Human-computer interaction interface operation redundancy comprehensive evaluation method and system based on interaction log and path data

Abstract

The application discloses a human-computer interaction interface operation redundancy comprehensive evaluation method and system based on interaction logs and path data. Interaction logs, actual interface state paths, effective candidate path sets formed according to different reasonable completion modes, interface element attributes and the like of a target task are acquired; eight indexes are calculated; the eight indexes are homogenized and standardized to form four first-level evaluation dimension scores, and CRI is calculated, and the eight indexes all enter CRI with positive secondary index combination weights; weighted quantile points are calculated based on CRI values and sample weights from a reference sample library to form four-level evaluation boundaries, and the redundancy level of an interface to be evaluated is determined according to the four-level evaluation boundaries; and according to abnormal indexes and contribution degrees, a problem type is matched and optimization suggestions are output. The application converts interface operation redundancy problems into quantified evaluation results that are calculable, traceable and optimizable, and provides comprehensive, adaptive and interpretable calculable basis for complexity identification, problem positioning and design optimization of a graphical human-computer interaction interface.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction technology, specifically a method and system for comprehensive evaluation of redundancy in human-computer interaction interface operations based on interaction logs and path data. This invention is applicable to graphical human-computer interaction interfaces with the ability to identify interface states, interactive elements, and record operation logs, including but not limited to mobile terminal interfaces, desktop software interfaces, vehicle-mounted human-computer interfaces, industrial control interfaces, smart screen interfaces, and virtual reality or augmented reality interactive interfaces. Background Technology

[0002] Graphical human-computer interfaces (HCIs) are widely used in mobile terminals, desktop software, vehicle cockpits, industrial control systems, smart screens, and virtual reality or augmented reality systems. To complete a target task, users need to drive continuous changes in the interface state through clicks, touches, buttons, knobs, gestures, voice, gaze, or controllers. When the information architecture, interactive feedback, label semantics, or visual presentation are unreasonable, it can lead to extra operations, invalid interactions, immediate rollbacks, state revisits, function retries, and path deviations, and can be affected by semantic competition or visual interference from non-target elements.

[0003] Several evaluation schemes for human-computer interaction interfaces already exist in the current technology. These schemes often use task completion time, success rate, subjective scales, or localized indicators for a single interface, making it difficult to uniformly identify various redundancies from the perspective of the interaction process and path structure. Some schemes only apply to mouse or touch clicks, failing to cover interaction modalities such as knobs, voice, and eye-tracking confirmation; some schemes use pages as path nodes, failing to distinguish between windows, modes, pop-ups, and operation states within the same page; and existing schemes typically lack adaptation mechanisms for determining level boundaries based on interface type, task type, and sample distribution. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a comprehensive evaluation method and system for the redundancy of human-computer interaction interface operations based on interaction logs and path data, addressing the shortcomings of the prior art. This method and system comprehensively covers user operation behavior, path results, and semantic and visual interference factors by setting eight secondary indicators and incorporating them into four primary evaluation dimensions, thus overcoming the deficiency of single evaluation dimensions in existing solutions. By uniformly representing page, window, mode, pop-up, and other state types with interface states and uniformly representing multiple input modalities with generalized interaction events, the method is applicable across terminals and modalities. By dynamically determining the level boundary based on the weighted quantiles of reference samples, the evaluation criteria can adapt to the sample distribution of different interface and task categories. At the same time, by calculating the contribution of indicators and locating abnormal indicators and outputting optimization suggestions through preset mapping rules, the evaluation results are interpretable and operable, forming a closed-loop design of "evaluation-diagnosis-optimization".

[0005] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows:

[0006] A comprehensive evaluation method for redundancy in human-computer interaction interface operations based on interaction logs and path data includes the following steps:

[0007] S101, obtain the interaction logs corresponding to the target task, the actual interface state path, one or more valid candidate paths, at least one of the functional branches or preset task rules corresponding to the actual interaction path to be evaluated, task target element attributes, and interface element attributes; wherein, at least one of the functional branches or preset task rules is used to distinguish different reasonable completion methods of the same target task and determine the set of valid candidate paths to which the actual interaction path to be evaluated belongs; the valid candidate path is the task completion path that meets the target task completion conditions, necessary state constraints, necessary interaction steps, and preset business or security constraints, and passes the log integrity verification.

[0008] S102, calculate eight secondary indicators, including extra operation ratio, invalid interaction rate, instant rollback rate, interface revisit rate, function retry rate, path deviation, semantic confusion, and visual interference. Among them, interface elements in each interface state that are not directly related to the current task target and have occlusion, motion, significant contrast, or continuous exposure characteristics are uniformly regarded as non-target visual elements. The visual interference is directly calculated based on the number, area ratio, spatial position weight, and interference morphology intensity of the non-target visual elements.

[0009] S103, the eight secondary indicators are homogenized and standardized to obtain the standardized values ​​of the eight secondary indicators;

[0010] S104, the extra operation ratio and invalid interaction rate are classified as unnecessary redundancy primary indicators, the instant rollback rate, interface revisit rate and function retry rate are classified as repetitive redundancy primary indicators, the path deviation degree is classified as path deviation redundancy primary indicators, and the semantic confusion degree and visual interference degree are classified as interference redundancy primary indicators. The scores of the four primary indicators are calculated according to the weights of the secondary indicators under each primary indicator and the standardized values ​​of the secondary indicators. When the path deviation redundancy dimension only includes the path deviation degree, its primary indicator score is directly determined by the standardized value of the path deviation degree.

[0011] S105. Calculate the comprehensive redundancy index (CRI) based on the weights of the four primary indicators, the weights of the secondary indicators within each primary indicator layer, and the standardized values ​​of the secondary indicators. The weights of the primary indicators and the weights of the secondary indicators within each secondary indicator layer are all positive and satisfy the normalization constraints. The product of the weights of the primary indicators and the weights of the secondary indicators within each secondary indicator layer is used as the combined weights of the secondary indicators, so that all eight secondary indicators enter the CRI with positive combined weights.

[0012] S106, Assign sample weights to the CRI values ​​of the reference samples and sort them according to the CRI values; determine at least two level boundaries based on the comprehensive redundancy index distribution in the reference sample set; compare the CRI of the interface to be evaluated with the level boundaries to determine the overall redundancy level of the interface to be evaluated.

[0013] S107, using the same reference samples and sample weights as in step S106, determine the abnormal limits of each indicator based on the reference distribution of the standardized values ​​of each secondary indicator; determine the indicator contribution by multiplying the standardized value of each secondary indicator of the interface to be evaluated with the weight of its corresponding secondary indicator combination, and compare the standardized value of each secondary indicator with the corresponding indicator abnormal limit to identify abnormal indicators; match the abnormal indicators or combinations of abnormal indicators with preset indicators and problem mapping rules, and output interface optimization suggestions containing preset problem types and adjustment actions according to the indicator contribution.

[0014] As a further improved technical solution of the present invention, the human-computer interaction interface is a graphical human-computer interaction interface with the ability to identify interface status, interactive elements and operation log recording, including at least one of mobile terminal interface, desktop software interface, vehicle human-computer interface, industrial control interface, smart screen interface and virtual reality or augmented reality interaction interface; the interaction log records at least one interaction event among click, touch, button, knob, gesture, voice trigger, gaze confirmation and controller input.

[0015] As a further improvement of the present invention, at least two of the following are combined into an interface state identifier: page identifier, window identifier, mode identifier, pop-up identifier, operation mode identifier, and interactive element tree summary. This allows page switching, mode switching, window switching, pop-up state change, and operation mode change to be uniformly represented as interface state transition.

[0016] As a further improved technical solution of the present invention, the additional operation ratio is determined based on the difference between the actual interaction step size and the reference interaction step size, the ratio of the difference to the reference interaction step size, and its bounded result.

[0017] The actual interaction path used for calculating the additional operation ratio is a valid actual interaction path that satisfies the target task completion conditions, necessary state constraints, necessary interaction steps, and preset business or security constraints. The valid candidate path is a task completion path that satisfies the target task completion conditions, necessary state constraints, necessary interaction steps, and preset business or security constraints, and passes log integrity verification. For the same target task with different reasonable completion methods, the valid candidate paths are distinguished according to functional branches or preset task rules, and corresponding valid candidate path sets are formed respectively. For the actual interaction path to be evaluated, the valid candidate path set to which it belongs is determined according to its corresponding functional branch or preset task rules, and then the path with the smallest total interaction step length is selected from the valid candidate path set as the valid benchmark path. When there is only one valid benchmark path, the valid benchmark path is selected as the selected valid benchmark path. When there are multiple valid candidate paths with a total interaction step length equal to the minimum value, all of them are used as valid benchmark paths, and the valid benchmark path with the smallest deviation from the actual interface state path is selected as the selected valid benchmark path for the index calculation of the current task sample. The benchmark interaction step length is the total interaction step length of the selected valid benchmark path.

[0018] The invalid interaction rate is determined by the sum of the number of invalid interaction events and the number of valid interaction events after removing redundant loops. Invalid interaction events are those that do not cause a change in the interface state within the response window, do not generate a system acceptance flag, and do not generate feedback related to the target task.

[0019] As a further improvement of the present invention, the instant rollback rate is determined based on the number of interface state jumps and the number of effective interface state jumps contained in the rollback segment in the actual interface state sequence after deduplication.

[0020] The interface revisit rate is determined based on the cumulative number of revisits and the number of valid visits between two adjacent visits to the same interface state, with an interval exceeding two state positions.

[0021] The function retry rate is determined by the number of function retry candidates exceeding the initial call within the same target task stage, after excluding those that are reasonable repeated calls required to complete the target task according to task rules, such as quantity increase / decrease, continuous parameter adjustment, pagination, scrolling, text editing, or others. This number is then combined with the number of valid interactions. The success of the initial call, whether the expected task state is reached, and whether the feedback is clear are only used for diagnosing the cause of the function retry and are not necessary conditions for identifying the function retry.

[0022] As a further improved technical solution of the present invention, the path deviation is determined based on the edit distance between the actual interface state sequence and the interface state sequence of the selected effective benchmark path, the number of interface state jumps contained in the interface state sequence of the effective benchmark path, and the bounded result; when there are multiple effective benchmark paths in the effective candidate path set to which the current task sample belongs, the path deviation is calculated based on the actual interface state sequence and the interface state sequence of each effective benchmark path, and the minimum value is taken, and the effective benchmark path corresponding to the minimum value is used as the selected effective benchmark path for the calculation of the current task sample index.

[0023] As a further improvement of the present invention, the semantic confusion degree is determined based on the semantic similarity between the target interactive element and each non-target interactive element in the same interface state, the number of non-target semantic interference elements, and the saturation coefficient of the number of semantic interference items; the semantic similarity is obtained based on at least one of text labels, icon semantic labels, function identifiers, system function codes, accessibility names, or operation descriptions.

[0024] The visual interference degree is determined based on the set of non-target visual elements in each interface state. The saturation coefficient of the number of non-target visual elements is determined according to a preset saturation upper limit. The area ratio, spatial position weight, and interference morphology intensity of a single non-target visual element are multiplied and aggregated sequentially within the interface state to obtain the visual interference degree of the interface state. The visual interference degree of the task sample is obtained by aggregating among the effective interface states of the same task sample. The visual interference degree of the evaluation unit is obtained by aggregating within the evaluation unit composed of one or more effective task samples. When the set of non-target visual elements is empty, the visual interference contribution corresponding to that interface state is zero.

[0025] As a further improved technical solution of the present invention, in step S103, the secondary index is subjected to truncated linear standardization using the lower reference quantile and upper reference quantile of the reference distribution.

