A multi-modal evaluation method for the influence of different emotional states on the conflict situation awareness ability of project managers

Abstract

The application discloses a kind of multi-modal evaluation methods of different emotional state influence project manager work conflict situation perception ability, by selecting six typical work conflicts and computer interactive interface presentation decision task, combined with standardized emotional induction scheme, make the test in positive or negative emotional state, through multi-modal data synchronous acquisition and analysis eye movement processing mode, and construct a kind of emotion-conflict bidirectional driving adaptive cross-modal fusion evaluation model, in the feature space of random forest, introduce the joint regularization operator constituted by decision reaction time penalty term and conflict sensitivity, realize the deep multi-modal nonlinear interweave of eye movement visual search strategy, operation behavior representation and subjective psychological measure.The application can objectively measure the cognitive preference and attention allocation difference of project manager to various work conflict situations under different emotional states, improve the information acquisition efficiency, judgment depth and future trend prediction ability of project manager in complex conflict situation.

Description

Technical Field

[0001] This invention belongs to the field of intelligent psychological state assessment technology, and involves the cross-disciplinary fields of intelligent project management, organizational behavior analysis and emotion computing. Specifically, it relates to a multimodal assessment method that utilizes the influence of different emotional states on a project manager's ability to perceive work conflict situations. Background Technology

[0002] In modern engineering project management practice, work conflict is a significant factor affecting project performance, teamwork, and project success rate. As the core decision-maker of the project team, the project manager's ability to perceive, understand, and predict information in conflict situations directly impacts the effectiveness of conflict resolution. However, with increasing project size, task complexity, and team member diversity, traditional conflict management methods, relying on experience-based judgment and subjective ratings, struggle to effectively quantify the project manager's situational awareness in conflict decision-making.

[0003] Studies have shown that project managers' emotional state has a significant impact on their cognitive processing and decision-making behavior. Positive emotions can improve attention allocation balance, information integration ability, and risk prediction accuracy, while negative emotions may lead to excessive focus on negative information, insufficient information integration, and decision-making bias towards unfavorable outcomes. However, existing quantitative research on the impact of emotions on situational awareness in conflict decision-making is still limited, mainly due to the following shortcomings:

[0004] (1) There is a lack of real-time measurement methods based on objective physiological indicators and behavioral data, and the assessment results are greatly affected by subjective bias;

[0005] (2) There is a lack of holistic analysis on the role of emotional interference in the three cognitive levels of perception, understanding and prediction, and the mechanism cannot be fully revealed.

[0006] (3) The lack of a quantitative assessment system applicable to project management training and decision support makes it difficult to provide a scientific basis for emotion management and conflict intervention.

[0007] (4) Traditional multimodal evaluation algorithms often use linear data superposition, which cannot effectively capture the nonlinear modulation effect of emotional state on the spatial distribution of eye movement trajectory under work conflict. In addition, the standard random forest algorithm is blind in feature selection and node splitting, resulting in insufficient model recognition ability of specific robust features in high cognitive load conflict situations. When faced with emotional fluctuations and cross-individual differences, the evaluation accuracy and generalization ability are low.

[0008] Therefore, there is an urgent need for a multimodal measurement and evaluation method that can simultaneously capture behavioral data, eye-tracking data, and subjective ratings to quantify project managers' situational awareness capabilities under different emotional states. This would enable the measurement and evaluation of project managers' situational awareness of work conflicts based on eye-tracking technology, thereby providing an operational scientific basis for conflict management and decision optimization. Summary of the Invention

[0009] (a) Technical problems to be solved

[0010] The purpose of this invention is to provide a multimodal assessment method for evaluating the impact of different emotional states on project managers' ability to perceive work conflict situations. This method addresses the technical problems of existing general machine learning models, which perform simple linear superposition when fusing multimodal data, thus failing to effectively capture the nonlinear modulation effect of emotional states on eye movement trajectories under work conflict, and the blindness of standard random forests in node splitting and feature extraction, resulting in low accuracy and generalization ability when facing emotional fluctuations and cross-individual differences. This method can objectively measure the differences in project managers' cognitive preferences and attention allocation in various work conflict situations under different emotional states, improve project managers' information acquisition efficiency, judgment depth, and future trend prediction ability in complex conflict situations, and promote the evolution of engineering project organizational behavior decision-making and conflict intervention mechanisms towards digital intelligence to improve quality and efficiency.

[0011] (II) Technical Solution

[0012] This invention discloses a multimodal assessment method for evaluating the impact of different emotional states on a project manager's ability to perceive work conflicts, comprising the following steps:

[0013] Step S1: Stimulate participants by playing standardized short videos and verify their emotional state using the simplified PANAS scale to ensure that the experimental conditions are controllable and effective; the induction process is synchronized with the conflict decision-making task to simulate the nonlinear interference of emotions on decision-making behavior in a real environment.

[0014] Step S2: Select six typical work conflicts: task conflict, relationship conflict, resource conflict, role conflict, power conflict, and value conflict. For each type, set three decision options with clear positive and negative feedback. Record the immediate results and cumulative performance of each choice through a computer interface, and simultaneously construct a behavioral characteristic matrix.

[0015] Step S3: Collect behavioral performance, eye movement trajectory, and subjective questionnaire data to complete multi-dimensional quantitative statistics and analysis. First, use behavioral data to quantify decision bias and strategy stability; second, use eye movement data to quantify information acquisition efficiency and judgment depth; and finally, use subjective data to comprehensively verify overall cognitive judgment performance.

[0016] Step S4: Use the dynamic time warping algorithm to align the eye-tracking multi-index feature matrix E, the decision-making behavior feature vector B, and the subjective rating scalar S along the time axis to construct a multimodal conflict cognition joint feature space. =[E, B, S]; The interaction scenario is divided into the perceptual interest zone corresponding to the question stem and the comprehension interest zone corresponding to the option; At the same time, the linear mixed-effects model (LME) is used to analyze the eye-tracking processing benchmark, and a group benchmark analysis is performed to quantify the impact of emotional state on information perception and attention allocation; Individual subjects are set as random effects, while interest zones, conflict types, and emotional states are set as fixed effects to quantify the impact of emotional state on information perception, attention allocation, and decision bias. Based on this, three core situational awareness indicators covering the entire process are calculated: perceptual efficiency (SE), comprehension depth (UD), and predictive ability (PA).

