Privacy computing-based AI decision system data processing method and system

Abstract

The invention relates to the technical field of data processing, in particular to an AI decision-making system data processing method and system based on privacy computing, and the method comprises the following steps: extracting texture segments and behavior fields, splitting positions and arranging a sequence, executing field interval comparison, screening difference fields and merging and compressing, extracting interval fields and executing hop segment screening. And comparing the repeated structures to generate fingerprint fragments, and constructing a merging sequence to generate an AI decision system data processing scheme. According to the method, sequential fusion of images and behavior fields is achieved through a unified field index structure, a cross-dimension mapping relation is constructed through field value intervals to improve the semantic association ability, similar fields are screened through entropy difference calculation to execute merging compression to enhance the structure expression efficiency, and the behavior track recognition ability is enhanced through hop segment screening and rearrangement. Fingerprint fragments are extracted through repeated structure comparison, multi-level merging and sequence compression are achieved, and the data organization level and the structure response flexibility are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a data processing method and system for AI decision-making systems based on privacy computing. Background Technology

[0002] The field of data processing technology encompasses the entire process of collecting, storing, transmitting, processing, and managing raw data. It covers core aspects such as data cleaning, classification, analysis, modeling, storage optimization, access control, and security protection. Widely applied in industries such as finance, healthcare, manufacturing, and government, it forms the foundation for supporting information system operation, intelligent decision-making, and big data mining. Within this technological field, with the development of artificial intelligence, data processing has gradually evolved towards higher efficiency, greater intelligence, and enhanced privacy protection. This is particularly true in multi-party collaborative data scenarios, where the requirements for data security and privacy are especially prominent. Traditional AI decision-making system data processing methods refer to the data operation processes such as data preprocessing, feature extraction, label generation, and model input transformation adopted in artificial intelligence applications to support algorithm training and inference decisions. They typically involve centralized data collection, storing raw datasets on local servers or cloud platforms for unified processing. In this process, traditional data partitioning management, static data anonymization, access restrictions, or encrypted transmission methods are mainly used to address data leakage and privacy risks. In scenarios where multiple parties cannot trust each other's data or where strict data privacy protection is required, processing methods are often based on static anonymization rules, anonymization strategies, or differential privacy to protect personal data. At the same time, machine learning models are used to build decision engines, but problems such as data silos, low efficiency in secure sharing, and limited modeling accuracy still exist.

[0003] Existing technologies use preset rules to statically partition and desensitize raw information when processing data, making it difficult to map dynamically evolving behavioral characteristics in detail. The lack of linkage mechanisms between fields results in fragmented information structure expression. Field sorting and combination are monotonous and cannot reflect the inherent correspondence between behavioral paths and image features. Data compression processing relies solely on general algorithm rules and lacks quantitative identification and reconstruction strategies for differences between fields. Jump structures between fields are not effectively identified and reorganized, resulting in coarse data organization granularity, insufficient contextual continuity, and repeated structures failing to form abstract expressions in the sequence, thus limiting the ability to deeply extract and structurally utilize decision data. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a data processing method for an AI decision-making system based on privacy computing, comprising the following steps: S1: Extract texture fragments from the image and structural fields from the behavior, call the corresponding texture location points and behavior record items to segment and split them, organize the positions according to the order of field appearance, rearrange them according to the index continuity, and generate a normalized field index structure. S2: Based on the normalized field index structure, extract the texture field value and behavior field value, compare them sequentially according to the interval between their occurrences, reorganize the field positions according to the comparison results, and generate the corresponding union set of the fields; S3: Based on the joint set corresponding to the fields, call the pathological texture field and the behavior index field, calculate the entropy difference of the field value distribution, filter similar field pairs according to the difference magnitude and linearly merge them to generate a field compressed expression sequence; S4: Based on the field sorting relationship of the compressed expression sequence, generate the flow trajectory of continuous positions of the fields, extract the interval fields and perform jump segment filtering according to the interval size, rearrange the jump segment positions in order, and form a jump segment index arrangement sequence; S5: Call the segment number in the segment index arrangement sequence, perform structural comparison according to the repetition order of the number, merge the repetitive structures into fingerprint segments according to the appearance direction, construct the merged fingerprint set according to the segment arrangement, and generate the AI ​​decision system data processing scheme.

[0005] As a further embodiment of the present invention, the normalized field index structure includes a texture location point index, behavior record item number, field order arrangement, and unified sequence label; the field corresponding union set includes a texture field mapping group, a behavior field mapping group, and a field parallel relationship identifier; the field compressed expression sequence includes difference close field pairs, merged field labels, and compressed field sorting information; the jump segment index arrangement sequence includes a field jump segment number, a jump segment interval identifier, and a jump segment order mark; and the AI ​​decision system data processing scheme includes a repeating structure fragment number, a fragment merging index, and an AI decision reference unit.

