Third-party communication software data extraction method based on non-intrusive detection

Abstract

The application relates to a third-party communication software data extraction method, system and device based on non-intrusive detection and a medium, and relates to the technical field of automatic data processing.The method comprises the following steps: obtaining a target software interface visual hot area through a perception layer; then constructing a spatial redundancy sampling matrix in the visual hot area through a collection layer; simulating interactive operation to obtain a plurality of text segments; then performing multi-frame comparison and splicing on the plurality of text segments through a cognitive layer to obtain a complete text sequence; and finally performing structured analysis on the complete text sequence through an output layer to obtain structured data in a preset format.The application can adapt to interface dynamic changes without intruding into the bottom layer of the software to solve the problems of missing and misplacement of fields and improve the accuracy of data restoration.

Description

Technical Field

[0001] This application relates to the field of automated data processing technology, and in particular to a method, system, device and medium for extracting data from third-party communication software based on non-invasive detection. Background Technology

[0002] As enterprises deepen their digital transformation, compliant application scenarios such as office automation, internal auditing of system-level collaboration software, automated testing, and user-initiated legitimate data backups are becoming increasingly common. In these clearly authorized application scenarios, efficiently and accurately extracting information displayed on third-party communication software interfaces has become a key aspect of improving business process automation. However, third-party communication software typically employs a closed ecosystem design, with complex and varied interface structures and often incomplete content display. Furthermore, most software lacks standardized data export interfaces, making the legal and efficient extraction of visual information from the interface, while ensuring strict compliance with personal privacy policies and relevant data compliance requirements, a significant technical challenge.

[0003] Currently, existing technologies primarily employ the following methods to address the aforementioned needs: First, protocol reverse engineering or hooking techniques, which acquire data by cracking underlying communication protocols or scanning memory signatures; second, traditional UI automation methods, which directly read the control attribute values ​​of interface elements through operating system APIs; and third, OCR (Optical Character Recognition) technology, which performs image analysis on screenshots to extract text. While these methods achieve data acquisition to some extent, their applicable scenarios and technical approaches differ, and a comprehensive solution that balances security, robustness, and accuracy has not yet been formed.

[0004] However, the aforementioned existing technologies all have significant drawbacks in practical applications. Protocol reverse engineering and hooking techniques are highly susceptible to triggering the target software's security detection mechanisms, posing a risk of account bans. Furthermore, protocol changes after software version updates lead to extremely high maintenance costs, essentially constituting an intrusive operation with unreliable compliance. Traditional UI automation methods are limited by encryption software's UI tree hardening or custom rendering techniques, often failing to reliably read text content through control identifiers, especially when interface elements dynamically change. While OCR technology does not rely on the software's internal structure, its recognition accuracy is significantly affected by factors such as resolution, font, and background interference. For data extraction scenarios requiring accuracy equivalent to copy-paste, its latency and error rate are high, making it difficult to meet efficient and accurate business needs. Therefore, how to achieve cross-platform, highly reliable structured reconstruction of front-end visual presentation information without intruding into the software's underlying layers or interfering with its normal operation has become a pressing technical problem in this field. Summary of the Invention

[0005] The purpose of this application is to provide a non-intrusive detection-based method for extracting data from third-party communication software. This method can adapt to dynamic changes in the interface without intruding on the underlying software, thereby solving the problem of missing or misaligned fields and improving the accuracy of data restoration.

[0006] Firstly, this application provides a method for extracting data from third-party communication software based on non-intrusive detection, employing the following technical solution: A method for extracting data from third-party communication software based on non-intrusive detection, comprising: The visual heatspots of the target software interface are obtained through the perception layer; By constructing a spatially redundant sampling matrix within the visual hotspot using the acquisition layer, multiple text fragments are obtained by simulating interactive operations. The cognitive layer performs multi-frame comparison and splicing of the multiple text fragments to obtain a complete text sequence; The output layer performs structured parsing on the complete text sequence to obtain structured data that conforms to a preset format.

[0007] By employing the above technical solution, the perception layer accurately locates the visual hotspots of the target software, defining the working range for subsequent operations; the acquisition layer constructs a spatially redundant sampling matrix within the hotspots, acquiring multiple text fragments through simulated interactive operations, overcoming the obstacle of not being able to directly read interface elements; the cognition layer performs multi-frame comparison and splicing of multiple fragments, restoring fragmented text into a complete sequence; the output layer performs structured parsing of the complete sequence, outputting data conforming to a preset format. This forms a complete closed loop from physical perception to logical processing, solving the problem of data fragmentation caused by dynamic changes in the interface structure, achieving automated data extraction under non-intrusive conditions, and improving the accuracy of data restoration.

[0008] In a preferred embodiment, this application can be further configured such that: the acquisition layer includes a multi-point scanning decision unit, an asynchronous action execution unit, and a clipboard capture unit; the step of constructing a spatially redundant sampling matrix within the visual hotspot through the acquisition layer to simulate interactive operations and acquire multiple text fragments includes: The multi-point scanning decision-maker generates a sequence of spatially nonlinearly distributed spatial contact points within the visual thermal zone. The asynchronous action execution unit performs simulated interactive operations on each touch point in the spatial touch point sequence, triggering text selection and copying actions at the operating system level; The clipboard capture unit acquires the text fragment corresponding to each copy operation.