[0026] In step S104, the expert's given first... The average importance score of the first-level indicator and the number of items within the same first-level indicator The weighted average scores of the secondary indicators were normalized to obtain the results for the second-level indicators. Item Level 1 Indicator Weight and the The first-level indicator Intra-layer weight of secondary indicators and with Determine the first The first-level indicator The weighting of the secondary indicators;

[0027] For the First-level indicators, by Calculate the scores for the primary indicators; among which, This indicates that the first step obtained in step S103 is... The first-level indicator Standardized values ​​of secondary indicators; Indicates the first The number of secondary indicators included in a primary indicator;

[0028] In step S105, the comprehensive redundancy index is calculated. , ,in , , , and ;

[0029] The weights of the primary indicators and the weights within the secondary indicator layers are all positive values. The sum of the weights of all primary indicators is 1. The sum of the weights within the secondary indicator layers under the same primary indicator is 1. The sum of the combined weights of all secondary indicators is 1.

[0030] As a further improvement of the present invention, the level boundary is adaptively determined according to the comprehensive redundancy index distribution of the reference sample set; the reference samples are stratified according to interface category, task category or a combination of the two, and sample weights are set for different categories;

[0031] Multiple ordered level regions are obtained based on the empirical distribution or clustering distribution of the weighted comprehensive redundancy index, and the level boundaries are determined by the boundary positions between adjacent level regions.

[0032] In the method of determining the class boundaries using weighted quantiles, when four redundant classes are set, the weighted 25%, 50% and 75% quantiles are used as three class boundaries respectively.

[0033] In the method of determining the level boundary using clustering, the comprehensive redundancy index of the reference sample is divided into multiple ordered clusters, and the corresponding level boundary is determined according to the intersection of adjacent cluster centers, adjacent cluster boundaries, or adjacent probability distributions.

[0034] The evaluation unit is the same as that used to determine the grade boundary for the sample to be evaluated; when the evaluation unit is a single valid execution record of a single user or a single session, the total number of samples is 1; when the evaluation unit is a group sample consisting of multiple valid execution records, each reference sample and the sample to be evaluated use the same number of valid execution records, aggregation method and sample weight.

[0035] As a further improvement to the present invention, the calculation of the first... Contribution of secondary indicators , For the first The combined weight of the secondary indicators is Ω when the k-th secondary indicator corresponds to the j-th secondary indicator under the i-th primary indicator. k =ω ij =w i ·v ij ; This is the standardized value of the k-th secondary indicator;

[0036] Abnormal indicators are determined based on the standardized values ​​of the secondary indicators and the corresponding abnormality limits.

[0037] For a single abnormal indicator or multiple indicators that are abnormal at the same time, the preset indicator and problem mapping rules are invoked to determine the corresponding problem type and the location of the problem; and after sorting by indicator contribution, interface optimization suggestions containing preset problem types and adjustment actions are output.

[0038] To achieve the above-mentioned technical objectives, another technical solution adopted by the present invention is as follows:

[0039] A comprehensive evaluation system for redundancy in human-computer interaction interface operations based on interaction logs and path data includes:

[0040] The data acquisition module is used to acquire the interaction logs corresponding to the target task, the actual interface state path, one or more valid candidate paths, at least one of the functional branches or preset task rules corresponding to the actual interaction path to be evaluated, task stage information, functional object identifiers, reasonable repeated call rules, task target element attributes, and interface element attributes; wherein, at least one of the functional branches or preset task rules is used to distinguish different reasonable completion methods and determine the set of valid candidate paths to which the actual interaction path to be evaluated belongs; the valid candidate paths are task completion paths that meet the target task completion conditions, necessary state constraints, necessary interaction steps, and preset business or security constraints, and pass the log integrity verification.

[0041] The metric calculation module is used to calculate the extra operation ratio, invalid interaction rate, instant rollback rate, interface revisit rate, function retry rate, path deviation, semantic confusion, and visual interference. Specifically, based on at least one of the functional branches or preset task rules corresponding to the actual interaction path to be evaluated, a set of valid candidate paths is determined. Within this set, the path with the smallest total interaction step length is selected as the valid baseline path. When only one valid baseline path exists, it is used as the selected baseline path for metric calculation of the current task sample. When multiple valid candidate paths with the same minimum total interaction step length exist, all are used as valid baseline paths, and the path with the smallest deviation from the actual interface state path is selected as the selected baseline path for metric calculation of the current task sample. The total interaction step length of the selected baseline path is used as the baseline interaction step length, and the selected baseline path is used to calculate the extra operation ratio and path deviation of the current task sample. Within the same target task stage, the second and subsequent calls to the same functional object are identified as candidate events. After excluding the corresponding candidate events according to reasonable repeated call rules, the function retry rate is calculated.

[0042] The standardization module is used to align and standardize the eight secondary indicators.

[0043] The dimension aggregation module is used to classify the extra operation ratio and invalid interaction rate into the unnecessary redundancy first-level indicator, the instant rollback rate, interface revisit rate and function retry rate into the repetitive redundancy first-level indicator, the path deviation degree into the path deviation redundancy first-level indicator, and the semantic confusion degree and visual interference degree into the interference redundancy first-level indicator. The scores of the four first-level indicators are calculated according to the weight of the second-level indicators under each first-level indicator and the standardized value of the second-level indicators.

[0044] The comprehensive evaluation module is used to calculate the comprehensive redundancy index (CRI) based on the weights of the four primary indicators, the weights of the secondary indicators under each primary indicator, and the standardized values ​​of the secondary indicators.

[0045] The grade determination module is used to assign sample weights to the CRI values ​​of the reference samples and sort them according to the CRI values, and determine at least two grade boundaries based on the comprehensive redundancy index distribution in the reference sample set.

[0046] The diagnostic suggestion module is used to determine the indicator contribution by multiplying the standardized value of each secondary indicator with the weight of its corresponding secondary indicator combination, and to determine abnormal indicators based on the standardized value of the secondary indicator and the corresponding indicator abnormality limit. The abnormal indicators or combinations of abnormal indicators are matched with preset indicators and problem mapping rules, and the module outputs interface optimization suggestions containing preset problem types and adjustment actions according to the indicator contribution.

[0047] The beneficial effects of this invention are as follows:

[0048] (1) Construct a multi-dimensional evaluation system that comprehensively covers operational redundancy:

[0049] To address the shortcomings of existing evaluation schemes that often rely on task completion time, success rate, subjective scales, or localized indicators for a single interface, making it difficult to uniformly identify various redundancies from the perspective of interaction process and path structure, this invention establishes eight secondary indicators: additional operation ratio, invalid interaction rate, instant rollback rate, interface revisit rate, function retry rate, path deviation, semantic confusion, and visual interference. These are categorized into four primary evaluation dimensions: unnecessary redundancy, repetitive redundancy, path deviation redundancy, and disruptive redundancy. The first three indicators primarily characterize the user's operations and path results during task execution based on interaction logs and interface state paths. Disruptive redundancy characterizes semantic and visual factors that may induce invalid operations, rollbacks, retries, path deviations, or attention shifts based on interface element attributes. All eight indicators are weighted with positive values ​​and incorporated into the Comprehensive Redundancy Index (CRI), achieving a comprehensive quantitative evaluation of various redundancy behaviors in the interaction process and path structure.

[0050] (2) Unify the representation of interface state and interaction events to achieve cross-terminal and cross-modal applicability:

[0051] To address the shortcomings of some solutions that only apply to mouse or touch clicks, fail to cover interaction modes such as knobs and voice commands, and cannot distinguish between windows, modes, pop-ups, and operation states when using pages as path nodes, this invention uses interface states to uniformly represent various state types such as pages, windows, modes, pop-ups, and operation modes. This unifies page switching, mode switching, window switching, pop-up state changes, and operation mode changes as interface state transitions. Furthermore, it uses generalized interaction events to uniformly represent various interaction events such as clicks, touches, buttons, knobs, gestures, voice triggers, gaze confirmations, and controller inputs. Simultaneously, visual interference is uniformly calculated as non-target visual elements that are not directly related to the current task objective and possess characteristics of occlusion, motion, significant contrast, or continuous exposure. This makes the evaluation method uniformly applicable to mobile terminals, desktop software, in-vehicle human-machine interfaces, industrial control interfaces, smart screen interfaces, and virtual reality or augmented reality interactive interfaces.

[0052] (3) Adaptively determine the grade boundary based on the weighted quantile of the reference sample:

[0053] To address the shortcomings of existing solutions in determining grade boundaries by lacking an adaptation mechanism for interface categories, task categories, and sample distribution, this invention uses the weighted quantiles of reference samples to form grade boundaries. The reference samples are stratified according to interface category, task category, or a combination of both, and sample weights are assigned to different categories. These weights balance the sample distribution across different interface or task categories. The reference sample library can be updated at a preset period or when new valid samples are added, samples become invalid, or the category structure changes, and the grade boundaries are redefined, enabling the evaluation criteria to adaptively adjust according to actual application scenarios.

[0054] (4) The evaluation results are interpretable and operable:

[0055] This invention determines the contribution of each indicator by calculating the product of the standardized value of each secondary indicator and its corresponding combined weight, thus accurately locating key abnormal indicators that cause operational redundancy. It identifies abnormal indicators by comparing the standardized indicator values ​​with the abnormal indicator limits determined based on reference samples. Through a preset indicator-problem mapping rule, it matches abnormal indicators to specific problem types and outputs optimization suggestions containing problem types and adjustment actions according to the indicator contribution ranking. This makes the evaluation results interpretable and operable, providing clear improvement directions and specific suggestions for interface design optimization, and forming a design closed loop of "evaluation-diagnosis-optimization-re-evaluation".

[0056] In summary, the eight indicators of this invention reflect user operation and path results, interface semantics, and visual interference factors, forming four primary evaluation dimensions: unnecessary redundancy, repetitive redundancy, path deviation redundancy, and interference redundancy. All are weighted with positive secondary indicators in the CRI (Content Recognition Index). The invention uses interface state and generalized interaction to uniformly represent multiple terminals and multiple input modalities. The weighted quantiles of the reference samples form the level boundaries, balancing different interface or task categories through sample weights. Evaluation results can be traced back to abnormal secondary indicators, problem types, and optimization suggestions, facilitating the formation of a design closed loop. This transforms the previously difficult-to-quantify interface operation redundancy problem into a calculable, comparable, traceable, and optimizable quantitative evaluation result, providing a calculable basis for complexity identification, problem localization, and design optimization of various graphical human-computer interaction interfaces, effectively improving the comprehensiveness, adaptability, and practicality of interface usability evaluation. This invention is applicable to graphical human-computer interaction interfaces with identifiable interface states, interactive elements, and log recording capabilities, such as mobile terminals, desktop software, automotive applications, industrial control systems, smart screens, and virtual reality or augmented reality. Attached Figure Description

[0057] Figure 1 This is a complete flowchart of the operational redundancy comprehensive evaluation method of the present invention.

[0058] Figure 2This is a schematic diagram illustrating the generation of the weighted quantile threshold based on the reference sample in this invention.

[0059] Figure 3 This is a schematic diagram of the operational redundancy comprehensive evaluation system of the present invention. Detailed Implementation

[0060] The present invention will be further described below with reference to the accompanying drawings and embodiments. The following embodiments are used to illustrate the technical solution and not to limit the scope of protection.