[0017] Step S5: Construct a collaborative split forest model that adaptively decouples conflict scenarios and emotional features. First, use a higher-order emotional state modulation operator to adaptively remap and dynamically weight the eye-tracking spatial feature channels in the joint feature space. Second, explicitly introduce a joint regularization constraint term consisting of a decision reaction time penalty factor and conflict cognitive information entropy in the decision tree node splitting calculation to construct an improved collaborative splitting function. Then, guide the model iteration by maximizing this function, adaptively decouple and calculate the globally optimal importance score based on three multimodal features: perceptual efficiency (SE), understanding depth (UD), and predictive ability (PA), and standardize it to generate an adaptive weight vector. Finally, combine the adaptive weight vector with the multimodal features, and output a comprehensive situational awareness score that accurately quantifies cognitive ability in complex conflict scenarios through matrix dot product.

[0018] The specific steps for S1-S5 are described below and will not be repeated here.

[0019] A further improvement of this invention lies in the following: the technical solution for emotion induction in step (1) involves playing positive or negative short videos to induce a preset emotional state in the project manager subject, while simultaneously using a simplified PANAS scale to verify the effectiveness of the emotional intensity and state. This approach can simulate real-world work-related emotional interference situations under controlled experimental conditions, ensuring that the impact of different emotional states on decision-making behavior and situational awareness can be repeatedly measured. Compared to traditional self-report or single emotional stimulus methods, this improvement enhances experimental controllability and the reliability of emotional interference, providing a precise baseline anchor point for evaluating higher-order emotional state regulation operators in the model.

[0020] A further improvement of this invention lies in the following: the technical solution for constructing the work conflict decision-making scenario in step (2) addresses six types of scenarios that project managers may encounter in complex engineering environments: task conflict, relationship conflict, resource conflict, role conflict, power conflict, and value conflict. Three decision options are set for each type, with clear positive and negative feedback guidance for each option, and the decision-making task is presented through a computer interactive interface. This solution can record the immediate results and cumulative performance of each choice made by the participants, and can quantitatively calculate the proportion of choosing the advantageous option and the cumulative decision accuracy rate, thereby ensuring that the experimental scenario closely resembles the real project management scenario and obtaining objective and quantifiable decision-making behavior data. Compared with traditional single-task or questionnaire methods, this improvement significantly enhances the experimental ecological validity and provides a high-dimensional, strongly correlated behavioral decision-making characteristic data stream for subsequent multimodal evaluation models.

[0021] A further improvement of this invention lies in the following: the technical solution for multimodal feature decoupling and group cognitive benchmark representation in step (3) is to simultaneously acquire behavioral performance, eye movement trajectory, and subjective questionnaire data, and conduct multi-dimensional quantitative statistical analysis. Specifically, the correctness and reaction time of the advantageous solution for each conflict type are calculated, the gaze features of the perception and understanding interest areas are extracted, and variance analysis is performed on the self-evaluation scores of each cognitive level. This improved solution quantifies decision bias and strategy stability through behavioral data, quantifies information acquisition efficiency and judgment depth through eye movement data, and comprehensively verifies overall cognitive performance through subjective data, thus constructing a three-in-one cross-validation system. Compared with the traditional single data source analysis mode, this significantly improves the comprehensiveness and objectivity of group characteristic benchmark representation.

[0022] A further improvement of this invention lies in the following: the technical solution for cross-modal feature space alignment and full-process index calculation in step (4) is to use the Dynamic Time Warping (DTW) algorithm to align the time axes of eye movements, behavior, and subjective scalars to construct a joint feature space, and to use the Linear Mixed Effects (LME) model to perform group benchmark analysis on the multidimensional eye movement data. Based on this, the three core capability indicators of perception efficiency (SE), understanding depth (UD), and prediction ability (PA) are precisely derived and decoupled using mathematical formulas. This improved scheme not only eliminates the temporal asynchrony between heterogeneous modalities, but also effectively removes heterogeneous cognitive noise across individuals, realizing the mathematical visualization of the entire situational awareness process, and providing a highly robust feature vector foundation for the adaptive evolution of subsequent machine learning algorithms.

[0023] A further improvement of this invention lies in the following: the technical solution for the adaptive fusion evaluation of the cross-modal algorithm driven by emotion and conflict in step (5) abandons the direct black-box splicing of traditional standard machine learning and innovatively constructs an adaptively decoupled collaborative split forest model. This improved scheme introduces a higher-order emotion state modulation operator to adaptively remap and dynamically weight the eye-tracking spatial feature channels; simultaneously, in the decision tree node splitting calculation of the forest model, a joint regularization constraint term composed of the decision reaction time penalty factor and the cognitive information entropy of the current conflict type is explicitly introduced to adaptively generate weight vectors that match different emotions and conflict types. This mechanism achieves the global optimal decoupling of the importance of multimodal features and accurately and dynamically outputs a comprehensive situational awareness score (CSAS) with high generalization ability.

[0024] This model can accurately quantify the fine-grained performance of project managers in terms of information acquisition efficiency, judgment depth, attention distribution, and strategy evolution trajectory, and adaptively output a comprehensive situational awareness score (CSAS) with high generalization robustness. Compared with traditional single-dimensional assessment or conventional superposition fusion algorithms, its comprehensive regression prediction accuracy is significantly improved by more than 11.2%. It has extremely high technical advancement, application value, and operability in multiple cognitive levels and multiple emotional states.

[0025] (III) Beneficial Effects

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] 1. Controllable Emotion Anchoring. A standardized video-induced and PANAS scale validation scheme was introduced to ensure that the participants' emotional state remained stable during the experiment, providing accurate baseline input for the higher-order emotion state regulation operators in the model.