[0009] By adopting the above technical solution, the multi-point scanning decision-maker generates a horizontally non-linearly distributed spatial touchpoint sequence within the visual hot zone, enabling sampling points to penetrate occluded areas such as interface icons. The asynchronous action execution unit performs simulated interactive operations on each touchpoint, triggering text selection and copying actions at the operating system level to acquire high-quality text. The clipboard capture unit acquires the text fragment corresponding to each copy operation. In summary, by dividing the acquisition layer into three functional units working collaboratively, reliable acquisition of interface text fragments is achieved without relying on software control properties.

[0010] In a preferred embodiment, this application can be further configured such that: the multi-point scanning decision unit includes a coordinate mapping unit and a step perturbation unit; the step of generating a spatially nonlinearly distributed sequence of spatial touch points within the visual hot zone via the multi-point scanning decision unit includes: The coordinate mapping unit generates a basic sampling interval based on the anchor point coordinates of the current probe row; The spatial contact sequence is generated by perturbing the sampling step size within the basic sampling interval using the step perturbation unit based on nonlinear discrete coefficients.

[0011] By adopting the above technical solution, the coordinate mapping unit generates a basic sampling interval based on the anchor point coordinates of the current detection row, ensuring the rationality of the sampling starting point; the step perturbation unit perturbs the sampling step size based on nonlinear discrete coefficients to generate a spatial touchpoint sequence, avoiding resonance blind spots caused by equidistant sampling and interface icon layout. In summary, through the nonlinear mapping segmentation mechanism, multiple detection points can be accurately positioned on lossless text segments within a wide threshold span, reducing the possibility of missing long texts from a mathematical probability perspective.

[0012] In a preferred embodiment, this application can be further configured such that: the asynchronous action execution unit includes a delay modulation unit and an action injection unit, and the step of performing a simulated interactive operation on each contact point of the spatial contact sequence through the asynchronous action execution unit includes: The delay modulation unit generates timing delay parameters for each contact point that follow a random distribution. The motion injection unit sequentially performs a simulated double-click operation on each contact point according to the timing delay parameters.

[0013] By adopting the above technical solution, the delay modulation unit generates a timing delay parameter that follows a random distribution for each touch point, making the interaction behavior conform to the natural rhythm of human operation; the action injection unit executes a simulated double-click operation on each touch point sequentially according to the timing delay parameter, triggering the operating system-level lexical boundary logic to obtain clean text. In this way, by introducing a random delay sequence, the automated operation is highly fitted to normal human user interaction behavior, effectively circumventing software security detection mechanisms and ensuring the legality of the operation and application compatibility.

[0014] In a preferred embodiment, this application can be further configured such that: the cognitive layer includes a multi-frame comparison unit, a segment splicing unit, and a noise filtering unit; the step of performing multi-frame comparison and splicing of the multiple text segments through the cognitive layer to obtain a complete text sequence includes: The multi-frame comparison unit performs overlapping region detection on the multiple text segments to determine the common substrings between the segments. The fragment splicing unit merges adjacent fragments based on the common substring to obtain spliced ​​text; The noise filtering unit removes unwanted characters from the concatenated text to obtain the complete text sequence.

[0015] By employing the above technical solution, the multi-frame comparison unit detects overlapping areas of multiple text segments, identifies common substrings between segments, and recognizes text boundaries that can be spliced; the segment splicing unit fuses adjacent segments based on the common substrings, restoring fragmented text to continuous content; and the noise filtering unit removes unexpected characters from the spliced ​​text, resulting in a clean and complete text sequence. This achieves seamless splicing and fusion of occluded text segments, solving the problem of data truncation caused by physical obstruction of text by interface icons.

[0016] In a preferred embodiment, this application can be further configured such that the step of detecting overlapping regions of the multiple text segments through the multi-frame comparison unit and determining common substrings among the segments includes: The multi-frame comparison unit uses dynamic programming to construct the character matching state space and takes the longest common substring between detected segments as the overlapping region.

[0017] By employing the above technical solution, the multi-frame comparison unit uses dynamic programming to construct a character matching state matrix and searches for the longest common substring between two text segments in memory. This accurately identifies overlapping parts between segments, providing precise fusion boundaries for subsequent splicing. Even when there are local differences in the text, it can still correctly identify overlapping areas, ensuring the accuracy and reliability of the splicing operation and improving the success rate of text reconstruction.

[0018] In a preferred embodiment, this application can be further configured such that: the output layer includes an anchor point recognition unit, a field inference unit, and a structured mapping unit; the step of performing structured parsing of the complete text sequence through the output layer to obtain structured data conforming to a preset format includes: The anchor point recognition unit identifies a unique identifier field from the complete text sequence based on character encoding features, and positions the unique identifier field as the spatial origin. Based on the relative position of the spatial origin, the field deduction unit infers the attribution of other attribute fields; The structured mapping unit maps the spatial origin and other attribute fields to a preset data structure to obtain the structured data.