[0061] First, this invention provides a comprehensive evaluation method for the redundancy of human-computer interaction interface operations based on interaction logs and path data. This invention is applicable to graphical human-computer interaction interfaces with identifiable interface states, interactive elements, and operation log recording capabilities, including but not limited to mobile terminal interfaces, desktop software interfaces, vehicle-mounted human-computer interfaces, industrial control interfaces, smart screen interfaces, and virtual reality or augmented reality interactive interfaces. Figure 1 As shown, the specific steps include:

[0062] Step S101: Obtain redundant indicator data:

[0063] For the task to be evaluated, the following information is obtained: task identifier, user or session identifier, interaction event sequence, timestamp, input modality, target element identifier, functional object identifier, target task stage identifier, interface state before and after interaction, initial call result, expected task state, feedback result, reasonable repeated call judgment rules, path attribution rules and their versions, at least one of the functional branches or preset task rules corresponding to the actual interaction path, whether the task is completed, the actual interface state path, one or more valid candidate paths that meet the target task completion conditions, necessary state constraints, necessary interaction steps and preset business or security constraints and pass log integrity verification, as well as interface element attributes. Among these, at least one of the functional branches or preset task rules corresponding to the actual interaction path is used to determine the set of valid candidate paths to which the actual interaction path belongs; the path attribution rules are used to specify how to distinguish different reasonable completion methods of the same target task based on the aforementioned functional branches or preset task rules. Target task stages can be divided according to sub-goals, necessary task steps, interface state combinations, or system process nodes; the same functional object can be determined according to the functional identifier, system function code, or the task target state it points to. Interface states can be determined by a combination of pages, windows, modes, pop-ups, operation modes, and interactive element tree summaries.

[0064] In this specification, the term "target task completion condition" refers to the preconditions or interaction context conditions used to distinguish different reasonable execution scenarios when executing the same target task in a human-computer interaction interface. These include at least one of the following: user permissions, interface operation mode, system or device operation status, interaction object type, and task execution environment. The term "necessary state" refers to the interface state or system state related to the interface interaction that the user operation process must reach, maintain, or pass through when executing the target task in a specific reasonable manner. The term "functional branch" refers to the different interaction path branches formed during the execution of the target task in the human-computer interaction interface due to different task sub-targets, interaction objects, parameter conditions, or function selections. The term "preset task rule" refers to the task execution rules pre-configured according to the task flow, functional logic, operation specifications, or security constraints of the human-computer interaction interface.

[0065] The path attribution rule is used to distinguish different reasonable completion methods under the same target task according to at least one of the above functional branches or preset task rules, to group valid candidate paths with the same task execution conditions and interaction path constraints into the same valid candidate path set, and to determine the valid candidate path set to which the actual interaction path to be evaluated belongs.

[0066] In the subsequent calculation formula, the symbol page represents the recognizable interface state; the symbol clk represents interactive events such as click, touch, button, knob, gesture, voice trigger, gaze confirmation, or controller input.

[0067] Step S102: Calculate the eight secondary indicators:

[0068] Eight secondary indicators were calculated: additional operation ratio, invalid interaction rate, instant rollback rate, interface revisit rate, function retry rate, path deviation, and semantic confusion. Visual interference was also calculated directly based on the unified attributes of non-target visual elements.

[0069] This implementation provides definitions, calculation formulas, and variable descriptions for eight indicators, with formula numbers used for reference between calculation steps. In the following sample aggregation formulas, u uniformly represents the u-th valid task sample within the evaluation unit, i.e., a verified user or session-task instance; N uniformly represents the total number of valid task samples within the evaluation unit. Valid task samples in this specification all satisfy the target task completion conditions, necessary state constraints, necessary interaction steps, and preset business or security constraints, and pass log integrity verification. Necessary constraints, as referred to in this specification, include necessary state constraints, necessary interaction steps, and preset business or security constraints; condition and constraint verification refers to verifying the target task completion conditions and the aforementioned necessary constraints; log integrity verification is performed separately. The set of valid candidate paths, as referred to in this specification, is a set formed by aggregating valid candidate paths corresponding to the same reasonable completion method according to functional branches or preset task rules; the set of valid candidate paths to which the actual interaction path to be evaluated belongs is determined according to at least one of the functional branches or preset task rules corresponding to the actual interaction path.

[0070] 1. Additional operations ratio:

[0071] The actual interaction step size L for completing the task based on the u-th valid task sample. u Total interaction step size with the selected effective baseline path used for calculating the current task sample metrics. Calculate the extra step size and its relative proportion, and then perform boundedness (as in formula (3), mapping the non-negative result to...). And calculate the mean of the N valid samples, as follows:

[0072] Formula for calculating the single-sample extra operation ratio:

[0073] (1);

[0074] (2);

[0075] Where, ΔL u For additional interaction steps; π u This is the actual interaction path sequence for this task sample; π ref,u L is the selected effective baseline path operation sequence used for calculating the metric of the u-th task sample; u =|π u | represents the total step length of the actual interaction path for this task sample; =|π ref,u | is the total interaction step size of the selected effective benchmark path used for calculating the metric of the u-th task sample.

[0076] Formula for single-sample normalized extra operation ratio:

[0077] (3);

[0078] in The purpose of using ∈[0,1) is to uniformly constrain the index values ​​within the [0,1) range, making it easier to conduct comprehensive evaluations with other dimensional indicators.

[0079] The normalized extra operation ratio for N valid task samples within the evaluation unit is calculated as follows, yielding the formula for calculating the normalized extra operation ratio of the evaluation unit:

[0080] (4);

[0081] in This is the index variable for the summation process, used to refer to the first... Group samples; The condition for summation is defined as the starting condition, indicating that the summation operation begins with the first set of samples. Take 1, 2, 3... in sequence until... ; The total number of valid task samples. The normalized extra operations ratio at the population level is obtained by using an arithmetic mean.

[0082] The index calculation relies on two types of operation path sequence data: selected valid baseline path data for the current task sample index calculation and actual interaction path data for valid task samples; the selected valid baseline path operation sequence for the u-th task sample index calculation is π. ref,u Its total interaction step size is L ref,u =|π ref,u |;The actual interaction path sequence of the effective task sample is π u Its total interaction step size is L u =|π u |

[0083] A valid candidate path is a task completion path that meets the target task completion conditions and necessary constraints and passes log integrity verification. For the same target task with different reasonable completion methods, valid candidate paths are distinguished according to functional branches or preset task rules, forming corresponding valid candidate path sets. For the u-th valid task sample, based on the functional branch or preset task rules corresponding to its actual interaction path, its valid candidate path set is determined, and the path with the smallest total interaction step length within this set is selected as the valid baseline path. When only one valid baseline path exists, this valid baseline path is used as the selected valid baseline path for the current task sample's index calculation. When multiple valid candidate paths with total interaction step lengths equal to the minimum value exist, all are used as valid baseline paths, and the path with the smallest deviation from the actual interface state path of the task sample is selected as the selected valid baseline path for the current task sample's index calculation. The total interaction step length of the selected valid baseline path is used as the baseline interaction step length. The actual interaction path of this task sample meets the target task completion conditions and necessary constraints corresponding to its valid candidate path set and passes log integrity verification; therefore, its total step length is not less than the baseline interaction step length, and the additional interaction step length is a non-negative value. The set of valid candidate paths, path attribution rules, and valid benchmark paths are stored after independent verification and version management; the actual paths to be evaluated are not automatically written into the set of valid candidate paths and the valid benchmark path library due to their short step size.

[0084] 2. Invalid interaction rate:

[0085] Invalid interactions are identified that do not cause a change in the effective interface state, do not generate a system acceptance flag, and do not generate task-related feedback. They are calculated as the sum of the number of invalid interactions and the number of redundant valid interactions, as follows:

[0086] Regarding operational-related indicators, For the first The actual operation sequence of a user-task sample. For the sample A collection of all operation events (including operation controls, operation behaviors, and other information).

[0087] Regarding interaction-related metrics, For the sample A collection of invalid interactive events. For the sample Invalid interaction count

[0088] For any interactive event If the system does not generate effective feedback after the interaction, then it is determined that... .

[0089] Single-sample invalid interaction rate:

[0090] (5);

[0091] in This refers to the number of unique interactions for this user in this task, retained only within the valid path (steps not involved in any backtracking or revisiting loops) after global denoising and cleaning (removing all redundant segments). In other words, if a step is deemed redundant due to being in a backtracking / revisiting loop, all interactions generated within that step (regardless of their validity) are not counted. This value represents the theoretically minimum necessary interaction cost for a user to complete the current task. After reasonable duplication is determined, the excluded reasonable duplicate calls are not considered redundant calls and are retained in the valid interaction count; only the net function retry calls after excluding reasonable duplicates are treated as redundant interactions.

[0092] The invalid interaction rate of the evaluation unit is calculated as follows for the N valid task samples within the evaluation unit:

[0093] (6);

[0094] in For sample index variables, The total number of valid task samples. The invalid interaction rate at the group level is obtained by using an arithmetic mean.

[0095] Metric calculations rely on two types of data: operation logs and timestamp data. These include interaction event records (containing interaction objects / areas, interactivity, system feedback results, etc.); undo / rollback / cancellation event records and their timestamps; and data used to construct... , And calculate .

[0096] 3. Immediate rollback rate:

[0097] Identification based on the actual user interface state sequence The instant rollback segment, together with the effective interface state transition, constitutes the ratio, calculated as follows:

[0098] Let the first The sequence of interface state transitions that occurs for each sample during the task is as follows:

[0099] (7);

[0100] In equation (7), p u,t This represents the interface state of the u-th valid task sample after the t-th interface state transition, where t=0,…,J u J uThis represents the total number of interface state transitions for this task sample. After global denoising and cleaning (removing all redundant segments and their associated transitions), the remaining valid path has the following number of transitions. This means the number of valid steps is reduced by 1. Any intermediate steps involved in a backtrack loop (such as B in A→B→A) are discarded and do not participate in the valid jump count.

[0101] Formula for calculating the instantaneous regression rate for a single sample:

[0102] (8);

[0103] in, For sequence Round trip jump occurs ( ) sequence, For round-trip jump sequences ( The number of interface state transitions. When multiple backflip segments overlap, nest, or point to the same interface state transition, first obtain the union of the transition positions involved in all backflip segments, and then count according to different transition positions. The same transition is counted only once.

[0104] The instant rollback rate of the N valid task samples within the evaluation unit is calculated as follows, yielding the formula for calculating the instant rollback rate of the evaluation unit:

[0105] (9);

[0106] in For sample index variables, The total number of valid task samples. The immediate backoff rate at the group level is obtained by using an arithmetic mean.

[0107] Metric calculation relies on the complete sequence of all interface state transitions during the execution of the user task. This includes the state codes of the starting and target interfaces for each jump, used to identify and count all ( () type of round-trip mode.

[0108] 4. Interface revisit rate:

[0109] Record the times when the user accesses each interface state in the actual path, and accumulate the revisit time of states with an interval of more than 2 between adjacent access locations. The specific calculation is as follows:

[0110] Let the first The interface state access sequence for each sample (user-task instance) is as follows:

[0111] (10);

[0112] Where, χ u,νLet R be the interface state encoding for the νth access in the access sequence of sample u. u =|V u page | represents the total number of visits. The interface state trajectory in equation (7) is numbered starting from 0 according to the jump sequence number, and the access sequence in equation (10) is numbered starting from 1 according to the access order. Therefore, R u =J u +1, and χ u,ν =p u,ν−1 In the entire text, ν only represents the position number of the access sequence.

[0113] In this task, after global noise reduction and cleaning, and removal of path segments deemed redundant and their associated state transitions, the user obtained the following valid interface state access sequence: The total number of times the interface status was accessed under this path is:

[0114] (11);

[0115] For any interface state p, define its set of access locations in sample u as Φ. u (p), and sorted by access location:

[0116] (12);

[0117] in < <⋯< , For the interface state p, the a-th access in V u page In the text, V represents the total number of times this interface state is accessed in the current sample, and a = 1, ..., V. Throughout the text, 'a' only indicates the access order of the same interface state.