[0028] 2. High ecological validity of the scenario design. Six types of engineering conflict decision-making scenarios, including tasks and relationships, were constructed. Each type had three options with positive and negative feedback. The test environment closely resembled real-world scenarios, which improved the ecological validity of the experiment and provided a highly robust multimodal data source.

[0029] 3. Multidimensional feature decoupling and group cognitive benchmark representation. The system integrates behavioral performance, eye movement trajectory and subjective questionnaire data, and conducts objective quantitative analysis from the aspects of decision bias and strategy stability, information acquisition efficiency and judgment depth, and overall cognitive performance. It constructs a three-in-one cross-validation system to effectively eliminate the subjective bias of single data patterns.

[0030] 4. Cross-modal temporal alignment and decoupling of full-process indicators. By combining dynamic time warping algorithm to eliminate temporal asynchrony between multimodal data and relying on linear mixture effect model to effectively eliminate heterogeneous cognitive noise across individuals, the three core indicators of perception efficiency, understanding depth and prediction ability are calculated in a refined manner, realizing accurate quantification of the entire process of situational awareness processing.

[0031] 5. Innovative underlying algorithms driven by both emotion and conflict. Abandoning the direct black-box invocation of traditional standard algorithms, an adaptive collaborative splitting forest model is constructed. Feature channels are adaptively remapped through a higher-order emotion state regulation operator, and a joint regularization constraint consisting of a decision reaction time penalty factor and cognitive information entropy is explicitly introduced during the decision tree node splitting process, achieving globally optimal decoupling of the importance of multimodal features.

[0032] 6. Multi-level, objective, and quantitative assessment of the entire cognitive process. By combining an adaptive weight vector automatically generated through feature importance standardization with a multi-modal core feature matrix, the system accurately outputs the project manager's comprehensive situational awareness score, comprehensively and dynamically quantifying the cognitive processing capabilities across the entire process under different emotional interference in complex conflict situations.

[0033] 7. Significantly improved quantitative assessment accuracy. Experimental evaluations verified that, in the regression prediction task of Integrated Situational Awareness Score (CSAS), the classification assessment accuracy of this model is 15.3% higher than that of the traditional Support Vector Machine (SVM); compared with the conventional random forest model, the root mean square error (RMSE) of the assessment is significantly reduced by 12.8%, and the overall prediction accuracy is improved by more than 11.2%, demonstrating extremely high robustness across individuals. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in this invention or the prior art, the accompanying drawings used in the embodiments will be briefly described below:

[0035] Figure 1 This is a breakdown diagram of the overall architecture and steps of the multimodal evaluation method of the present invention;

[0036] Figure 2 Distribution map and interest area division map of decision-making multiple-choice questions;

[0037] Figure 3 Box plots of reaction time for different conflict types under different emotion groups;

[0038] Figure 4 Heatmaps illustrating six types of conflict examples under different emotion groups;

[0039] Figure 5 Example trajectory diagrams of six types of conflict under different emotion groups;

[0040] Figure 6 A graph showing the changes in the original pupil diameter of a single subject under different emotional states;

[0041] Figure 7 This is a self-decoupled distribution diagram of the global weight (W), the core evaluation index of the improved adaptive collaborative forest model of this invention;

[0042] Figure 8 This is a diagram of the overall architecture of the topology network of the adaptive cross-modal fusion evaluation model algorithm driven by emotion and conflict in this invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0044] like Figure 1 and Figure 8 As shown, this embodiment discloses a multimodal assessment model for the impact of different emotional states on a project manager's ability to perceive work conflicts. By integrating conflict situation interaction, standardized emotion induction, cross-modal temporal alignment, and improved machine learning algorithms, it achieves an objective and highly generalized quantitative assessment of a project manager's cognitive abilities throughout the complex engineering environment. The specific implementation architecture is as follows:

[0045] Firstly, state-controlled adaptive emotional anchoring: By playing standardized positive or negative short videos, the target emotional state of the subjects is activated, and the effectiveness of the emotional intensity and state is verified in real time using the simplified PANAS scale, providing accurate baseline state input for evaluating the higher-order emotional state regulation operators in the model.

[0046] Secondly, the construction of a conflict decision-making foundation with high ecological validity: For six types of situations that project managers may encounter in complex engineering environments, such as task conflict, relationship conflict, resource conflict, role conflict, power conflict and value conflict, three decision options with clear positive and negative feedback are set for each type. The immediate results and cumulative performance of each choice are recorded by relying on the computer interactive interface, providing high-dimensional behavioral characteristic data for the model.

[0047] Third, multimodal feature decoupling and group cognitive benchmark representation: The system simultaneously acquires behavioral performance, eye-tracking trajectory, and subjective questionnaire data. Decision bias and strategy stability are objectively quantified through behavioral data, eye-tracking interest area features are extracted to quantify information acquisition efficiency and judgment depth, and variance analysis of subjective data is combined to verify overall cognitive performance, constructing a multidimensional cross-validation underlying data matrix.

[0048] Fourth, cross-modal temporal alignment and full-process index calculation: The Dynamic Time Warping (DTW) algorithm is used to eliminate temporal asynchrony between heterogeneous modalities to construct a joint feature space, and the Linear Mixed Effects (LME) model is used to analyze the group benchmark to eliminate individual heterogeneous cognitive noise. On this basis, three core indices—perceptual efficiency, understanding depth, and predictive ability—are precisely derived and calculated to achieve accurate quantification of the entire cognitive processing process.

[0049] Fifth, adaptive fusion of cross-modal algorithms driven by emotion and conflict: constructing an adaptive collaborative split forest model. A higher-order emotion state modulation operator is used to perform nonlinear remapping weighting on eye-tracking feature channels. During the decision tree node splitting process, a joint regularization constraint term consisting of a decision reaction time penalty factor and conflict cognitive information entropy is explicitly introduced to adaptively decouple multimodal feature weights, ultimately outputting a highly generalized and robust Project Manager Comprehensive Situation Awareness Score (CSAS).