[0019] By adopting the above technical solution, the anchor point recognition unit identifies the unique identifier field from the complete text sequence based on character encoding features and positions it as the spatial origin, solving the problem of field misalignment that is easily caused by traditional fixed line number matching when interface data is missing; the field inference unit infers the attribution of other attribute fields based on the relative position of the spatial origin, and realizes field placement by utilizing the spatial consistency of UI design; the structured mapping unit maps the recognition results to a preset data structure. Thus, adaptive parsing is achieved in the case of missing fields.

[0020] In a preferred embodiment, this application can be further configured such that: the anchor point recognition unit includes a feature calculation unit and a filtering unit, and the step of identifying a unique identifier field from the complete text sequence based on character encoding features by the anchor point recognition unit includes: The feature calculation unit calculates and obtains the character purity index and regular expression matching index for the candidate substrings in the complete text sequence, respectively. The unique identifier field is selected by the filtering unit based on the weighted result of the character purity index and the regular expression matching index.

[0021] By employing the above technical solution, the feature calculation unit calculates character purity and regularity matching indices for candidate substrings, objectively measuring the probability that a candidate word carries a unique identifier; the filtering unit selects the unique identifier field based on the weighted results, prioritizing the candidate word with the highest overall confidence as the anchor point. This approach accurately filters out mixed noise, ensuring accurate identification of unique identifiers even when the interface contains a large number of non-standard characters.

[0022] Secondly, this application provides a third-party communication software data extraction system based on non-invasive detection, employing the following technical solution: A third-party communication software data extraction system based on non-invasive detection includes: The perception module is used to acquire the visual heatspots of the target software interface; The acquisition module is used to construct a spatially redundant sampling matrix within the visual hotspot and simulate interactive operations to acquire multiple text fragments. The cognitive module is used to perform multi-frame comparison and splicing of the multiple text fragments to obtain a complete text sequence; The output module is used to perform structured parsing on the complete text sequence to obtain structured data that conforms to a preset format.

[0023] Thirdly, this application provides an electronic device that adopts the following technical solution: An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for extracting third-party communication software data based on non-invasive detection.

[0024] In summary, this application has the following beneficial technical effects: 1. This application uses a non-intrusive detection method, which does not require modification or cracking of the code of third-party communication software, effectively avoiding software malfunction, data corruption and legal compliance risks that may be caused by intrusive operations, and ensuring the security and stability of the data extraction process; 2. This application utilizes visual heat map recognition and the construction of a spatial redundancy sampling matrix to accurately capture key interactive areas in the target software interface. Combined with multi-frame comparison and splicing technology, it can completely extract text sequences and output data that conforms to a preset format after structured parsing, thereby improving the accuracy and practicality of data extraction from third-party communication software. Attached Figure Description

[0025] Figure 1 This is a flowchart of a third-party communication software data extraction method based on non-invasive detection, as described in one embodiment of this application.

[0026] Figure 2 This is a flowchart of a sub-step of step S2 in one embodiment of this application.

[0027] Figure 3 This is a flowchart of a sub-step of step S20 in one embodiment of this application.

[0028] Figure 4 This is a flowchart of a sub-step of step S21 in one embodiment of this application.

[0029] Figure 5 This is a flowchart of a sub-step of step S3 in one embodiment of this application.

[0030] Figure 6This is a flowchart of a sub-step of step S30 in one embodiment of this application.

[0031] Figure 7 This is a flowchart of a sub-step of step S4 in one embodiment of this application.

[0032] Figure 8 This is a flowchart of a sub-step of step S40 in one embodiment of this application.

[0033] Figure 9 This is a schematic diagram of the structure of a third-party communication software data extraction system based on non-invasive detection, according to one embodiment of this application.

[0034] Figure 10 This is a schematic block diagram of an electronic device in one embodiment of this application.

[0035] Attached reference numerals: 1. Sensing module; 2. Acquisition module; 3. Cognition module; 4. Output module. Detailed Implementation

[0036] The following is in conjunction with the appendix Figure 1-10 This application will be described in further detail.

[0037] It should be noted that all actions involving the acquisition of data or information in this application are carried out in accordance with the relevant data protection laws and policies of the country where the application is located, and with the authorization of the relevant users.

[0038] refer to Figure 1 A method for extracting data from third-party communication software based on non-intrusive detection, specifically including: S1. Obtain the visual hotspots of the target software interface through the perception layer.

[0039] Specifically, the perception layer introduces an adaptive fluid coordinate system positioning algorithm. In this embodiment, the real-time handle of the target main window and the screen anchor coordinates are first dynamically obtained through the operating system kernel API. .

[0040] Subsequently, using a multi-level relative offset reference space calculation model, the logical hot zone border vector of the feature container on the details page is derived. ,in , , , It uses pre-calibrated adaptive baseline coefficients for different UI versions to accurately lock the visual perception hotspots under various system scaling ratios (DPI) and resolutions, thus defining the working range for subsequent operations.