[0118] To distinguish between immediate rollback (A→B→A, continuous access behavior), we only calculate the case where the number of visits between two adjacent visits is greater than 2. The total number of revisit loop steps for interface p is defined as follows: when the same interface is visited at least twice, it is calculated according to equation (13); where, if V>2, equation (13) is summed continuously for multiple adjacent access pairs; the total number of visits between each two adjacent visits (including the number of visits to interface p in the second visit):

[0119] (13);

[0120] Here, 1(·) is an indicator function (takes 1 if the condition is true, otherwise takes 0), (τ a+1 −τ a ) represents the number of visits experienced from the a-th visit (excluding) to the (a+1)-th visit (including) of interface state p.

[0121] When the same interface is accessed only twice and the position difference between the two accesses is greater than 2, the revisit is counted according to formula (13); when a new access occurs later, only the newly formed adjacent access pair is judged, and the counting result of the previous access pair is not traced back and changed.

[0122] Let the sample The total number of redundant revisits is the sum of the number of revisits for all interfaces:

[0123] (14);

[0124] Formula for calculating the single-sample interface revisit rate:

[0125] (15);

[0126] The interface revisit rate of the N valid task samples within the evaluation unit is calculated as follows, yielding the formula for calculating the interface revisit rate of the evaluation unit:

[0127] (16);

[0128] in For sample index variables, The total number of valid task samples. The interface revisit rate at the group level is obtained by using an arithmetic mean.

[0129] This metric is calculated based on the interface state access sequence V of the u-th valid task sample. u page Φ is the set of access locations for each interface state. u (p) and the total number of visits V for the current interface state p.

[0130] 5. Function retry rate:

[0131] The function call sequence within the same target task phase is statistically analyzed. The second and subsequent calls to the same function object are first identified as candidate events for function retry. Then, events such as increases or decreases in the number of calls, continuous parameter adjustments, pagination, scrolling, text editing, and other reasonable repeated events that need to be called repeatedly according to task rules are excluded. The remaining candidate events are used as the net function retries and together with the number of effective interactions, they constitute the ratio. The specific calculation is as follows:

[0132] The sequence of all operation events for the u-th valid task sample can be represented as:

[0133] (17);

[0134] Among them, c u,oFor the identifier of the 0th operation, each operation must be associated with at least the function object identifier, the target task stage identifier, a reasonable repetition type, the initial call result, the expected task status, and feedback attributes; fun(c u,o ) represents the functional object corresponding to this operation, stage(c u,o ) indicates the target task phase to which the operation belongs; L u clk =|c u | represents the total number of operations. The 'o' in the text only indicates the position of the operation event in the operation sequence.

[0135] For any target task stage μ and its functional interaction object f, the number of calls within a stage is defined as:

[0136] (18);

[0137] Within the same target task phase, the second and subsequent calls to the same functional object form candidate counts; then, based on the task rules, the number of these counts that are reasonable duplicate calls is determined, resulting in the net functional retries count after excluding reasonable duplicates.

[0138] (19);

[0139] Formula for calculating single-sample functional retry rate:

[0140] (20);

[0141] Where u is the index of valid task samples, μ is the index of the target task stage, and f is the functional object; M u Let F be the set of target task stages contained in sample u; u,μ Let n be the set of functional objects that appear in sample u during the target task phase μ; cand (u,μ,f) represents the number of function retry candidates formed by sample u making a second and subsequent call to function object f within the target task phase μ; n rr (u,μ,f) represents the number of times that are determined to be reasonable repeated calls according to the task rules from among the candidate calls; n rep (u,μ,f) represents the net number of function retries after excluding reasonable duplicate calls; N rep (u) represents the sum of net functional retries obtained by aggregating sample u across all target task stages and functional objects; n valid-clk,u This is the number of deduplicated interactions retained in the valid path after global denoising and cleaning of the task sample. Reasonable repeated calls will continue to be counted in this valid interaction count.

[0142] The function retry rate for N valid task samples within the evaluation unit is calculated as follows, yielding the formula for calculating the function retry rate of the evaluation unit:

[0143] (twenty one);

[0144] in For sample index variables, This represents the total number of valid task samples. The functional retry rate at the population level is obtained by using an arithmetic mean.

[0145] when A high value indicates that, after excluding reasonable duplicate calls required for the task, a large number of operations are still repeatedly calling the same functional object within the same target task stage. This may mean that system feedback is not obvious, changes in the functional object's state are not easily perceived, operation results are uncertain, or the operation path is interrupted. When this value is low, it indicates that, after excluding reasonable duplicates, there are fewer unnecessary duplicate calls to the same functional object within the same target task stage. Whether the first call is successful, whether the expected task state is achieved, and whether the feedback is timely and clear are only used for diagnosing the reasons for functional retry and are not used as criteria for identifying candidate events or net functional retry counts.

[0146] This metric is calculated based on user task interaction logs, functional object identifiers, target task stage identifiers, and reasonable repeated call rules. It also stores the initial call result, expected task status arrival, system response time, and diagnostic attributes such as the timeliness and clarity of feedback. The sample to be evaluated uses the same task stage division, functional object identification, and reasonable repeated call exclusion rules as the reference sample.

[0147] 6. Path deviation:

[0148] The path difference is measured by the edit distance between the actual interface state sequence and the interface state sequence of the selected effective baseline path used for calculating the current task sample index. The path difference is normalized and bounded by the number of interface state jumps contained in the interface state sequence of the selected effective baseline path (normalization, such as: Equation (26) using the number of interface state jumps of the selected effective baseline path). The edit distance is normalized and bounded, as shown in equation (27). The monotonic transformation maps the normalized path deviation to the interval When there are multiple valid baseline paths in the valid candidate path set to which the current task sample belongs, all of which have a total interaction step length equal to the minimum value, calculate the path deviation of the actual interface state sequence relative to the interface state sequence of each valid baseline path and take the minimum value. Select the valid baseline path corresponding to the minimum value as the selected valid baseline path for the index calculation of the current task sample. The specific calculation is as follows:

[0149] Let the first The interface state transition sequence for each sample (user-task instance) is as follows:

[0150] (twenty two);

[0151] in Indicates the first The interface state encoding after the next jump. This is the number of times the interface state transitions, therefore:

[0152] (twenty three);

[0153] In this specification, the vertical bar symbol |·| uniformly represents the number of elements in the sequence or the cardinality of the set. For the interface state sequence shown in equation (22), the first state node is the jump start point, and the number of state nodes is equal to the number of interface state jumps plus 1. Therefore, equation (23) is written as | |=J u +1.

[0154] When multiple valid baseline paths with the minimum total interaction step size exist within the valid candidate path set to which the current task sample belongs, the path that minimizes the actual path deviation is selected as the chosen valid baseline path for calculating the current task sample's metrics. The interface state sequence of this selected valid baseline path is as follows:

[0155] (twenty four);

[0156] Among them, J ref,u =| |-1 is the interface state transition step size for the selected valid benchmark path used for calculating the u-th task sample index; This represents the selected valid baseline path interface state sequence used for calculating the metric of the u-th task sample. When there are multiple valid baseline paths in the set of valid candidate paths to which the task sample belongs, all with a total interaction step size equal to the minimum value, this represents the interface state sequence corresponding to the path that minimizes the deviation of the actual path. The subscripts ref, u uniformly represent the valid baseline path used for calculating the metric of the u-th task sample. and These represent the operation sequence and interface state sequence for the same valid baseline path, respectively.

[0157] The minimum edit distance (LevenshteinDistance) is defined as:

[0158] (25);

[0159] Indicates will Convert to Minimum number of editing operations (insert, delete, replace) required.

[0160] Formula for calculating single-sample path deviation:

[0161] (26);

[0162] To constrain the indicator values ​​to the [0,1) range and facilitate comprehensive evaluation with other dimensional indicators, the formula for calculating the single-sample normalized path deviation is as follows:

[0163] (27);

[0164] The normalized path deviation of the N valid task samples within the evaluation unit is calculated as follows, yielding the formula for calculating the normalized path deviation of the evaluation unit:

[0165] (28);

[0166] in For sample index variables, This represents the total number of valid task samples. The overall path deviation level at the group level is obtained using an arithmetic mean.

[0167] When the path deviation is high, it indicates that the user has deviated significantly during task execution, and the interface may have problems such as a chaotic information architecture, insufficient guidance, and unclear task entry.

[0168] 7. Semantic Confusion:

[0169] The semantic confusion is determined based on the semantic similarity between the target interactive element and the non-target interactive element and the number of non-target semantic interference elements. The semantic similarity is obtained based on at least one of text labels, icon semantic labels, function identifiers, system function codes, accessibility names or operation descriptions, and is determined by at least one of preset semantic relationships, vector similarity, semantic model output values ​​or pre-calibrated semantic scores.

[0170] Semantic confusion is used to characterize the degree of functional semantic proximity between the target interactive element and non-target interactive elements in the same interface state, and the degree of potential competition they form. First, the semantic attributes of the target and non-target interactive elements are obtained. These semantic attributes include at least one of text labels, function names, icon semantic labels, system function codes, accessibility names, and operation descriptions. For any non-target interactive element, a single-element semantic confusion value is obtained based on its semantic similarity to the target interactive element. The semantic similarity can be determined by at least one of the following methods: a preset semantic relationship matrix, text or label vector similarity, function code association, semantic model output value, or a pre-calibrated expert semantic score. In one embodiment, a five-level semantic relationship score is used to pre-calibrate the semantic relationship between the target and non-target interactive elements, where 1 indicates complete irrelevance and 5 indicates approximately equivalent functional meaning. The score is normalized to the range of 0 to 1. Further, the semantic confusion of the current interface state is determined by combining the number of non-target semantic interference elements and the saturation coefficient of the number of semantic interference items, and then aggregated within the relevant interface states traversed by the target task. The specific calculation is as follows:

[0171] For the d-th semantic interference element (d=1,…,Q) on interface p p The semantic relationship score between the element and the target interactive element is defined as follows:

[0172] (29);

[0173] Correspondence:

[0174] 1: Completely irrelevant;

[0175] 2: Slightly related (same field but different tasks);

[0176] 3: Moderately related (same task chain but different steps);

[0177] 4: Highly related (likely to be mistaken for the same function);

[0178] 5: Almost equivalent / alias (same function, different names);

[0179] Linearly map it to :

[0180] (30);

[0181] Let Q be the number of semantic interference elements on interface p. p The average semantic similarity is denoted as Specifically:

[0182] (31);

[0183] When Q p When the average semantic similarity is 0, the semantic confusion degree D of the interface state is set to 0. sem (p)=0; when Q p When the value is greater than 0, calculate according to formula (31).

[0184] Let Q be the number of semantic interference elements on interface p. p The maximum number of interferences within a reasonable interface is Q. max Then the saturation coefficient κ(p) for the number of semantic interference terms is:

[0185] (32);

[0186] Single-interface semantic obfuscation:

[0187] (33);

[0188] sample The set of all interfaces involved in the target operation during the task is as follows: Then, the semantic confusion of a single sample is:

[0189] (34);

[0190] The semantic confusion of the N valid task samples within the evaluation unit is calculated as follows to obtain the semantic confusion of the evaluation unit:

[0191] (35);

[0192] in For sample index variables, The total number of valid task samples. The average semantic similarity of single interaction elements at the group level is obtained by using an arithmetic mean.