[0050] The concept of this invention is as follows:

[0051] (1) Anchoring a controllable emotional interference state. The target emotional state is activated by standardized audiovisual stimuli and verified in real time using the PANAS scale to simulate the nonlinear interference of different emotions on the project manager's ability to acquire, understand and predict information, so as to achieve accurate anchoring of the baseline state.

[0052] (2) Construct an integrated ecological conflict decision-making scenario. Transform six types of engineering conflicts, such as tasks and relationships, into computer-interactive decision-making tasks with real-time feedback mechanisms. Deeply simulate the real pressure and strategic biases faced by project managers, and provide the model with high-dimensional and objective behavioral decision-making feature sources.

[0053] (3) Multimodal feature decoupling and group cognitive benchmark representation. The system synchronously acquires behavioral performance, eye movement trajectory and subjective questionnaire data, and conducts cross-validation from three dimensions: operation bias and strategy stability, visual information acquisition efficiency and judgment depth, and subjective overall cognitive performance, to construct a multidimensional objective underlying data matrix.

[0054] (4) Cross-modal temporal alignment and decoupling of full-process indicators. The dynamic time warping algorithm is used to eliminate the temporal asynchrony between multimodal data to construct a joint feature space. The linear mixture effect model is used to remove individual differences and eliminate cognitive noise. On this basis, the three core indicators of perception efficiency, understanding depth and prediction ability are precisely derived and calculated to achieve accurate quantification of the entire situational awareness process.

[0055] (5) Adaptive deep fusion of algorithms driven by both emotion and conflict. A collaborative split forest model is constructed, and an emotion state regulation operator is introduced to perform spatial adaptive remapping weighting of the eye-tracking channel; at the same time, a decision response time penalty term and conflict sensitivity information entropy are explicitly introduced into the node splitting function. This adaptively decouples the core feature weights and finally outputs a comprehensive situational awareness score with high generalization robustness, which is used to quantitatively guide the optimization of emotion intervention and conflict handling strategies.

[0056] To demonstrate the advantages of the evaluation method in this invention, steps 1 through 5 will be described in detail below.

[0057] like Figure 1 As shown, a multimodal assessment method for evaluating the impact of different emotional states on a project manager's ability to perceive work conflict situations is presented. The method specifically includes the following calculation steps:

[0058] Step S1: Dynamic Standardized Emotion Induction and State Anchoring: Stimulate participants by playing standardized short videos and verify emotional states using a simplified PANAS scale to ensure that experimental conditions are controllable and effective; the induction process is synchronized with the conflict decision-making task to simulate the nonlinear interference of emotions on decision-making behavior in a real environment.

[0059] In one embodiment, step S1 specifically includes: activating the target conflict emotion in real time by playing a standardized positive or negative audiovisual video of a preset length of 4 minutes; immediately after the video ends, using a simplified PANAS scale to perform a dual mapping verification of the subject's emotional arousal and valence intensity, so as to achieve stable output and intensity verification of the emotional interference source state under controllable experimental conditions.

[0060] Step S2: Constructing a foundation for work conflict decision-making scenarios and a data matrix: Select six typical types of work conflicts: task conflict, relationship conflict, resource conflict, role conflict, power conflict, and value conflict. For each type, set three decision options with clear positive and negative feedback orientations. Record the immediate results and cumulative performance of each choice through a computer interactive interface, and simultaneously construct a behavioral characteristic matrix.

[0061] In one embodiment, step S2 specifically includes: extracting and designing the six types of work conflicts into three decision-making interaction questions for each type in a complex engineering environment, for a total of 18 questions; each question is configured with quantifiable real-time performance indicators, and the project manager's choice of each solution, cumulative accuracy rate, and micro-operational behavior flow triggered by mouse and keyboard are fully recorded through a computer interactive interface under unlimited time pressure, so as to ensure that the test environment is close to the actual scenario and obtains objective behavioral decision characteristics.

[0062] Step S3: Multimodal Feature Decoupling and Group Cognitive Benchmark Representation: Collect behavioral performance, eye movement trajectory, and subjective questionnaire data to complete multi-dimensional quantitative statistics and analysis. This step systematically integrates three aspects of data analysis: first, quantifying decision bias and strategy stability through behavioral data; second, quantifying information acquisition efficiency and judgment depth through eye movement data; and finally, comprehensively verifying overall cognitive judgment performance through subjective data.

[0063] In one embodiment, step S3 utilizes a high-precision hardware eye tracker to capture an eye-tracking processing feature matrix in real time, including the duration of the first fixation, the total duration of fixation, and the number of visits. Simultaneously, an event-triggered marker is used to capture behavioral data streams such as decision-making reaction time, and subjective situational awareness scores are collected. Based on this, decision bias and strategy stability are quantified through behavioral data, information acquisition efficiency and judgment depth are quantified through eye-tracking data, and overall cognitive judgment performance is comprehensively verified through subjective data. This achieves multi-dimensional quantitative analysis of heterogeneous features based on high-precision alignment and verification.

[0064] In addition, step S3 includes steps S3-1 to S3-3:

[0065] S3-1, Behavioral Data Analysis: See [link / reference] Figure 3 As shown, the correctness of the choice of the dominant option and the reaction time of the six types of conflict under different emotion groups are calculated. Repeated measures ANOVA is used to quantify decision bias and strategy stability, and the decision behavior feature vector B is obtained.

[0066] S3-2, Eye-tracking data analysis: See [link / reference] Figure 2 As shown, fixation time, number of fixations, duration of first fixation, total fixation time, number of visits, heat map, trajectory map, and pupil diameter were collected. The decision-making scenario was divided into the perceptual interest area (AOI1, question stem) and the comprehension interest area (AOI2, options). Fixation features were extracted and nonparametric tests were used to quantify the information acquisition efficiency and judgment depth under different emotions, and the eye movement multi-index feature matrix E was obtained.

[0067] S3-3, Subjective Data Analysis: An analysis of variance was performed on the scores of the three cognitive levels of perception, understanding, and prediction in the situational awareness self-assessment scale to verify the performance of cognitive judgment in the decision-making process and obtain the subjective score scalar S.