[0041] S2. A spatially redundant sampling matrix is ​​constructed within the visual hot zone through the acquisition layer to simulate interactive operations and obtain multiple text fragments.

[0042] Specifically, the acquisition layer constructs a dynamic step-by-step sampling matrix. By spatiotemporally adding noise to the micro-interaction trajectory of the mouse, the underlying components of the platform are driven to trigger the selection of the clipboard copy action system even with slight deviations when the mouse moves down point by point.

[0043] To overcome the data truncation caused by the physical obstruction of text by interface icons (such as gender identifiers, avatars, and buttons), this embodiment employs a horizontal 7-point nonlinear redundant sampling matrix logic.

[0044] The above describes how the acquisition layer mainly constructs a spatially redundant sampling matrix within the hot zone and performs simulated interactive operations to obtain multiple text fragments, overcoming the obstacle that interface elements cannot be directly read.

[0045] S3. The cognitive layer performs multi-frame comparison and splicing of multiple text fragments to obtain a complete text sequence.

[0046] Specifically, the cognitive layer performs multimodal text flow graph denoising and decomposition. First, it performs multi-frame comparison and overlap analysis on the chaotic text flow stack captured by the temporal clipboard to achieve seamless splicing and fusion of occluded text segments. Then, it performs dimensionality reduction and stripping of invisible control characters and facial expression noise within the extracted string units, and performs logical assignment of field roles. Finally, it achieves multi-frame comparison and splicing of multiple segments, restoring fragmented text to a complete sequence.

[0047] S4. The output layer performs structured parsing on the complete text sequence to obtain structured data that conforms to the preset format.

[0048] Specifically, the output layer applies tree-structured semantic verification and entity topology projection. The output layer injects the non-standard flat key-value pairs calculated by the cognitive layer into the local field relation projection model, forcibly transforming and mapping them into structured object data that satisfies the third normal form. Before submission, it performs cross-field logical self-consistency conflict detection. If the confidence level of any core attribute detection falls below a preset threshold, a closed-loop backtracking data collection action is immediately triggered, ensuring that the output data has complete temporal consistency and anti-phantom read characteristics. This solves the problem of data fragmentation caused by dynamic changes in the interface structure, achieves automated data extraction under non-intrusive conditions, and improves the accuracy of data reconstruction.

[0049] refer to Figure 2 Furthermore, in one embodiment, the acquisition layer includes a multi-point scanning decision-maker, an asynchronous action execution unit, and a clipboard capture unit, and step S2 is refined into the following sub-steps: S20. A sequence of spatially nonlinearly distributed contact points in the visual hot zone is generated by a multi-point scanning decision unit.

[0050] Specifically, the multi-point scanning decision-maker calculates the vertical step sequence based on the height of the hot zone. Then, in each A one-dimensional spatial displacement lattice is laid horizontally along the axis slice height. This allows the sampling points to penetrate areas obscured by interface icons, etc.

[0051] S21. Simulated interactive operations are performed on each contact point in the spatial contact sequence through the asynchronous action execution unit, triggering text selection and copying actions at the operating system level.

[0052] Specifically, the engine in the asynchronous action execution unit introduces pseudo-random Brownian motion variables ( Spatiotemporal noise is added to the micro-interaction trajectory of the mouse to trigger text selection and copying actions at the operating system level in order to obtain high-quality text.

[0053] S22. Obtain the text fragment corresponding to each copy operation through the clipboard capture unit.

[0054] Specifically, the final clipboard capture unit drives the underlying components of the platform to trigger a streaming asynchronous action system of "probing down - slightly biased selection - clipboard copy" point by point, realizing the splitting of the acquisition layer into three functional units working together. Therefore, it can reliably acquire interface text fragments without relying on software control properties.

[0055] In addition, refer to Figure 3 Furthermore, in one embodiment, the multi-point scanning decision maker includes a coordinate mapping unit and a step perturbation unit, and step S20 is refined into the following sub-steps: S200. Based on the anchor point coordinates of the current probe row, the coordinate mapping unit generates the basic sampling interval.

[0056] Specifically, the multi-point scanning decision-maker reduces the likelihood of missing long texts from a mathematical probability perspective through a non-uniform coordinate matrix generation algorithm. The baseline target point construction stage of the non-uniform probability coordinate matrix generation algorithm is completed by the coordinate mapping unit. The system uses the current probe row (with its logical ordinate set to 1) as the reference point. Feature interaction starting point coordinates Using the baseline target point, a horizontal acquisition vector containing 7 spatial contact points is constructed. This ensures the rationality of the sampling starting point.

[0057] S201. The sampling step size within the basic sampling interval is perturbed by the step perturbation unit based on the nonlinear discrete coefficient to generate a spatial contact sequence.