[0193] 8. Visual interference:

[0194] Visual interference degree is used to characterize the degree to which interface elements unrelated to the current task objective interrupt, obstruct, or divert attention during the execution of a target task, due to features such as spatial occlusion, foreground occupancy, dynamic changes, significant visual contrast, or continuous exposure, thereby affecting the acquisition of target information, judgment of task status, and interactive operations. This indicator directly uses non-target visual elements as a unified evaluation object and comprehensively characterizes the spatial and dynamic interference generated by these elements.

[0195] Interference features generated by non-target visual elements include spatial interference features and dynamic interference features. Spatial interference features are used to characterize the visual interference caused by non-target visual elements due to foreground hierarchy, visible area, spatial relationship with the target area, and degree of occlusion of the target area; dynamic interference features are used to characterize the attentional attraction caused by non-target visual elements due to displacement, scrolling, flickering, size changes, brightness or contrast changes, and continuous dynamic presentation.

[0196] The degree of non-target visual interference is determined jointly by the two types of features mentioned above. The spatial interference feature and the dynamic interference feature are internal calculation features of the same non-target visual interference index and are not used as independent evaluation indicators.

[0197] In one implementation, the interference contribution of a single non-target visual element is determined based on the element area factor, spatial location factor, and interference morphology intensity factor formed by the degree of occlusion, motion salience, visual contrast, or continuous exposure, and the non-target visual interference degree is obtained by aggregating the interference within the interface state and task sample range.

[0198] For the first The first task sample For each interface state, first identify the visible elements within that state; then, include elements that are not directly related to the current task objective and possess at least one of the following characteristics: occupies the target area, occupies the foreground layer, generates motion attraction, forms significant visual contrast, or is continuously exposed, into the non-target visual element set. The calculation process is as follows.

[0199] make For the first The task sample in the th... The set of non-target visual elements in each interface state, the number of elements in the set is:

[0200] (36);

[0201] In this equation (36), the left side represents the number of non-target visual elements; when the set is empty, the number of elements is 0.

[0202] For sets any element in The area factor is calculated based on the proportion of the visible area of ​​the element to the visible area of ​​the current interface.

[0203] (37);

[0204] in, Let e ​​be the visible area of ​​the non-target visual element e in the current interface state. The area of ​​the visible region in the current interface state, and the area factor. It is located between 0 and 1.

[0205] Based on the degree of interference that the element's location causes to the current task target area or main task information area, assign spatial location weights to the elements and normalize them:

[0206] (38);

[0207] in, The spatial position weight of the non-target visual element e. This represents the upper limit of the spatial position weight; the closer an element is to, covers, or occupies the current task target area, the higher the spatial position factor. It can be configured to a larger value, and is between 0 and 1.

[0208] Determine and normalize the intensity of the interfering morphology of the element based on at least one of the following: degree of occlusion, motion salience, visual contrast, or degree of sustained exposure.

[0209] (39);

[0210] in, The intensity of the interference shape of the non-target visual element e. The upper limit of the interference morphology intensity, the interference morphology intensity factor It is located between 0 and 1. When the same element has multiple interference characteristics, the comprehensive intensity value can be obtained according to preset rules.

[0211] Multiplying the area factor, spatial location factor, and interference morphology intensity factor yields the visual interference contribution of a single element:

[0212] (40);

[0213] The left side of equation (40) represents the element The contribution of visual interference is determined by three factors: area, location, and interference morphology.

[0214] Sum the contributions of all non-target visual elements in the current interface state and average them by the number of elements:

[0215] (41);

[0216] The left side of equation (41) represents the average visual interference contribution of the elements. The denominator is the larger of the number of elements and 1; when the set is empty, the sum of the sets and the average contribution are both taken as 0 to avoid division by zero.

[0217] Considering that the marginal interference effect of non-target visual elements tends to saturate after the number of non-target visual elements increases to a certain extent, the saturation coefficient of the number of non-target visual elements is calculated:

[0218] (42);

[0219] in, This is a preset saturation upper limit for the number of non-target visual elements in a single interface state, and >0; when ≥ At that time, the saturation coefficient of the number of non-target visual elements Take 1; the coefficient on the left side of equation (42) is between 0 and 1, and only reflects the number of non-target visual elements.

[0220] Multiply the saturation coefficient of the number of non-target visual elements by the average visual interference contribution of the elements to obtain the visual interference level of the current interface state:

[0221] (43);

[0222] The left side of equation (43) represents the visual interference level of the current interface state, with a value between 0 and 1. When there are many non-target visual elements and the average visual interference contribution of these elements is high, the visual interference level of the interface state increases accordingly.

[0223] For the first The visual interference level of a task sample is obtained by averaging all valid interface states in the sample.

[0224] (44);

[0225] in, This is the sequence number of the valid interface state record or sampling position in the u-th task sample. This represents the number of valid interface state records or the total number of sampling locations for this task sample. =1,…, ,and ≥1; The denominator of equation (44) is the number of valid interface state records or sampling positions for the task sample, and its value is not less than 1. The same interface state sampling rules are used for the same evaluation batch; when it is necessary to reflect the state dwell time, sampling is performed at fixed time intervals so that states with longer durations correspond to more sampling positions. In the whole text, η only represents the valid interface state records or sampling positions in the calculation of visual interference.

[0226] For the evaluation unit The visual interference level of the evaluation unit is obtained by calculating the arithmetic mean of the visual interference levels of each valid task sample.

[0227] (45);

[0228] in, For task sample indexing, The number of valid task samples is not less than 1; the left side of equation (45) is the original index value of visual interference degree for subsequent standardization and comprehensive redundancy index (CRI) calculation. The visual interference degree at the group level is obtained by using the arithmetic mean method.

[0229] When the evaluation unit contains multiple valid task samples, i.e. When the result is greater than 1, the result at the group level is the aggregated result of that evaluation unit; when the evaluation unit contains only one valid task sample, that is... An expression equal to 1 still belongs to the evaluation unit, but is not referred to as the group level.

[0230] Step S103: Standardize the eight indicators:

[0231] All eight indicators were aligned in the direction that larger values ​​indicate greater operational redundancy. Regarding the first... The indicator is obtained from the lower reference quantile of a reference sample similar to the target task. and upper reference quantile Perform truncated linear standardization:

[0232] (46);

[0233] in, This is the original value of the k-th indicator. For standardized values, ε is a positive number to prevent the denominator from being zero. The lower and upper reference quantiles can be taken as the 5% and 95% quantiles of the reference distribution, respectively, or configured according to the sample size. Here, clip(x,0,1) indicates that x is truncated to the closed interval [0,1]. Throughout the text, k uniformly represents the global index of the eight secondary indicators, k=1,…,8.

[0234] Step S104: Generate scores for the four primary indicators:

[0235] The four primary indicators are unnecessary redundancy, repetitive redundancy, path deviation redundancy, and disruptive redundancy. The number of secondary indicators included in each of the four primary indicators is as follows: , , and The weights can be determined by expert importance scores, data-driven calibration of historical evaluation samples, preset weight schemes for different interface types, or a combination of the above methods. The following explanation uses the mean normalization of expert importance scores as an example.

[0236] The first three items mainly reflect user operations and path results, while interference redundancy mainly reflects interface semantics and visual interference factors. All four items are used together to evaluate the comprehensive degree of operational redundancy of a specific interface under the target task.

[0237] Let the total number of experts be H, and the importance score of the h-th expert for the i-th primary indicator be . First, calculate the average expert score for the i-th primary indicator. Then, normalize the average scores of the four primary indicators to obtain the weights of the primary indicators. (i.e. Figure 1 (Inter-layer weights in the text). In the full text, H represents only the total number of experts, and h represents only the expert's serial number.

[0238] (47);

[0239] (48);

[0240] (49);

[0241] (50);

[0242] (51);

[0243] (52);

[0244] (53);

[0245] In equation (47), The expert average importance score for the i-th primary indicator; in equation (48) This is a dummy index for the normalized summation of only the first-level indices. In equation (49), The expert average weight score of the j-th secondary indicator under the i-th primary indicator; in equation (50) This is a dummy index that is only used for the normalized summation within this primary indicator. Let be the number of secondary indicators contained in the i-th primary indicator, where =2、 =3、 =1、 =2.

[0246] in, The weighted score given by the h-th expert to the j-th secondary indicator under the i-th primary indicator. The weights of secondary indicators within the same primary indicator layer (i.e.) Figure 1 (intra-layer weights in the middle). The weights of the secondary indicator combinations are 1, and the sum of the weights of all secondary indicator combinations is 1. The standardized value of the j-th secondary indicator under the i-th primary indicator obtained in step S103. The scores are then assigned to the corresponding primary indicators. This results in the scores for the four primary indicators: F1, F2, F3, and F4.

[0247] Step S105: Calculate the Comprehensive Redundancy Index (CRI):

[0248] The Comprehensive Redundancy Index (CRI) is calculated based on the primary weights of the four primary indicators and the combined weights of the secondary indicators. The hierarchical structure and the expanded formulas of the eight secondary indicators are as follows:

[0249] (54);

[0250] (55);

[0251] In the expansion, , , , , , , and These represent the standardized values ​​for the additional operation ratio, invalid interaction rate, instant rollback rate, interface revisit rate, function retry rate, path deviation, semantic confusion, and visual interference, respectively. , , , , , , and The weights are the combined weights of the corresponding secondary indicators, in that order. When all expert scores are positive, the combined weights of all secondary indicators are greater than 0, and therefore all eight secondary indicators are included in the CRI.

[0252] The weighting of the secondary indicator combination referred to in this specification is the product of the weight of the primary indicator and the weight of the secondary indicator within the primary indicator layer. The global index k and the hierarchical index (i,j) of the eight secondary indicators adopt a fixed mapping: k=1↔(i,j)=(1,1), k=2↔(i,j)=(1,2), k=3↔(i,j)=(2,1), k=4↔(i,j)=(2,2), k=5↔(i,j)=(2,3), k=6↔(i,j)=(3,1), k=7↔(i,j)=(4,1), k=8↔(i,j)=(4,2). Under this fixed mapping, z k With z ij Ω represents the standardized value of the same secondary indicator in both global and hierarchical forms.k =ω ij =w i ·v ij This indicates the same combined weight for the secondary indicator under both numbering formats. All eight indicators have calculable values ​​and their corresponding combined weights are greater than 0; when an indicator is applicable but the observation result is zero, the indicator value is set to 0 and it is still included in CRI; records where any indicator cannot be reliably calculated due to missing data, collection errors, or unrecognizable interface status are not considered valid evaluation samples.

[0253] Step S106: Determine the level using the weighted quantile method:

[0254] When establishing the reference sample library, a single valid execution record of a single user or a single session for a target task is used as the reference sample. The reference samples are stratified according to interface category, task category, or a combination of both, and each stratum is designated as a category. Interface categories may include mobile terminal interfaces, desktop software interfaces, vehicle-mounted human-machine interfaces, industrial control interfaces, smart large screen interfaces, and virtual reality or augmented reality interactive interfaces; task categories may include information query, content browsing, parameter setting, information entry, transaction confirmation, navigation operation, and monitoring and handling.

[0255] The above-mentioned single user or single session's one valid execution record corresponds to a single evaluation mode. In the group evaluation mode, a preset number of valid execution records are aggregated in the same way and with the same sample weight to form an evaluation unit; the sample to be evaluated and each reference sample use the same evaluation unit, and the number of valid execution records contained in each reference evaluation unit is the same as that in the unit to be evaluated.