[0068] Step S4: Cross-modal feature space alignment and full-process index calculation: The Dynamic Time Warping (DTW) algorithm is used to align the eye-tracking multi-index feature matrix E, the decision-making behavior feature vector B, and the subjective rating scalar S along the time axis to construct a multimodal conflict cognition joint feature space. =[E, B, S]; The interaction scenario is divided into the perceptual interest zone (AOI1) corresponding to the question stem and the comprehension interest zone (AOI2) corresponding to the options; a linear mixed-effects model (LME) is used to analyze the eye-tracking processing baseline, and a group baseline analysis is performed to quantify the impact of emotional state on information perception and attention allocation; the individual subjects are set as random effects, while the interest zone, conflict type, and emotional state are set as fixed effects to quantify the impact of emotional state on information perception, attention allocation, and decision bias. Based on this, three core situational awareness ability indicators covering the entire process are calculated: perceptual efficiency (SE), comprehension depth (UD), and predictive ability (PA).

[0069] In one embodiment, the perceptual efficiency (SE), understanding depth (UD), and predictive ability (PA) are calculated as follows:

[0070] ①Perceived efficiency (SE):

[0071]

[0072] Among them, T AOI1 and T AOI2 These represent the fixation time for the perceptual interest area and the comprehension interest area, respectively.

[0073] ② Understanding Depth Undefined (UD):

[0074]

[0075] Among them, V AOI2 To understand the number of visits to areas of interest, P advantage The proportion of participants who chose the dominant option.

[0076] ③ Predictive ability PA:

[0077]

[0078] Where N is the total number of decision questions, C i For the consistency of decision options in question i (a binary classification indicator, 1 for consistency and 0 for inconsistency), w rt,i These are the weighting coefficients calculated based on reaction time.

[0079] Step S5: Construct a collaborative split forest model that adaptively decouples conflict scenarios and emotional features. First, use a higher-order emotional state modulation operator to adaptively remap and dynamically weight the eye-tracking spatial feature channels in the joint feature space. Second, explicitly introduce a joint regularization constraint term consisting of a decision reaction time penalty factor and conflict cognitive information entropy in the decision tree node splitting calculation to construct an improved collaborative splitting function. Then, guide the model iteration by maximizing this function, adaptively decouple and calculate the globally optimal importance score based on three multimodal features: perceptual efficiency (SE), understanding depth (UD), and predictive ability (PA), and standardize it to generate an adaptive weight vector. Finally, combine the adaptive weight vector with the multimodal features, and output a comprehensive situational awareness score that accurately quantifies cognitive ability in complex conflict scenarios through matrix dot product.

[0080] In one embodiment, step S5 involves adaptive fusion evaluation based on the emotion-conflict dual-driven collaborative split forest algorithm: constructing an adaptive collaborative split forest model to perform deep nonlinear interweaving and decoupling of the perceptual efficiency (SE), understanding depth (UD), and predictive ability (PA) features, specifically including steps S5-1 to S5-4:

[0081] S5-1, Emotion-Driven Channel Adaptive Dynamic Weighting: Constructing Higher-Order Emotional State Regulation Operators ,in Emotional space channel adaptive remapping operator This represents a combination of a situational awareness self-rating scale reflecting subjective assessment and an emotional baseline state vector quantified by dynamic pupil diameter. For the emotion prior modulation coefficients extracted analytically based on the linear mixture effect model, the feature space is... The eye-tracking channel, which maps to the perceived and understood regions of interest, undergoes adaptive remapping to explicitly capture the spatial modulation effect of emotional states on the visual search trajectory, resulting in a weighted feature matrix. ':

[0082]

[0083] in, (E, B) represents the nonlinear interaction matrix between eye-tracking features and behavioral representations. This is the compensatory cognitive load adjustment coefficient.

[0084] S5-2, Conflict Context Sensitive Tree Node Cooperative Splitting: To overcome the blindness of feature selection, a joint regularization constraint term consisting of a reaction time penalty factor and conflict context sensitivity is explicitly introduced into the calculation of the Gini Index during the decision tree node splitting process of the cooperative forest model, thus constructing an improved cooperative splitting function. H(m):

[0085] Let the dataset containing a node t in a decision tree be D, and let the dataset contain K categories. Then the formula for calculating the Gini index Gini(t) of node t is:

[0086]

[0087] Where, p k This represents the proportion of samples belonging to class k in the dataset D of node t (i.e., the probability that a sample belongs to class k).

[0088] Let m be the current decision split node, and also represent the set of samples entering this node, containing a total of N samples. m Feature X m To evaluate perceptual efficiency (SE), understanding depth undefined (UD), or predictive ability (PA), let feature X... m The candidate input features selected at node m for split evaluation; when based on feature X m Split node m into its left child node m L and right child node m R When the number of samples contained in these two child nodes is N, it is denoted as N. L and N R The feature X m Improved cooperative splitting function before and after classification at node m The expression for H(m) is:

[0089]

[0090] Among them, RT m H(C) represents the mean decision reaction time corresponding to the current feature. m ) represents the cognitive information entropy of the current conflict type. 1 and 2 represents the adaptive balancing hyperparameter.

[0091] S5-3, Adaptive Calculation and Standardization of Feature Importance: In a forest model containing T trees, by maximizing... H(m) guides the model in adaptive iteration, decoupling the original importance scores of each feature.

[0092]

[0093] Where t is the traversal index of a single decision tree, t=1, 2, ..., T; M t This represents the set of all split nodes in the t-th decision tree that use the current specific feature as the splitting criterion. This formula represents the summation of the total improvement gain generated by the specific feature across all node splits in the entire forest model.

[0094] Finally, let the set of original importance scores for the three core features extracted be V = {V SE V UD V PA The data is standardized using Min-Max and mapped to the [0, 1] interval, thereby automatically generating an adaptive weight vector that matches the current emotional state and conflict type of the participant. , specifically Figure 8 As shown; where any feature corresponds to a weight component. The specific calculation relationship is as follows:

[0095]

[0096] Among them, V i V represents the original importance score corresponding to this feature. max and V min These represent the maximum and minimum values ​​in the original feature importance score set V, respectively.