[0058] Specifically, the step perturbation unit completes the step size perturbation and coordinate calculation stages of the non-uniform probability coordinate matrix generation algorithm. To avoid the resonance sampling blind zone that may be caused by equidistant sampling and UI embedded icons (such as the usual 32×32px or 48×48px icons), the lateral coordinates of each touch point are... It abandons the traditional fixed step size method and instead adopts a feature scaling formula that introduces discrete coefficients: ; in, This is the base pixel stride for dynamic DPI calculation in current multi-screen systems, and For a predefined spatial nonlinear discrete sequence {-2.2,-1.0,0,0.8,1.5,2.3,3.4}, at least a few detection points can safely penetrate the occlusion area and accurately land on the lossless text segment within a wide threshold span of approximately -100px to +200px.

[0059] In addition, refer to Figure 4 Furthermore, in one embodiment, the asynchronous action execution unit includes a delay modulation unit and an action injection unit, and step S21 is refined into the following sub-steps: S210. The delay modulation unit generates timing delay parameters for each contact that follow a random distribution.

[0060] Specifically, after completing the calculation and projection of 7 independent coordinates, the engine in the delay modulation unit will inject an asynchronous delay sequence that follows a Markov chain distribution, so that the interactive behavior conforms to the natural rhythm of human operation.

[0061] S211. The motion injection unit performs a simulated double-click operation on each contact point in sequence according to the timing delay parameters.

[0062] Specifically, the action injection unit executes the anti-masking double-click action on each dot matrix in sequence according to the timing delay parameter, thereby inducing the underlying lexical boundary logic of the system to obtain the local clean text snapshot fragment stack that is segmented by the operating system level and thrown to the upper layer.

[0063] In summary, randomized delayed sequences enable automated operations to closely resemble normal human user interactions, effectively circumventing software security detection mechanisms and thus ensuring the legitimacy of the operation and application compatibility.

[0064] In addition, refer to Figure 5 Furthermore, in one embodiment, the cognitive layer includes a multi-frame comparison unit, a segment splicing unit, and a noise filtering unit, and step S3 is refined into the following sub-steps: S30. The overlapping area detection of multiple text segments is performed by the multi-frame comparison unit to determine the common substrings between the segments.

[0065] Specifically, the cognitive layer employs a multi-frame fingerprint end-to-end stitching and self-healing splicing engine. In this embodiment, the multi-frame comparison unit aggregates the clipboard string fragments asynchronously thrown by each discrete detection point to construct a raw data stream buffer pool. By detecting the overlapping areas between adjacent fragments, the boundaries of common substrings that can be used for splicing are determined.

[0066] S31. The adjacent segments are merged based on the common substring by the segment splicing unit to obtain the spliced ​​text.

[0067] Specifically, once the common substring is determined, the lossless splicing operator is triggered. ,in This represents performing clipping and fusion on the overlapping LCS intersections, thus restoring fragmented text to continuous content.

[0068] S32. Remove unexpected characters from the concatenated text using the noise filtering unit to obtain the complete text sequence.

[0069] Specifically, to completely resolve the issue caused by invisible control characters or custom graphic text forcibly separating the text... For extreme cases of text bridge breaks, this algorithm further incorporates a tolerance-based jump matching of sequence features. This means that within a noise tolerance of 1-3 placeholders in one-dimensional space, the algorithm uses the heat weight of edge characters to forcibly implement approximate feature vector bridging, ensuring that even UUIDs of up to 50 characters or extremely long nicknames containing multiple obscure languages ​​can be completely structured and restored to the database unit to remove unexpected characters from the spliced ​​text and obtain a pure and complete text sequence.

[0070] The combined efforts of the multi-frame comparison unit, the segment splicing unit, and the noise filtering unit have enabled seamless splicing and fusion of occluded text slices, ultimately solving the problem of data truncation caused by physical occlusion of text by interface icons.

[0071] In addition, refer to Figure 6 Furthermore, in one embodiment, step S30 is refined into the following sub-steps: The S300 multi-frame comparison unit uses dynamic programming to construct the character matching state space, and takes the longest common substring between detected segments as the overlapping region.

[0072] In this embodiment, the multi-frame comparison unit uses a longest common substring algorithm based on dynamic programming to search for logical bridges between segments. For any two spatiotemporally adjacent candidate text blocks S a With S b The system constructs in memory with a space complexity of O(n). Given the character matching state transition matrix M, calculate the length L of the longest common overlapping suffix-prefix. Determine if the condition is met. If the condition is met, the longest common substring is determined to be the overlapping region. In practice, the optimal overlap threshold was found to be 2-3 characters.

[0073] In summary, the multi-frame comparison unit can accurately identify overlapping parts between segments, providing precise fusion boundaries for subsequent splicing. Even when there are local differences in the text, it can still correctly identify overlapping areas, ensuring the accuracy and reliability of the splicing operation and improving the success rate of text restoration.

[0074] In addition, refer to Figure 7 Furthermore, in one embodiment, the output layer includes an anchor point identification unit, a field inference unit, and a structured mapping unit, and step S4 is refined into the following sub-steps: S40. The anchor point recognition unit identifies the unique identifier field from the complete text sequence based on the character encoding features, and positions the unique identifier field as the spatial origin.

[0075] Specifically, the output layer addresses the core challenges of data non-standardization and missing data cases through a self-healing dynamic anchor point and field relationship deduction algorithm. In this embodiment, the system abandons traditional fixed line number matching and instead uses a dynamic model where content determines identity.