[0256] Reference sample data can be derived from anonymous interaction logs generated during actual operation, controlled evaluation records, task operation records of prototype or simulation systems, and authorized and verified historical evaluation records. Each candidate sample collects the following: category identifier, task identifier, anonymous user or session identifier, interaction event sequence, timestamp, input modality, functional object identifier, target task stage identifier, interface state before and after interaction, initial call result, expected task state, system feedback, reasonable repeated call rules and their versions, path attribution rules and their versions, at least one of the following: functional branch or preset task rules corresponding to the actual interaction path, valid candidate path set identifier, valid candidate path set identifier to which the actual interaction path belongs, valid baseline path identifier used for calculation of the current sample, task completion status, actual interface state path, one or more valid candidate paths that meet the target task completion conditions, necessary state constraints, necessary interaction steps and preset business or security constraints and pass log integrity verification, verification information of target task completion conditions and necessary constraints, log integrity verification information, target element attributes and interface element attributes.

[0257] For each candidate sample, the same data preprocessing rules, task stage division, functional object identification, reasonable duplicate call exclusion rules, effective candidate path verification rules, effective candidate path set division rules, actual interaction path set attribution rules, indicator definitions, standardized parameters, and weight calculation processes as the samples to be evaluated are applied. The extra operation ratio, invalid interaction rate, immediate rollback rate, interface revisit rate, function retry rate, path deviation, semantic confusion, and visual interference are calculated sequentially, and the CRI is further calculated. Duplicate records, records with missing key fields, abnormal event timing, unrecognizable interface states, or records that cannot complete indicator calculations due to collection errors are removed. The sample number, category identifier, eight indicator values, standardized indicator values, CRI value, calculation parameter version, reasonable duplicate call rule version, effective candidate path set version, path attribution rule version, set identifier, effective baseline path identifier, effective candidate path version, and entry time are associated and stored for those that pass the integrity, consistency, and computability verification.

[0258] When the group sample is used as the evaluation unit, the single sample index is calculated for each valid execution record in the evaluation unit. Then, the eight indicators at the group level are obtained by aggregation according to the formulas (4), (6), (9), (16), (21), (28), (35) and (45). The CRI calculated from the group index is stored as a reference value.

[0259] For categories containing valid samples Configure normalized class weights The sum of the weights for each category is 1. When no prior category weights are configured, each category receives equal weight. =1 / G; When data on actual usage frequency or task importance is available, the corresponding proportion can be normalized and used as the category weight. Let the categories be... The number of valid samples in the sample is The category weights are evenly distributed among the samples of that category, and the weight of the r-th reference evaluation unit is denoted as α. r This makes the sum of the weights of all samples in that category equal to... The specific calculations are as follows:

[0260] (56);

[0261] (57);

[0262] (58);

[0263] Where r represents only the number of the reference evaluation unit before sorting, α rLet g(r) be the weight of the r-th reference evaluation unit, and g(r) be its category; ℓ represents only the position after sorting by CRI value from smallest to largest, and α is the weight of the r-th reference evaluation unit. (ℓ) y represents the weight of the ℓth reference evaluation unit after synchronous rearrangement. (ℓ) For the ℓth CRI reference value after sorting, A ℓ λ represents the normalized cumulative weight after sorting; λ in equation (57) is only the dummy index used for summing when accumulating sorted weights. w (·) represents the weighted quantile function, G is the total number of categories containing valid samples, and R is the total number of valid reference evaluation units participating in the calculation. K only represents the preset redundancy level number, b only represents the level boundary number, and θ b Let θ1, θ2, and θ3 be the boundary of the b-th CRI level. When K=4, θ1, θ2, and θ3 are the weighted 25%, 50%, and 75% quantiles, respectively.

[0264] (59);

[0265] Level represents the redundancy level of the unit to be evaluated, determined according to equation (59). Levels 1 to 4 can be named low redundancy, mild redundancy, high redundancy, and severe redundancy, respectively. Figure 2 As shown.

[0266] The reference sample library can be updated according to a preset period, or when new valid samples are added, samples become invalid, or the category structure changes. During an update, the new records undergo a collection, calculation, cleaning, and storage process, and the number of valid samples and sample weights for each category are re-determined. When the indicator algorithm, standardization parameters, or weight parameters change, the eight indicators and CRI of the relevant samples are recalculated according to the same parameter version, and the weighted quantiles and rank boundaries are updated accordingly.

[0267] Step S107: Problem Diagnosis and Optimization Suggestion Output:

[0268] Based on the k-th standardized index value Combined weights with corresponding secondary indicators Calculate the contribution value of the indicator = · The sum of the contribution values ​​of each indicator satisfies Σ k =CRI; When the k-th secondary indicator corresponds to the j-th secondary indicator under the i-th primary indicator. , and Σ k =1. The abnormal limits for each indicator use the same reference sample and sample weights as in step S106, based on the standardized value of the k-th indicator. The reference distribution is calculated to obtain the weighted quantiles corresponding to the preset abnormal quantile levels; Indicators that reach or exceed the corresponding abnormal thresholds are identified as abnormal indicators. These abnormal thresholds are used to identify specific abnormal indicators. The CRI level boundary in step S106 is determined based on the reference distribution of the Comprehensive Redundancy Index (CRI) and used to determine the overall redundancy level. Both are determined independently based on the distribution of individual standardized indicators and the comprehensive CRI distribution, respectively. For a single abnormal indicator or multiple indicators that are simultaneously abnormal, a preset indicator-problem mapping rule is invoked to determine the corresponding problem type and location. For functional objects with abnormal retry rates, attributes such as the initial call result, expected task status arrival status, system response time, and whether feedback is timely and clear are further read to determine possible causes of functional retries. These attributes are used for cause diagnosis and optimization suggestion matching, but not for determining candidate events for functional retries or the net number of functional retries.

[0269] The location of the problem includes at least one of the following: interface state, interactive element, functional object, or task path segment; based on the type of abnormal indicator, the degree of contribution of the indicator, and the location of the problem, the corresponding adjustment action is matched from the interface adjustment rule base.

[0270] Output a structured diagnostic record, which includes at least one of the following: task identifier, interface status identifier, interactive element or functional object identifier, abnormal indicator type, indicator value, indicator contribution level, problem type, redundancy level, and adjustment action.

[0271] The adjustment actions include at least one of the following: process structure adjustment, interface state transition relationship adjustment, interactive hotspot or interactive attribute adjustment, feedback state adjustment, navigation information adjustment, semantic label adjustment, and non-target visual element display attribute adjustment.

[0272] In an interface system that supports automatic configuration, the adjustment action can be further converted into machine-readable interface adjustment parameters and output to the corresponding interface configuration module.

[0273] Table 1 shows an example of the mapping rules between indicators, problem types, and optimization suggestions:

[0274] Additional operations The process is too long and the necessary steps are not properly broken down. Interface status: Confirmation status after information entry is completed; Interface elements: "Next" and "Confirm" buttons; Specific problem: The same confirmation operation is split into multiple low-value steps; Severity: High redundancy; Adjustment rules: Process simplification rules - merge consecutive confirmation steps and delete intermediate pages that do not affect the task result. Invalid interaction rate Hotspots, interactivity, or feedback anomalies Interface status: Form editing state; Interface element: Drop-down selection box; Specific problem: Clicking the text area does not expand the options, and there is no unavailable prompt; Severity: Severe redundancy; Adjustment rules: Interactivity rules - expand the effective hotspot, unify the clickable range of controls, and clearly display the disabled status and operation feedback. Real-time rollback rate The expected entry point is inconsistent with the actual content. Interface status: The state after entering the landing page from the function entry point; Interface elements: "Advanced Settings" entry point; Specific problem: The entry name is inconsistent with the landing page content, and the user returns immediately after entering; Severity: High redundancy; Adjustment rules: Entry consistency rule - rewrite the entry name to make the entry label, target content and landing point level consistent. Interface revisit rate Information fragmentation and cross-state information transfer Interface state: Repeatedly switching between list page and details page; Interface elements: List items and "View Details" entry; Specific problem: Information required to complete the current task is scattered across different pages, requiring users to repeatedly go back and forth to view it; Severity: High redundancy; Adjustment rules: Information integration rules - Display key summaries on the same screen in the current interface, while retaining filter conditions, scroll position, and entered content. Functional retry rate Unclear status feedback or delayed response Interface Status: Processing status after submission; Interface Element: "Submit" button; Specific Issue: The second call to the "Submit" function within the same target task stage is still counted as a net function retry after reasonable duplicate rules are excluded; Combining the processing status and feedback attributes after the first call, the cause is diagnosed as unclear feedback or delayed response; Severity: Severe redundancy; Adjustment Rules: Feedback and anti-retry rules - display progress immediately after the first click and temporarily disable the button, while adopting an anti-duplicate submission mechanism. Path deviation Navigation or information architecture anomaly Interface status: Task execution status in multi-level navigation; Interface elements: Navigation menu and target function entry point; Specific problems: The main task entry point is not prominent and there are too many similar branches, causing users to deviate from the effective baseline path determined by their set of effective candidate paths; Severity: Severe redundancy; Adjustment rules: Main path optimization rules—prioritize high-frequency entry points, reduce unnecessary branches, and highlight recommended operation paths through navigation labels. Semantic confusion Tag or functional semantic competition Interface status: Function selection status; Interface elements: "Save" and "Save and Submit" buttons; Specific problem: Adjacent element names and visual styles are too similar, making it difficult for users to judge the operation result; Severity: High redundancy; Adjustment rules: Semantic differentiation rules - adopt result-oriented differentiated naming, and differentiate functions through icons, explanatory text, and visual hierarchy. Visual interference Obscuring, dynamic attraction, or attention competition Interface status: Key task confirmed status; Interface elements: Target confirmation button and its dynamic pop-up layer above it; Specific problem: Non-target pop-up layers obscure the operation area and compete for attention through animation; Severity: Severe redundancy; Adjustment rules: Visual obstruction reduction rules - delay unnecessary pop-up layers, reduce the prominence of animation effects, avoid key operation areas and strengthen the hierarchy of target elements.

[0275] System Implementation Method:

[0276] like Figure 3As shown, the system includes a data acquisition module, an indicator calculation module, a standardization module, a dimension aggregation module, a comprehensive evaluation module, a level determination module, and a diagnostic suggestion module, corresponding one-to-one with steps S101 to S107. It also has access to the effective candidate path set and effective benchmark path library, the path attribution rule library, the reasonable repeated call rule library, the reference sample library, the indicator-problem rule library, and the problem-suggestion rule library. The effective candidate path set and effective benchmark path library store effective candidate paths that meet the target task completion conditions and necessary constraints and pass log integrity verification, the set identifiers of each effective candidate path, and the effective benchmark paths determined within each set according to the minimum interaction step size rule. The path attribution rule library stores the functional branches used to distinguish different reasonable completion methods and determine the effective candidate path set to which the actual interaction path belongs, preset task rules, and their versions. The reasonable repeated call rule library stores the reasonable repeated function types, applicable task stages, allowed repeated conditions, and rule versions. The metric calculation module first determines the set of valid candidate paths to which the actual path to be evaluated belongs based on the functional branch or preset task rules corresponding to the actual path to be evaluated. Within this set, a valid baseline path is determined. If multiple valid baseline paths exist with total interaction steps equal to the minimum value, the path with the smallest deviation from the actual interface state path is selected as the chosen valid baseline path for metric calculation of the current task sample. The total interaction step length of this selected valid baseline path is used as the baseline interaction step length. The additional operation ratio and path deviation are then calculated. Simultaneously, the function retry rate is calculated based on reasonable repeated call rules. The actual path to be evaluated is not automatically added to the set of valid candidate paths or the valid baseline path library simply because its step length is short. Each module can be deployed on the same electronic device or distributed and exchange data via an interface.