[0097] S5-4, Cross-modal Integrated Situational Awareness Score Output: Combining the adaptive weight vector W with multimodal features, the final multimodal integrated situational awareness score (CSAS) is output in matrix dot product form, accurately quantifying the entire cognitive ability under different emotional interferences.

[0098]

[0099] Where F is the core cognitive multimodal feature vector, These represent the baseline prediction regression output values ​​of perception efficiency, understanding depth, and prediction ability in the adaptive collaborative forest, respectively.

[0100] Since this model is a collaborative forest ensemble architecture containing T regression trees, let the local regression prediction function of a single decision tree t for the i-th cognitive dimension be h. t,i ( Then the regression prediction function f in the three dimensions i ( The specific calculation formulas for ) are expanded as follows:

[0101]

[0102] Where T is the total number of decision trees in the adaptive collaborative forest, h t,i ( ) represents the t-th single regression tree for the feature matrix. A local regression prediction function on the i-th cognitive dimension (perceptual efficiency SE, understanding depth UD, or predictive ability PA).

[0103] Based on steps 1-5 above, the experimental methods used in this invention are verified and analyzed as follows:

[0104] (A) Verification of behavioral decision-making characteristics and strategy evolution

[0105] This invention first analyzes the behavioral data of project managers in completing work conflict decision-making tasks, divides the experimental trials into positive and negative emotion groups and six types of work conflict, statistically analyzes the correctness of the selection of the superior solution and the reaction time, and uses repeated measures ANOVA to evaluate the robustness of these data as input features of the algorithm.

[0106] Preferred Solution Selection Accuracy: Significant differences in accuracy were observed across different emotion groups (F=16.547, P<0.001). Within each emotion group, the accuracy for different conflict types exhibited highly non-linear fluctuations (positive F=7.787, negative F=6.045, both P<0.001). The data indicates that task conflict and relationship conflict are highly susceptible to emotional fluctuations, providing core behavioral representation features for the subsequent non-linear weighting of the algorithm. Decision Reaction Time: Significant differences in decision reaction time were observed across different emotion groups (F=8.668, P<0.05), and the interaction effect between emotion and conflict type was highly significant (F=7.995, P<0.001). This result fully confirms the effectiveness of introducing a decision reaction time penalty factor (RT) into the node splitting function of the collaborative splitting forest model in this invention. m The mathematical and logical correctness of ).

[0107] (B) Eye-tracking cognitive processing model and LME benchmark validation

[0108] This invention divides the key areas in a decision-making task into regions of interest (AOIs), including six types of conflict options and areas for progress and feedback information. The extracted core indicators include first fixation time, total access time, average fixation duration, number of fixations and accesses, and pupil diameter. Simultaneously, heatmaps and trajectory maps are generated for spatial and path visualization analysis. The results of the linear mixed-effects model analysis are shown in Table 1 below. Example heatmaps for the six conflict types under different emotion groups are shown below. Figure 4 As shown, the trajectory diagrams of six types of conflict examples under different emotion groups are as follows: Figure 5 As shown in the figure, the changes in the original pupil diameter of a single subject under different emotions are illustrated in the figure below. Figure 6 As shown.

[0109] Project managers prioritized focusing on advantageous / conflicting options during the information acquisition phase, exhibiting shorter initial fixation times and longer total access times, indicating high information acquisition efficiency. For high-risk or disadvantageous options, the average fixation duration and number of fixations increased, suggesting improved depth of understanding and evaluation of decision-making information. The low-frequency switching characteristic of access frequency is consistent with the risk-avoidance trend in behavioral data, validating the decrease in exploratory behavior and the increase in stable choices during strategy evolution. Furthermore, heatmaps and trajectory maps show that attention is concentrated on key options, the scanning path exhibits a strategic distribution, and changes in pupil diameter reflect task stress and cognitive load levels.

[0110] Further analysis revealed differences in attention allocation across different types of conflict. Gaze distribution was more concentrated on key options in task and relationship conflicts, while it was more dispersed in resource and role conflicts, indicating that project managers flexibly adjust information processing strategies based on conflict type. Overall, eye-tracking analysis comprehensively reflects participants' visual information search, cognitive processing, and attention allocation during conflict decision-making, providing reliable physiological support for behavioral data.

[0111] Table 1. Analysis results of the linear mixed-effects model

[0112]

[0113] Note: 1. * indicates P < 0.001, ** indicates P < 0.05, *** indicates P < 0.01. Emotions are compared with positive emotions as the reference group, and conflicts are compared with resource conflicts as the reference group.

[0114] 2. The LME results show that emotional state has a significant nonlinear modulation effect on the perception (AOI1) and understanding (AOI2) stages (e.g., fixation time increases significantly by 1.50 s under negative emotions), validating the model's inclusion of an emotional state modulation operator. The necessity of cross-modal channel weighting.

[0115] (C) Evaluation of model algorithm performance and ablation comparison verification

[0116] To verify the effectiveness of the proposed improved cooperative splitting forest regression algorithm in multimodal fusion and Project Manager Comprehensive Situation Awareness Score (CSAS) prediction, the algorithm was compared with traditional black-box models (Support Vector Machine (SVM) and standard Random Forest) on a full-sample dataset (80% training set, 20% independent test set). Ablation experiments were also conducted on the two core improved modules of the algorithm (sentiment weighting module and improved cooperative splitting function). The accuracy of the prediction regression was evaluated using the conventional root mean square error (RMSE) and overall prediction accuracy. The verification results are shown in Table 2.

[0117] Table 2 Comparison of multimodal fusion prediction performance of different evaluation models and ablation experiment results

[0118]

[0119] The evaluation data in Table 2 strongly demonstrates that by incorporating decision reaction time penalty terms and conflict sensitivity information entropy into the splitting rules of cascaded tree decision nodes, this invention can adaptively eliminate cross-individual data noise and accurately decouple compensatory features under high cognitive processing load.