[0076] This dynamic model requires structured extraction of core anchor points. Because the target interface often exhibits horizontal breaks or blank spaces in the data, the system abandons the traditional fixed line numbers and coordinate mapping, instead using a complete text sequence stitched together from multiple frames. The multidimensional character feature logic operator is used to dynamically search for the unique identifier field (UID) and establish it as the origin of the field space (i.e., the lower-level axis).

[0077] S41. Based on the relative position of the spatial origin, the field deduction unit deduces the attribution of other attribute fields.

[0078] Specifically, the field deduction unit models field relationships, deduces the attribution of other attribute fields based on the relative position of the spatial origin, and utilizes the spatial consistency of the UI design to achieve field placement. Once the UID physical row is locked, the system immediately establishes a new row based on that row. Relative coordinate system of axes: Scope: Search by name and supplementary description (Nickname). If there are logically separated lines, they are associated with nicknames.

[0079] Field: Search for the region / property.

[0080] Furthermore, the field deduction unit possesses self-healing logic. That is, when a missing attribute causes sequence collapse (e.g., no nickname), the system will detect Nickname == UID through field semantic conflict detection. At this point, the system triggers self-healing compensation, automatically clearing the misjudged redundant fields and ensuring the atomicity of the data flow.

[0081] S42. The spatial origin and other attribute fields are mapped to a preset data structure through the structured mapping unit to obtain structured data.

[0082] Specifically, the structured mapping unit injects the non-standard flat key-value pairs calculated by the cognitive layer into the local field relation projection model, forcibly transforming and mapping them into structured object data that satisfies the third normal form (3NF). Before submission, it performs cross-field logical consistency conflict detection, and checks the confidence level of any core attribute. If the value is below the preset threshold, a closed-loop backtracking collection action is immediately triggered, ultimately achieving adaptive parsing in the case of missing fields.

[0083] In addition, refer to Figure 8 Furthermore, in one embodiment, the anchor point recognition unit includes a feature calculation unit and a filtering unit. The step S40, which involves the anchor point recognition unit identifying a unique identifier field from a complete text sequence based on character encoding features, is further refined into the following sub-steps: S400: The feature calculation unit calculates and obtains the character purity index and regular expression matching index for the candidate substrings in the complete text sequence.

[0084] Specifically, This represents the total number of characters in the substring that fall entirely within the standard range of legal English letters, numbers, underscores, and hyphens. This represents the total length of the substring displayed in the layout. The algebraic term representing the ratio between the two. The purity of the underlying ASCII characters of the candidate word was objectively measured. The higher the purity, the higher the probability that it carries the system UID.

[0085] For a business structure hard-matching function based on Turing-complete expressions, its discrete value space is: When the candidate word fully meets the underlying validation rules set by the target third-party communication software for the UID, such as when the UID has strict attributes like "must start with an English letter" or "the entire string cannot contain any Chinese characters or spaces," a boolean truth value of 1 is returned.

[0086] S401. Select the unique identifier field based on the weighted result of the character purity index and the regular expression matching index through the filtering unit.

[0087] Specifically, to accurately filter out mixed and noisy nicknames, pure Chinese characters, and emoticons, the system extracts a set of candidate string sequences. The following core confidence level determination formula is introduced: ; The specific physical meanings of each calculation factor and parameter in the above core calculation formula are distributed as follows: Indicates the first candidate substrings The final overall confidence weight is established as the unique absolute UID (such as WeChat ID) anchor point for this contact. The system will select the item with the highest value as the benchmark at the end of the calculation.

[0088] This is a penalty feature value decay term for unexpected character sets. This algebraic term is specifically used to identify and penalize disguised strings that appear similar to the system UID but are internally user-defined non-standard nickname attributes, such as complex nickname paragraphs with special full-width characters, Martian prefix combinations, or system-invisible control characters.

[0089] This represents the weight hyperparameter vector under neural network adaptive feedback or manual calibration, indicating the algorithm's computational power preference allocation for the three extraction dimensions mentioned above. For example, the optimal recommendation threshold empirical group, after large-sample calibration, is usually set as... .

[0090] The system's computation engine performs a multi-threaded exhaustive search of the current solution cache space, prioritizing and filtering out those that meet the requirements. Furthermore, candidate words with the largest positive peak in absolute score are used as structured physical calibration anchors. The comprehensive introduction of this mathematical model enables this invention to achieve extremely high accuracy and dynamic fault tolerance to prevent deviation. In other words, even in extremely incomplete layouts where the target has not actively set a nickname, the region is left blank, or key attribute entries are severely missing in large areas, this verification algorithm still has strong autonomous addressing capabilities. This ensures that fatal data misplacement disasters will not occur when extracting the relative displacement of subsequent fields, thus building an absolute algorithmic barrier for the robustness of the parsing system.

[0091] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0092] This application also provides a third-party communication software data extraction system based on non-intrusive detection, which corresponds one-to-one with the third-party communication software data extraction method based on non-intrusive detection in the embodiments.