[0277] Application example:

[0278] Taking a target task of a graphical human-computer interaction interface as an example, generalized interaction logs and interface state sequences from multiple valid sessions are collected, and eight secondary indicators are obtained according to the aforementioned calculation formula. Regarding visual interference, non-target visual elements in each interface state are uniformly included in the set, and the intensity is directly calculated based on the number of elements, area ratio, spatial position weight, and interference morphology. After standardizing the eight indicators, the CRI is calculated according to predetermined intra-layer and inter-layer weights.

[0279] Mobile, desktop, vehicle, and industrial control interface samples relevant to the task to be evaluated are selected from the reference sample library. Equal category weights are assigned to each interface category, and samples are evenly distributed within each category. Reference CRIs are sorted from smallest to largest, and cumulative weights are calculated to obtain weighted 25%, 50%, and 75% quantiles as θ1, θ2, and θ3. The CRI of the interface to be evaluated is compared with the thresholds to obtain its level. Major problems are identified based on the contribution of the indicators, and corresponding optimization suggestions for processes, feedback, navigation, semantics, or visual aspects are output.

[0280] The following provides a set of recalculated numerical examples. A group evaluation model is used, with four valid sessions forming one evaluation unit. Each session determines its set of valid candidate paths based on its corresponding functional branch or preset task rules, and within this set, a valid baseline path is selected for calculating the current task's sample metrics. The shortest paths in a row are all considered valid baseline paths, and the one with the smallest deviation from the actual interface state path is chosen as the selected valid baseline path for calculating the current sample metrics. In this example, the total interaction step length of the selected valid baseline paths for each session... Number of times the interface state jumps All are 10. The number of function retries for each session is the net number of function retries obtained after identifying the second and subsequent calls within the same target task phase, excluding reasonable repeated calls such as quantity increases or decreases, continuous parameter adjustments, pagination, scrolling, and text editing. According to the order of "actual interaction step size, number of invalid interactions, number of effective interactions after cleaning, number of deduplicated jumps contained in the rollback segment, number of effective jumps after cleaning, number of redundant revisits, number of effective accesses after cleaning, net number of function retries, and path editing distance", the data for the four sessions are (12, 1, 9, 2, 9, 3, 10, 1, 2), (14, 2, 10, 2, 8, 4, 11, 2, 3), (13, 1, 9, 4, 8, 2, 10, 2, 4) and (15, 3, 11, 2, 10, 5, 12, 3, 3).

[0281] The extra operation ratio (ARR) was calculated using equations (1) to (28), and the average was calculated for the four sessions. The ARR was 0.254121, the invalid interaction rate was 0.145238, the instant rollback rate was 0.220455, the interface revisit rate was 0.239555, the function retry rate was 0.165693, and the path deviation was 0.228480. Taking the first session as an example, the bounded ARR was 0.166667, the invalid interaction rate was 0.100000, the instant rollback rate was 0.181818, the interface revisit rate was 0.230769, the function retry rate was 0.100000, and the bounded path deviation was 0.166667. The other sessions were calculated using the same method.

[0282] For semantic confusion, the maximum number of interferences is taken as 5. Each of the four sessions contains one interface state involving the target operation. The number of non-target semantic interference elements and the five-level semantic relationship scores are as follows: 2 with scores of 3 and 4; 3 with scores of 3, 4, and 4; 2 with scores of 4 and 5; and 4 with scores of 2, 3, 4, and 5. According to equations (29) to (35), the semantic confusion of the four sessions is 0.250000, 0.400000, 0.350000, and 0.500000, with a group average of 0.375000. For visual interference, the upper limit of the number of elements is taken as 5. Each of the four sessions takes one valid interface state. The area factor, spatial location factor, and disturbance morphology intensity factor of each element are taken as (0.80, 0.50, 1.00), (0.80, 0.625, 1.00); (0.60, 0.60, 1.00), (0.75, 0.60, 1.00), (0.90, 0.60, 1.00); (0.70, 0.80, 1.00), (0.80, 0.80, 1.00); (0.50, 0.60, 1.00), (0.50, 0.80, 1.00), (0.625, 0.80, 1.00) and (0.75, 0.80, 1.00). The visual interference contributions of a single element calculated according to equations (37) to (40) are (0.40, 0.50), (0.36, 0.45, 0.54), (0.56, 0.64) and (0.30, 0.40, 0.50, 0.60), respectively, with corresponding interface visual interference degrees of 0.18, 0.27, 0.24 and 0.36. The group visual interference degree is 0.262500 according to equation (45).

[0283] During standardization, a positive number of 0.001 is used to prevent the denominator from being zero. The lower reference quantiles of the eight indicators are 0.05, 0.02, 0.03, 0.04, 0.02, 0.04, 0.08, and 0.05, respectively, and the upper reference quantiles are 0.40, 0.25, 0.35, 0.36, 0.28, 0.35, 0.60, and 0.50, respectively. According to formula (46) and retaining 6 decimal places, the standardized values ​​of the eight indicators are 0.581541, 0.542156, 0.593316, 0.621667, 0.558209, 0.606045, 0.566219, and 0.471175, respectively, all of which are in the range of 0 to 1.

[0284] The weights of the four primary indicators are 0.30, 0.35, 0.15 and 0.20 respectively; the weights in the unnecessary redundancy layer are 0.60 and 0.40, the weights in the repetitive redundancy layer are 0.50, 0.30 and 0.20, the weight in the path deviation redundancy layer is 1, and the weights in the interference redundancy layer are 0.50 and 0.50. From equation (51), the combined weights of the eight secondary indicators are 0.18, 0.12, 0.175, 0.105, 0.07, 0.15, 0.10 and 0.10 respectively, and their sum is 1; from equation (53), the scores of the four primary indicators are 0.565787, 0.594800, 0.606045 and 0.518697 respectively; from equation (54) or equation (55), the CRI is 0.572562. The contributions of the eight indicators were 0.104677, 0.065059, 0.103830, 0.065275, 0.039075, 0.090907, 0.056622 and 0.047118 respectively, and their sum was consistent with the CRI.

[0285] The reference sample library includes four types of interfaces: mobile terminals, desktop software, automotive, and industrial control. Each type has four reference evaluation units, and each unit is generated by aggregating four valid sessions in the same way as the unit being evaluated. The category weight for each of the four types is 0.25, therefore the weight of each reference evaluation unit is 0.0625. The CRIs of the 16 reference evaluation units, sorted from smallest to largest, are 0.12, 0.18, 0.23, 0.28, 0.32, 0.37, 0.42, 0.46, 0.50, 0.55, 0.60, 0.65, 0.70, 0.76, 0.82, and 0.90. The cumulative weights reach 0.25, 0.50 and 0.75 at the 4th, 8th and 12th reference evaluation units, respectively. Therefore, the three level boundaries obtained by equations (56) to (58) are 0.28, 0.46 and 0.65, respectively. The CRI of the unit to be evaluated is greater than 0.46 and not greater than 0.65. According to equation (59), it is determined to be level 3, that is, high redundancy.

[0286] In this example, using the same reference sample and sample weights as in step S106, weighted quantiles corresponding to the preset abnormal quantile levels are calculated for the reference distribution of the eight standardized indicator values, and the obtained quantiles are used as the indicator abnormality limits for each indicator; the abnormality limits for the eight indicators are 0.65, 0.60, 0.58, 0.60, 0.60, 0.60, 0.65, and 0.60, respectively. The instant rollback rate, interface revisit rate, and path deviation exceed the corresponding abnormality limits, and their indicator contributions are 0.103830, 0.065275, and 0.090907, respectively; ranked from highest to lowest contribution, they are instant rollback rate, path deviation, and interface revisit rate. Based on this, the following optimization suggestions are provided: First, check the interface jumps corresponding to instant rollback, clarify the jump target, current position and return semantics, and provide perceptible feedback; second, adjust the main task entry and navigation hierarchy to reduce unnecessary branches; third, continuously display task progress in cross-interface tasks, retain the entered status and provide an entry point to directly return to the target step, so as to reduce interface revisits.

[0287] This invention understands the complexity of graphical human-computer interaction interfaces from a cognitive-behavioral perspective. It argues that interface complexity manifests not only in the complexity of interface elements, visual information, or functional structures, but also in the additional operations and cognitive load incurred by users during real-world task execution due to factors such as interface structure, interactive feedback, information organization, and visual presentation. Therefore, this invention not only evaluates the visual or structural complexity of the interface but also identifies redundant behaviors such as unnecessary operations, repetitive operations, path deviations, and attentional distractions caused by interface complexity. These are summarized into four primary evaluation dimensions: unnecessary redundancy, repetitive redundancy, path deviation redundancy, and distracting redundancy. The first three dimensions primarily characterize the user's operations and path results during task execution based on interaction logs and interface state paths. Distracting redundancy primarily characterizes semantic and visual factors that may induce invalid operations, rollbacks, retries, path deviations, or attentional shifts based on interface element attributes. These four primary evaluation dimensions are used together to evaluate the overall degree of operational redundancy of a specific interface under a target task. By analyzing interaction logs, interface state paths, and interface element attributes, the aforementioned complex experience is transformed into eight calculable, comparable, and machine-detectable evaluation indicators. This constructs a comprehensive evaluation method for operational redundancy that simultaneously covers the overall task process and individual interface state nodes. All eight indicators are weighted as positive secondary indicators into a comprehensive redundancy index, and the weighted quantiles of reference samples are used to form level boundaries. This enables a quantitative evaluation of the overall redundancy level, as well as a classification and diagnosis of abnormal performance, negative experiences, triggering causes, and design attributions. This provides a calculable basis for the complexity identification, problem localization, and design optimization of various graphical human-computer interaction interfaces.

[0288] The scope of protection of this invention includes, but is not limited to, the above embodiments. The scope of protection of this invention is defined by the claims. Any substitutions, modifications, or improvements to this technology that are easily conceived by those skilled in the art fall within the scope of protection of this invention.

Claims

1. A comprehensive evaluation method for redundancy in human-computer interaction interface operations based on interaction logs and path data, characterized in that, Includes the following steps: S101, obtain the interaction logs corresponding to the target task, the actual interface state path, one or more valid candidate paths, at least one of the functional branches or preset task rules corresponding to the actual interaction path to be evaluated, task target element attributes, and interface element attributes; wherein, at least one of the functional branches or preset task rules is used to distinguish different reasonable completion methods of the same target task and determine the set of valid candidate paths to which the actual interaction path to be evaluated belongs; the valid candidate path is the task completion path that meets the target task completion conditions, necessary state constraints, necessary interaction steps, and preset business or security constraints, and passes the log integrity verification. S102, calculate eight secondary indicators, including extra operation ratio, invalid interaction rate, instant rollback rate, interface revisit rate, function retry rate, path deviation, semantic confusion, and visual interference. Among them, interface elements in each interface state that are not directly related to the current task target and have occlusion, motion, significant contrast, or continuous exposure characteristics are uniformly regarded as non-target visual elements. The visual interference is directly calculated based on the number, area ratio, spatial position weight, and interference morphology intensity of the non-target visual elements. S103, the eight secondary indicators are homogenized and standardized to obtain the standardized values ​​of the eight secondary indicators; S104. The extra operation ratio and invalid interaction rate are classified as unnecessary redundancy primary indicators. The instant rollback rate, interface revisit rate and function retry rate are classified as repetitive redundancy primary indicators. The path deviation is classified as path deviation redundancy primary indicators. The semantic confusion and visual interference are classified as interference redundancy primary indicators. The scores of the four primary indicators are calculated according to the weights of the secondary indicators contained in each primary indicator and the standardized values ​​of the secondary indicators. When a primary indicator contains only one secondary indicator, the score of the primary indicator is directly taken as the standardized value of that secondary indicator. S105. Calculate the comprehensive redundancy index (CRI) based on the weights of the four primary indicators, the weights of the secondary indicators within each primary indicator layer, and the standardized values ​​of the secondary indicators. The weights of the primary indicators and the weights of the secondary indicators within each secondary indicator layer are all positive and satisfy the normalization constraints. The product of the weights of the primary indicators and the weights of the secondary indicators within each secondary indicator layer is used as the combined weights of the secondary indicators, so that all eight secondary indicators enter the CRI with positive combined weights. S106, Assign sample weights to the CRI values ​​of the reference samples and sort them according to the CRI values; determine at least two level boundaries based on the comprehensive redundancy index distribution in the reference sample set; compare the CRI of the interface to be evaluated with the level boundaries to determine the overall redundancy level of the interface to be evaluated. S107, using the same reference samples and sample weights as in step S106, determine the abnormal limits of each indicator based on the reference distribution of the standardized values ​​of each secondary indicator; determine the indicator contribution by multiplying the standardized value of each secondary indicator of the interface to be evaluated with the weight of its corresponding secondary indicator combination, and compare the standardized value of each secondary indicator with the corresponding indicator abnormal limit to identify abnormal indicators; match the abnormal indicators or combinations of abnormal indicators with preset indicators and problem mapping rules, and output interface optimization suggestions containing preset problem types and adjustment actions according to the indicator contribution.