[0120] The comprehensive score generated by multimodal fusion shows high statistical consistency with individual visual search features, operational behavior features, and subjective psychological measurement results, confirming the effectiveness of the adaptive weight vector of this invention. The robustness.

[0121] Therefore, it can be seen that the present invention has made the following innovative improvements compared to the prior art:

[0122] This invention proposes a multimodal assessment model for the impact of different emotional states on project managers' ability to perceive work conflicts, focusing on the refined design and cognitive measurement of conflict scenarios. In the experimental design, six typical work conflicts—task conflict, relationship conflict, resource conflict, role conflict, power conflict, and value conflict—are selected. Each type has three decision options with two possible outcomes (positive and negative). The decision-making task is presented through a computer interface to ensure the simulated environment closely resembles real project management situations. A standardized emotion induction scheme is used to induce positive or negative emotional states in participants, quantifying the interference of emotions on decision-making behavior. Eye movement features are collected in real time using a high-precision eye tracker, simultaneously recording behavioral data and subjective situational perception scores. A linear mixed-effects model is used to analyze eye movement processing patterns. Based on this, an adaptive cross-modal fusion assessment model driven by emotion and conflict is constructed. This model introduces a joint regularization operator consisting of a penalty term for decision-making reaction time and conflict sensitivity into the feature space of a random forest, optimizing the collaborative splitting function of tree nodes to achieve a deep multimodal nonlinear interweaving of eye-tracking visual search strategies, operational behavior representations, and subjective psychological measurements. This invention can objectively measure the differences in cognitive preferences and attention allocation of project managers in various work conflict situations under different emotional states, improve project managers' information acquisition efficiency, judgment depth and future trend prediction ability in complex conflict situations, and help promote the evolution of engineering project organizational behavior decision-making and conflict intervention mechanisms towards digitalization to improve quality and efficiency.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multimodal assessment method for evaluating the impact of different emotional states on project managers' ability to perceive work conflict situations, characterized in that... The steps include the following: Step S1: Stimulate participants by playing standardized short videos and verify their emotional state using the simplified PANAS scale to ensure that the experimental conditions are controllable and effective; the induction process is synchronized with the conflict decision-making task to simulate the nonlinear interference of emotions on decision-making behavior in a real environment. Step S2: Select six typical work conflicts: task conflict, relationship conflict, resource conflict, role conflict, power conflict, and value conflict. For each type, set three decision options with clear positive and negative feedback. Record the immediate results and cumulative performance of each choice through a computer interface, and simultaneously construct a behavioral characteristic matrix. Step S3: Collect behavioral performance, eye movement trajectory, and subjective questionnaire data to complete multi-dimensional quantitative statistics and analysis. First, use behavioral data to quantify decision bias and strategy stability; second, use eye movement data to quantify information acquisition efficiency and judgment depth; and finally, use subjective data to comprehensively verify overall cognitive judgment performance. Step S4: Use the dynamic time warping algorithm to align the eye-tracking multi-index feature matrix E, the decision-making behavior feature vector B, and the subjective rating scalar S along the time axis to construct a multimodal conflict cognition joint feature space. =[E, B, S]; The interaction scenario is divided into the perceptual interest zone corresponding to the question stem and the comprehension interest zone corresponding to the option; At the same time, the linear mixed-effects model (LME) is used to analyze the eye-tracking processing benchmark, and a group benchmark analysis is performed to quantify the impact of emotional state on information perception and attention allocation; Individual subjects are set as random effects, while interest zones, conflict types, and emotional states are set as fixed effects to quantify the impact of emotional state on information perception, attention allocation, and decision bias. Based on this, three core situational awareness indicators covering the entire process are calculated: perceptual efficiency (SE), comprehension depth (UD), and predictive ability (PA). Step S5: Construct a collaborative split forest model that adaptively decouples conflict scenarios and emotional features. First, use a higher-order emotional state modulation operator to adaptively remap and dynamically weight the eye-tracking spatial feature channels in the joint feature space. Second, explicitly introduce a joint regularization constraint term consisting of a decision reaction time penalty factor and conflict cognitive information entropy in the decision tree node splitting calculation to construct an improved collaborative splitting function. Then, guide the model iteration by maximizing this function, adaptively decouple and calculate the globally optimal importance score based on three multimodal features: perceptual efficiency (SE), understanding depth (UD), and predictive ability (PA), and standardize it to generate an adaptive weight vector. Finally, combine the adaptive weight vector with the multimodal features, and output a comprehensive situational awareness score that accurately quantifies cognitive ability in complex conflict scenarios through matrix dot product.

2. The multimodal assessment method for assessing the impact of different emotional states on a project manager's ability to perceive work conflict situations, as described in claim 1, is characterized in that... Step S1 specifically includes: activating the target conflict emotion in real time by playing a standardized positive or negative audiovisual video with a preset length of 4 minutes; immediately after the video ends, using a simplified PANAS scale to perform a dual mapping verification of the subject's emotional arousal and valence intensity, so as to achieve stable output and intensity verification of the emotional interference source state under controllable experimental conditions.

3. The multimodal assessment method for assessing the impact of different emotional states on a project manager's ability to perceive work conflict situations, as described in claim 1, is characterized in that... Step S2 specifically includes: extracting and designing the six types of work conflicts into three decision-making interaction questions for each type in a complex engineering environment, for a total of 18 questions; each question is equipped with quantifiable real-time performance indicators, and the project manager's choice of each solution, cumulative accuracy rate, and micro-operational behavior flow triggered by mouse and keyboard are fully recorded through a computer interactive interface under unlimited time pressure, in order to ensure that the test environment is close to the actual scenario and to obtain objective behavioral decision characteristics.