[0093] refer to Figure 9 A data extraction system for third-party communication software based on non-invasive detection includes: a sensing module 1, a data acquisition module 2, a cognitive module 3, and an output module 4. Detailed descriptions of each functional module are as follows: Perception module 1 is used to acquire the visual heatspots of the target software interface; Acquisition module 2 is used to construct a spatially redundant sampling matrix within the visual hotspot and simulate interactive operations to acquire multiple text fragments; Cognitive module 3 is used to perform multi-frame comparison and splicing of multiple text fragments to obtain a complete text sequence; Output module 4 is used to perform structured parsing on the complete text sequence and obtain structured data that conforms to a preset format.

[0094] The system comprises the following modules: Perception Module 1 acquires the visual hotspots of the target software interface, defining a precise working range for subsequent data acquisition operations and ensuring that interactive actions accurately land on the effective area; Acquisition Module 2 constructs a spatially redundant sampling matrix within the visual hotspots, simulating mouse clicks and double-clicks to acquire text fragments from multiple sampling points, achieving physical-layer acquisition of the interface text; Cognition Module 3 performs multi-frame comparison, overlap detection, and splicing fusion on the multiple fragmented text fragments acquired by the acquisition module to reconstruct a complete text sequence and filter out unexpected characters; Output Module 4 performs structured parsing on the complete text sequence generated by the cognition module, identifying unique identifier fields and deducing the attribution of other attribute fields, ultimately outputting structured data conforming to a preset format. Through the combined action of these modules, the system can adapt to dynamic interface changes without intruding into the underlying software, resolving issues of missing or misaligned fields and improving the accuracy of data reconstruction.

[0095] Specific limitations regarding the third-party communication software data extraction system based on non-intrusive detection can be found in the context of limitations on the third-party communication software data extraction method based on non-intrusive detection, and will not be repeated here. Each module in the aforementioned third-party communication software data extraction system based on non-intrusive detection can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the electronic device, or stored in the memory of the electronic device as software, so that the processor can call and execute the operations corresponding to each module. In one embodiment, an electronic device is provided, which is a user terminal. (Reference) Figure 10The electronic device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and the database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores detection data tables. The network interface allows communication with external terminals via a network connection. When executed by the processor, the computer program implements a data extraction method for third-party communication software based on non-intrusive detection.

[0096] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: S1. Obtain the visual hotspots of the target software interface through the perception layer.

[0097] S2. A spatially redundant sampling matrix is ​​constructed within the visual hot zone through the acquisition layer to simulate interactive operations and obtain multiple text fragments.

[0098] S3. The cognitive layer performs multi-frame comparison and splicing of multiple text fragments to obtain a complete text sequence.

[0099] S4. The output layer performs structured parsing on the complete text sequence to obtain structured data that conforms to the preset format.

[0100] In one embodiment, the acquisition layer includes a multi-point scanning decision-maker, an asynchronous action execution unit, and a clipboard capture unit. Step S2 is further refined into the following sub-steps: S20. A sequence of spatially nonlinearly distributed contact points in the visual hot zone is generated by a multi-point scanning decision unit.

[0101] S21. Simulated interactive operations are performed on each contact point in the spatial contact sequence through the asynchronous action execution unit, triggering text selection and copying actions at the operating system level.

[0102] S22. Obtain the text fragment corresponding to each copy operation through the clipboard capture unit.

[0103] In one embodiment, the multi-point scanning decision unit includes a coordinate mapping unit and a step perturbation unit, and the sub-steps refined in step S20 include: S200. Based on the anchor point coordinates of the current probe row, the coordinate mapping unit generates the basic sampling interval.

[0104] S201. The sampling step size within the basic sampling interval is perturbed by the step perturbation unit based on the nonlinear discrete coefficient to generate a spatial contact sequence.

[0105] In one embodiment, the asynchronous action execution unit includes a delay modulation unit and an action injection unit, and the sub-steps refined in step S21 include: S210. The delay modulation unit generates timing delay parameters for each contact that follow a random distribution.

[0106] S211. The motion injection unit performs a simulated double-click operation on each contact point in sequence according to the timing delay parameters.

[0107] In one embodiment, the cognitive layer includes a multi-frame comparison unit, a segment splicing unit, and a noise filtering unit, and the refined sub-steps of step S3 include: S30. The overlapping area detection of multiple text segments is performed by the multi-frame comparison unit to determine the common substrings between the segments.

[0108] S31. The adjacent segments are merged based on the common substring by the segment splicing unit to obtain the spliced ​​text.

[0109] S32. Remove unexpected characters from the concatenated text using the noise filtering unit to obtain the complete text sequence.

[0110] In one embodiment, the sub-steps of step S30 are further refined as follows: The S300 multi-frame comparison unit uses dynamic programming to construct the character matching state space, and takes the longest common substring between detected segments as the overlapping region.

[0111] In one embodiment, the sub-steps of step S4 refinement include: S40. The anchor point recognition unit identifies the unique identifier field from the complete text sequence based on the character encoding features, and positions the unique identifier field as the spatial origin.