2. The comprehensive evaluation method for redundancy of human-computer interaction interface operation based on interaction logs and path data according to claim 1, characterized in that, The human-machine interface is a graphical human-machine interface with the ability to identify interface status, interactive elements, and record operation logs, including at least one of mobile terminal interfaces, desktop software interfaces, vehicle-mounted human-machine interfaces, industrial control interfaces, smart screen interfaces, and virtual reality or augmented reality interactive interfaces; the interaction log records at least one interactive event among clicks, touches, buttons, knobs, gestures, voice triggers, gaze confirmations, and controller inputs.

3. The method according to claim 1, characterized in that, Combine at least two of the following: page identifier, window identifier, mode identifier, pop-up identifier, operation mode identifier, and interactive element tree summary, to form the interface state identifier, so that page switching, mode switching, window switching, pop-up state change, and operation mode change are uniformly represented as interface state transition.

4. The method according to claim 1, characterized in that, The additional operation ratio is determined based on the difference between the actual interaction step size and the reference interaction step size, the ratio of the difference to the reference interaction step size, and their bounded results. The actual interaction path used for calculating the additional operation ratio is a valid actual interaction path that satisfies the target task completion conditions, necessary state constraints, necessary interaction steps, and preset business or security constraints. The valid candidate path is a task completion path that satisfies the target task completion conditions, necessary state constraints, necessary interaction steps, and preset business or security constraints, and passes the log integrity verification. For the same target task with different reasonable completion methods, the valid candidate paths are distinguished according to the functional branches or preset task rules, and corresponding valid candidate path sets are formed respectively. For the actual interaction path to be evaluated, the valid candidate path set to which it belongs is determined according to its corresponding functional branch or preset task rules, and then the path with the smallest total interaction step length is selected from the valid candidate path set as the valid benchmark path. When there is only one valid benchmark path, the valid benchmark path is selected as the valid benchmark path. When there are multiple valid candidate paths with a total interaction step length equal to the minimum value, all of them are taken as valid baseline paths, and the valid baseline path with the smallest deviation from the actual interface state path is selected as the selected valid baseline path. The reference interaction step size is the total interaction step size of the selected valid reference path; The invalid interaction rate is determined by the sum of the number of invalid interaction events and the number of valid interaction events after removing redundant loops. Invalid interaction events are those that do not cause a change in the interface state within the response window, do not generate a system acceptance flag, and do not generate feedback related to the target task.

5. The method according to claim 1, characterized in that, The instant rollback rate is determined based on the number of interface state jumps and the number of effective interface state jumps contained in the rollback segment in the actual interface state sequence after deduplication. The interface revisit rate is determined based on the cumulative number of revisits and the number of valid visits between two adjacent visits to the same interface state, with an interval exceeding two state positions. The function retry rate is determined by the number of function retry candidates exceeding the first call within the same target task stage, after excluding reasonable repeated calls that fall under the task rules such as quantity increase / decrease, continuous parameter adjustment, pagination, scrolling, and text editing, and is combined with the number of valid interactions. Among them, the second and subsequent calls of the same function object are first identified as function retry candidate events, and the remaining candidate events after excluding reasonable repeated calls are included in the function retry count. Whether the first call is successful, whether the expected task state is reached, and whether the feedback is clear are only used as diagnostic attributes for function retry reasons, and are not necessary conditions for function retry identification.

6. The method according to claim 4, characterized in that, The path deviation is determined based on the edit distance between the actual interface state sequence and the interface state sequence of the selected effective benchmark path, the number of interface state jumps contained in the interface state sequence of the effective benchmark path, and the bounded result. When there are multiple effective benchmark paths in the effective candidate path set to which the current task sample belongs, the path deviation is calculated based on the actual interface state sequence and the interface state sequence of each effective benchmark path, and the minimum value is taken. The effective benchmark path corresponding to the minimum value is selected as the effective benchmark path and used for the calculation of the current task sample index.

7. The method according to claim 1, characterized in that, The semantic confusion level is determined based on the semantic similarity between the target interactive element and each non-target interactive element in the same interface state, the number of non-target semantic interference elements, and the saturation coefficient of the number of semantic interference items; the semantic similarity is determined based on at least one of text labels, icon semantic labels, function identifiers, system function codes, accessibility names, or operation descriptions. The visual interference degree is determined based on the set of non-target visual elements in each interface state. The saturation coefficient of the number of non-target visual elements is determined according to a preset saturation upper limit. The area ratio, spatial position weight, and interference morphology intensity of a single non-target visual element are multiplied and aggregated sequentially within the interface state to obtain the visual interference degree of the interface state. The visual interference degree of the task sample is obtained by aggregating among the effective interface states of the same task sample. The visual interference degree of the evaluation unit is obtained by aggregating within the evaluation unit composed of one or more effective task samples. When the set of non-target visual elements is empty, the visual interference contribution corresponding to that interface state is zero.

8. The method according to claim 1, characterized in that, In step S103, the secondary index is subjected to truncated linear standardization using the lower and upper reference quantiles of the reference distribution. In step S104, the expert's given first... The average importance score of the first-level indicator and the number of items within the same first-level indicator The weighted average scores of the secondary indicators were normalized to obtain the results for the second-level indicators. Item Level 1 Indicator Weight and the The first-level indicator Intra-layer weight of secondary indicators and with Determine the first The first-level indicator The weighting of the secondary indicators; For the First-level indicators, by Calculate the scores for the primary indicators; among which, This indicates that the first step obtained in step S103 is... The first-level indicator Standardized values ​​of secondary indicators; Indicates the first The number of secondary indicators included in a primary indicator; In step S105, the comprehensive redundancy index is calculated. , ,in , , , and ; The weights of the primary indicators and the weights within the secondary indicator layers are all positive values. The sum of the weights of all primary indicators is 1. The sum of the weights within the secondary indicator layers under the same primary indicator is 1. The sum of the combined weights of all secondary indicators is 1.

9. The method according to claim 1, characterized in that, The level boundary is adaptively determined based on the comprehensive redundancy index distribution of the reference sample set; the reference samples are stratified according to interface category, task category, or a combination of both, and sample weights are set for different categories; Multiple ordered level regions are obtained based on the empirical distribution or clustering distribution of the weighted comprehensive redundancy index, and the level boundaries are determined by the boundary positions between adjacent level regions. In the method of determining the class boundaries using weighted quantiles, when four redundant classes are set, the weighted 25%, 50% and 75% quantiles are used as three class boundaries respectively. In the method of determining the level boundary using clustering, the comprehensive redundancy index of the reference sample is divided into multiple ordered clusters, and the corresponding level boundary is determined according to the intersection of adjacent cluster centers, adjacent cluster boundaries, or adjacent probability distributions. The evaluation unit is the same as that used to determine the grade boundary for the sample to be evaluated; when the evaluation unit is a single valid execution record of a single user or a single session, the total number of samples is 1; when the evaluation unit is a group sample consisting of multiple valid execution records, each reference sample and the sample to be evaluated use the same number of valid execution records, aggregation method and sample weight.

10. The method according to claim 1, characterized in that, Abnormal indicators are determined based on the standardized values ​​of the secondary indicators and the corresponding abnormality limits. For a single abnormal indicator or multiple indicators that are abnormal at the same time, the preset indicator and problem mapping rules are invoked to determine the corresponding problem type and the location of the problem; and after sorting by indicator contribution, interface optimization suggestions containing preset problem types and adjustment actions are output.

11. A comprehensive evaluation system for redundancy in human-computer interaction interface operations based on interaction logs and path data, characterized in that, include: The data acquisition module is used to acquire the interaction logs corresponding to the target task, the actual interface state path, one or more valid candidate paths, at least one of the functional branches or preset task rules corresponding to the actual interaction path to be evaluated, task stage information, functional object identifiers, reasonable repeated call rules, task target element attributes, and interface element attributes; wherein, at least one of the functional branches or preset task rules is used to distinguish different reasonable completion methods and determine the set of valid candidate paths to which the actual interaction path to be evaluated belongs; the valid candidate paths are task completion paths that meet the target task completion conditions, necessary state constraints, necessary interaction steps, and preset business or security constraints, and pass the log integrity verification. The indicator calculation module is used to calculate the extra operation ratio, invalid interaction rate, instant rollback rate, interface revisit rate, function retry rate, path deviation, semantic confusion, and visual interference. Specifically, based on at least one of the functional branches or preset task rules corresponding to the actual interaction path to be evaluated, a set of valid candidate paths is determined. Within this set, the path with the smallest total interaction step length is selected as the valid baseline path. When only one valid baseline path exists, that path is selected. When multiple valid candidate paths with the same minimum total interaction step length exist, all are selected as valid baseline paths, and the path with the smallest deviation from the actual interface state path is chosen. The total interaction step length of the selected valid baseline path is used as the baseline interaction step length. Within the same target task stage, the second and subsequent calls to the same functional object are identified as candidate events. After excluding the corresponding candidate events according to reasonable repeated call rules, the function retry rate is calculated. The standardization module is used to align and standardize the eight secondary indicators. The dimension aggregation module is used to classify the extra operation ratio and invalid interaction rate into the unnecessary redundancy first-level indicator, the instant rollback rate, interface revisit rate and function retry rate into the repetitive redundancy first-level indicator, the path deviation degree into the path deviation redundancy first-level indicator, and the semantic confusion degree and visual interference degree into the interference redundancy first-level indicator. The scores of the four first-level indicators are calculated according to the weight of the second-level indicators under each first-level indicator and the standardized value of the second-level indicators. The comprehensive evaluation module is used to calculate the comprehensive redundancy index (CRI) based on the weights of the four primary indicators, the weights of the secondary indicators under each primary indicator, and the standardized values ​​of the secondary indicators. The grade determination module is used to assign sample weights to the CRI values ​​of the reference samples and sort them according to the CRI values, and to determine at least two grade boundaries based on the comprehensive redundancy index distribution in the reference sample set. The diagnostic suggestion module is used to determine the indicator contribution by multiplying the standardized value of each secondary indicator with the weight of its corresponding secondary indicator combination, and to determine abnormal indicators based on the standardized value of the secondary indicator and the corresponding indicator abnormality limit. The abnormal indicators or combinations of abnormal indicators are matched with preset indicators and problem mapping rules, and the module outputs interface optimization suggestions containing preset problem types and adjustment actions according to the indicator contribution.

Images