4. The multimodal assessment method for assessing the impact of different emotional states on a project manager's ability to perceive work conflict situations, as described in claim 1, is characterized in that... Step S3 specifically includes: Step S3-1: Behavioral data analysis: As shown in Figure 3, calculate the correctness of the choice of the superior solution and the reaction time for the six types of conflict under different emotion groups, use repeated measures ANOVA to quantify decision bias and strategy stability, and obtain the decision behavior feature vector B; Step S3-2: Eye movement data analysis: As shown in Figure 2, collect fixation time, number of fixations, duration of first fixation, total fixation time, number of visits, heat map, trajectory map and pupil diameter. Divide the decision-making scenario into perceptual interest area AOI1 and comprehension interest area AOI2; extract fixation features and use nonparametric tests to quantify the information acquisition efficiency and judgment depth under different emotions, and obtain the eye movement multi-index feature matrix E; Step S3-3: Subjective Data Analysis: Perform ANOVA on the scores of the three cognitive levels of perception, understanding and prediction in the Situation Awareness (SA) self-rating scale to verify the cognitive judgment performance in the decision-making process and obtain the subjective rating scalar S.

5. The multimodal assessment method for assessing the impact of different emotional states on a project manager's ability to perceive work conflict situations, as described in claim 1, is characterized in that... Step S4 specifically includes the following calculation methods for the perception efficiency SE, understanding depth UD, and prediction capability PA: ①Perceived efficiency (SE): Among them, T AOI1 and T AOI2 These represent the fixation time in the perceptual interest area and the comprehension interest area, respectively; ② Understanding Depth Undefined (UD): Among them, V AOI2 To understand the number of visits to areas of interest, P advantage The proportion of participants who chose the dominant option; ③ Predictive ability PA: Where N is the total number of decision questions, C i For consistency of decision options in question i, w rt,i These are the weighting coefficients calculated based on reaction time.

6. The multimodal assessment method for assessing the impact of different emotional states on a project manager's ability to perceive work conflict situations, as described in claim 5, is characterized in that... Step S5 also includes: Step S5-1: Emotion-Driven Channel Adaptive Dynamic Weighting: Constructing Higher-Order Emotional State Modulation Operators ,in Emotional space channel adaptive remapping operator This represents a combination of a situational awareness self-rating scale reflecting subjective assessment and an emotional baseline state vector quantified by dynamic pupil diameter. For the emotion prior modulation coefficients extracted analytically based on the linear mixture effect model, the feature space is... The eye-tracking channel, which maps to the perceived and understood regions of interest, undergoes adaptive remapping to explicitly capture the spatial modulation effect of emotional states on the visual search trajectory, resulting in a weighted feature matrix. ': in, (E, B) represents the nonlinear interaction matrix between eye-tracking features and behavioral representations. This is the compensatory cognitive load adjustment coefficient.

7. The multimodal assessment method for assessing the impact of different emotional states on a project manager's ability to perceive work conflict situations, as described in claim 6, is characterized in that... Step S5 also includes: Step S5-2: Conflict Context Sensitive Tree Node Cooperative Splitting: To overcome the blindness of feature selection, during the decision tree node splitting process of the cooperative forest model, a joint regularization constraint term consisting of the reaction time penalty factor and conflict context sensitivity is explicitly introduced into the Gini Index calculation to construct an improved cooperative splitting function. H(m): Let the dataset containing a node t in a decision tree be D, and let the dataset contain K categories. Then the formula for calculating the Gini index Gini(t) of node t is: Where, p k This represents the proportion of samples belonging to the k-th class in the dataset D for node t; Let m be the current decision split node, and also represent the set of samples entering this node, containing a total of N samples. m Feature X m To evaluate perceptual efficiency (SE), understanding depth undefined (UD), or predictive ability (PA), let feature X... m The candidate input features selected at node m for split evaluation; when based on feature X m Split node m into its left child node m L and right child node m R When the number of samples contained in these two child nodes is N, it is denoted as N. L and N R The feature X m Improved cooperative splitting function before and after classification at node m The expression for H(m) is: Among them, RT m H(C) represents the mean decision reaction time corresponding to the current feature. m ) represents the cognitive information entropy of the current conflict type. 1 and 2 represents the adaptive balancing hyperparameter.

8. The multimodal assessment method for assessing the impact of different emotional states on a project manager's ability to perceive work conflicts, as described in claim 7, is characterized in that... Step S5 also includes: Step S5-3: Adaptive Calculation and Standardization of Feature Importance: In a forest model containing T trees, by maximizing... H(m) guides the model in adaptive iteration, decoupling the original importance scores V of each feature. m Where t is the traversal index of a single decision tree, t=1, 2, ..., T; M t It represents the set of all split nodes in the t-th decision tree that are split based on the current specific feature; Finally, let the set of original importance scores for the three core features extracted be V = {V SE V UD V PA The data is standardized using Min-Max and mapped to the [0, 1] interval, thereby automatically generating an adaptive weight vector that matches the current emotional state and conflict type of the participant. Where, any feature corresponds to a weight component. The specific calculation relationship is as follows: Among them, V i V represents the original importance score corresponding to this feature. max and V min These represent the maximum and minimum values ​​in the original feature importance score set V, respectively.

9. The multimodal assessment method for assessing the impact of different emotional states on a project manager's ability to perceive work conflict situations, as described in claim 8, is characterized in that... Step S5 also includes: Step S5-4: Cross-modal Integrated Situational Awareness Score Output: Combining the adaptive weight vector W with multimodal features, the final multimodal integrated situational awareness score (CSAS) is output in matrix dot product form, accurately quantifying the full-process cognitive ability under different emotional interferences. Where F is the core cognitive multimodal feature vector, These represent the baseline prediction regression output values ​​of perception efficiency, understanding depth, and prediction ability in the adaptive collaborative forest, respectively.

10. The multimodal assessment method for assessing the impact of different emotional states on a project manager's ability to perceive work conflict situations, as described in claim 9, is characterized in that... Let h be the local regression prediction function of a single decision tree t for the i-th cognitive dimension. t,i ( Then the regression prediction function f in the three dimensions i ( The specific calculation formulas for ) are expanded as follows: Where T is the total number of decision trees in the adaptive collaborative forest, h t,i ( ) represents the t-th single regression tree for the feature matrix. The local regression prediction function on the i-th cognitive dimension.

Images