction operations; Based on the process log data and result snapshot data, the content genealogy tracing and ownership evolution algorithm is executed to dynamically calculate the text evolution source flow of each content node in the content canvas, and output the source tracing quantitative index that represents the relative proportion weight of human native input, artificial intelligence fully generated and human-machine hybrid enhanced attributes. Throughout the entire human-computer collaborative interaction process, the multi-dimensional collaborative data and traceability quantitative indicators are continuously input into the collaborative efficiency evaluation model. By integrating the operational behavior weight logic, the semantic parsing features of human-computer interaction instructions, and the cumulative ratio of human-computer text output, the model calculates and generates explicit canvas feature evaluation indicators visible to the user, as well as implicit collaborative state feature data used to characterize the user's current implicit cognitive state. Visual color mapping is performed based on the explicit canvas feature evaluation index to generate a spatial collaboration heatmap; at the same time, the implicit collaboration state feature data is input into a preset template matching and behavior clustering algorithm to output behavior clustering profile labels that map the current collaboration mode, and drive the visualization state update and data persistence storage of the human-computer collaborative interaction interface.

2. The method according to claim 1, characterized in that, The multidimensional collaboration data comprises multiple feature subsets representing the explicit and implicit collaboration states of users. Specifically, the multidimensional collaboration data includes: The resulting snapshot data includes: the absolute coordinates of each node in the spatial canvas, the number of characters contained in the node, the identity attribution attribute label, and the directed graph topology connection state matrix composed of nodes and connection lines. The procedural log data includes the following low-level silent records of operation categories: topology-level spatial operation events for nodes, text-level morphological modification events for nodes, and instruction interaction frequency and instruction semantic evolution paths for artificial intelligence computing power or interactive systems.

3. The method according to claim 1, characterized in that, The evaluation process of tracing the genealogy of the executed content, calculating evaluation indicators, and outputting profile tags specifically includes: Capture the lifecycle start trigger source of any content node in the content canvas space to assign an initial attribution attribute, dynamically monitor node reorganization and text modification behavior during the interaction process; based on the accumulated action type weight and text editing distance difference, drive the node attribute to undergo a relative magnitude smooth drift between pure human native input, fully generated by artificial intelligence and human-machine hybrid enhanced state, and output the source tracing quantitative index. For unstructured operation flow, calculate nonlinear operation integrals and instruction interaction semantic analysis weights, and solve for explicit canvas feature evaluation indicators that include at least the information clipping ratio and the proportion of native canvas content, as well as implicit collaborative state calculation matrices that characterize user cognitive load level, divergent thinking activity, algorithm dependence and cognitive overload dissipation state. The implicit collaborative state calculation matrix is ​​matched with a preset multi-type typical behavior clustering model. When the matching degree exceeds a preset threshold, the corresponding behavior clustering profile label is extracted and output for explicit display and archiving. Furthermore, the method also includes a closed-loop intervention mechanism based on the implicit collaborative state: the process log data synchronously monitors the intervention event stream initiated by the system, and the system dynamically triggers the interface's early warning prompts, interaction restriction locks, or permission releases based on the real-time integral dissipation state.

4. A human-computer collaboration intent evaluation system based on multi-dimensional data fusion, characterized in that, The system includes: The multidimensional data monitoring module is used to collect result-oriented snapshot data based on the spatial topology environment in parallel, and process log data that records various user interaction events; The content genealogy tracing module is used to calculate the dynamic evolution and attribution of content node attributes in the human, artificial intelligence and human-machine augmented states based on behavioral weights and editing distance differences. A comprehensive evaluation engine is used to integrate behavioral operations, semantic parsing, and text output to calculate explicit canvas evaluation metrics and implicit collaborative feature data. The visualization and profiling clustering module is used to generate collaborative heatmaps representing the density of logical interactions based on evaluation data, and output corresponding behavioral clustering profile labels.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 3.

Images