[0112] S41. Based on the relative position of the spatial origin, the field deduction unit deduces the attribution of other attribute fields.

[0113] S42. The spatial origin and other attribute fields are mapped to a preset data structure through the structured mapping unit to obtain structured data.

[0114] In one embodiment, the sub-steps of step S40 are further refined as follows: S400: The feature calculation unit calculates and obtains the character purity index and regular expression matching index for the candidate substrings in the complete text sequence.

[0115] S401. Select the unique identifier field based on the weighted result of the character purity index and the regular expression matching index through the filtering unit.

[0116] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0117] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

Claims

1. A method for extracting data from third-party communication software based on non-invasive detection, characterized in that, include: The visual heatspots of the target software interface are obtained through the perception layer; By constructing a spatially redundant sampling matrix within the visual hotspot using the acquisition layer, multiple text fragments are obtained by simulating interactive operations. The cognitive layer performs multi-frame comparison and splicing of the multiple text fragments to obtain a complete text sequence; The complete text sequence is structured and parsed through the output layer to obtain structured data that conforms to a preset format. The acquisition layer includes a multi-point scanning decision unit, an asynchronous action execution unit, and a clipboard capture unit. The step of constructing a spatially redundant sampling matrix within the visual hotspot through the acquisition layer to simulate interactive operations and acquire multiple text fragments includes: The multi-point scanning decision-maker generates a sequence of spatially nonlinearly distributed spatial contact points within the visual thermal zone. The asynchronous action execution unit performs simulated interactive operations on each touch point in the spatial touch point sequence, triggering text selection and copying actions at the operating system level; The clipboard capture unit acquires the text fragment corresponding to each copy operation.

2. The method according to claim 1, characterized in that, The multi-point scanning decision unit includes a coordinate mapping unit and a step perturbation unit. The step of generating a spatially nonlinearly distributed sequence of spatial touch points in the visual hot zone using the multi-point scanning decision unit includes: The coordinate mapping unit generates a basic sampling interval based on the anchor point coordinates of the current probe row; The spatial contact sequence is generated by perturbing the sampling step size within the basic sampling interval using the step perturbation unit based on nonlinear discrete coefficients.

3. The method according to claim 1, characterized in that, The asynchronous action execution unit includes a delay modulation unit and an action injection unit. The step of performing simulated interactive operations on each contact point of the spatial contact sequence through the asynchronous action execution unit includes: The delay modulation unit generates timing delay parameters for each contact point that follow a random distribution. The motion injection unit sequentially performs a simulated double-click operation on each contact point according to the timing delay parameters.

4. The method according to claim 1, characterized in that, The cognitive layer includes a multi-frame comparison unit, a segment splicing unit, and a noise filtering unit. The step of performing multi-frame comparison and splicing of the multiple text segments through the cognitive layer to obtain a complete text sequence includes: The multi-frame comparison unit performs overlapping region detection on the multiple text segments to determine the common substrings between the segments. The fragment splicing unit merges adjacent fragments based on the common substring to obtain spliced ​​text; The noise filtering unit removes unwanted characters from the concatenated text to obtain the complete text sequence.

5. The method according to claim 4, characterized in that, The step of detecting overlapping regions of the multiple text segments and determining common substrings among the segments through the multi-frame comparison unit includes: The multi-frame comparison unit uses dynamic programming to construct the character matching state space and takes the longest common substring between detected segments as the overlapping region.

6. The method according to claim 1, characterized in that, The output layer includes an anchor point recognition unit, a field inference unit, and a structured mapping unit. The step of performing structured parsing of the complete text sequence through the output layer to obtain structured data conforming to a preset format includes: The anchor point recognition unit identifies a unique identifier field from the complete text sequence based on character encoding features, and positions the unique identifier field as the spatial origin. Based on the relative position of the spatial origin, the field deduction unit infers the attribution of other attribute fields; The structured mapping unit maps the spatial origin and other attribute fields to a preset data structure to obtain the structured data.

7. The method according to claim 6, characterized in that, The anchor point recognition unit includes a feature calculation unit and a filtering unit. The step of identifying a unique identifier field from the complete text sequence based on character encoding features by the anchor point recognition unit includes: The feature calculation unit calculates and obtains the character purity index and regular expression matching index for the candidate substrings in the complete text sequence, respectively. The unique identifier field is selected by the filtering unit based on the weighted result of the character purity index and the regular expression matching index.

8. A third-party communication software data extraction system based on non-invasive detection, characterized in that, The system for performing the method of claim 1 includes: The perception module (1) is used to acquire the visual hotspots of the target software interface; The acquisition module (2) is used to construct a spatially redundant sampling matrix within the visual hot zone and simulate interactive operations to acquire multiple text fragments. The cognitive module (3) is used to perform multi-frame comparison and splicing on the multiple text fragments to obtain a complete text sequence; The output module (4) is used to perform structured parsing on the complete text sequence to obtain structured data that conforms to a preset format.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as any one of the non-intrusive detection-based third-party communication software data extraction methods as described in claims 1 to 7.

Images