[0006] As a further aspect of the present invention, the step of filtering similar field pairs based on the magnitude of difference and linearly merging them refers to filtering out the different field pairs based on the entropy difference of the field value distribution and linearly merging them according to the value trend.

[0007] As a further aspect of the present invention, the index continuity refers to the fields being arranged in the original order of appearance in the structure without any missing parts, thus verifying a continuous index number sequence.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Based on the image-end texture fragments and behavior-end structure fields, extract the index values ​​of texture location points and behavior record items, call the two types of index values ​​to perform segmentation and splitting, and reposition the fields according to their order in the original sequence to generate a field position mapping table; S102: Based on the field order in the field position mapping table, combine the texture and behavior fields to form a field fragment sequence, extract the index value, and generate a field rearrangement index set; S103: Call the index values ​​in the field rearrangement index set, perform continuous rearrangement, combine the fields into a unified sequence structure, and generate a normalized field index structure.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Call the field items in the normalized field index structure, retrieve the texture field value and behavior field value respectively, locate the position of the two types of fields in the sequence based on the field index, establish a correspondence table between field value and position index, and generate a field value index mapping table; S202: Based on the field position index in the field value index mapping table, calculate the interval between the texture field value and the behavior field value in the sequence, and match the interval value with the set field order comparison rules to establish a field combination group that conforms to the order relationship and generate a field order matching set; S203: Call the field order matching set, rearrange and organize the fields according to their parallel relationship in the original index structure, combine and aggregate the related fields into corresponding items, and generate a joint set of fields.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the union set corresponding to the fields, call the pathological texture field and the behavior index field, count the frequency distribution range of the field values, calculate the dispersion of the field based on the distribution range, and obtain the information on the distribution of the two types of fields to generate a field frequency difference matrix. S302: Based on the entropy difference value of the field pairs in the field frequency difference matrix, determine whether the difference magnitude is less than the preset entropy difference proximity threshold, filter all field pairs that meet the conditions, mark the original field index, and generate a set of field pairs with close entropy differences. S303: Call the field pairs in the entropy difference close field pair set, perform linear mapping fusion operation according to the field index order, and serialize and arrange the fusion result to generate a field compressed expression sequence.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Call the field items in the field compression expression sequence, construct the continuous flow path of the fields in the structure according to the index sorting relationship in the sequence, and generate the field flow trajectory sequence based on the direction and magnitude of index change; S402: Based on the field flow trajectory sequence, extract the index difference between field positions, extract the interval field index that is greater than the preset jump threshold, filter and judge according to the interval size, and mark the original sequence number of the jump field to generate a jump field index set; S403: Call the field indexes in the jump field index set, rearrange them according to the order of appearance in the compressed sequence, and build a continuous structure for all jump field indexes to generate a jump index arrangement sequence.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Call all the jump segment numbers in the jump segment index arrangement sequence, identify the repeated field arrangement structure based on the order of the numbers in the structure, compare the structural form of the corresponding field values ​​of the same number in the sequence, and extract the structural combination with consistent features to generate a repeated structure comparison set. S502: Based on the field arrangement direction of the repeated structure comparison set, perform field-level aggregation on the structure groups with the same direction, call the start and end positions of each group of structures in the sequence to complete the aggregation process, and construct the fragment structure to generate a merged fingerprint fragment set. S503: Call the fingerprint fragments in the merged fingerprint fragment set, perform structural continuation and sequence reconstruction according to the arrangement relationship in the original fragment index sequence, integrate all fragments and map them to the decision process to generate an AI decision system data processing scheme.

[0013] The data processing system for AI decision-making based on privacy-preserving computation includes: The field index normalization module is used to implement S1: extract texture fragments from the image end and structural fields from the behavior end, call the corresponding texture location points and behavior record items to segment and split, normalize the positions according to the order of field appearance, rearrange according to index continuity, and generate a normalized field index structure. The field union matching module is used to implement S2: Based on the normalized field index structure, extract the texture field value and behavior field value, compare them in order according to the interval between their occurrence, reorganize the field positions according to the comparison results, and generate the field corresponding union set; The field entropy difference merging module is used to implement S3: based on the joint set corresponding to the fields, the pathological texture field and the behavior index field are called to calculate the entropy difference of the field value distribution, and similar field pairs are selected and linearly merged according to the difference magnitude to generate a field compressed expression sequence; The jump segment trajectory rearrangement module is used to implement S4: based on the field sorting relationship of the field compressed expression sequence, generate the flow trajectory of continuous field positions, extract the interval field and perform jump segment filtering according to the interval size, rearrange the jump segment positions in order, and form a jump segment index arrangement sequence; The fingerprint structure generation module is used to implement S5: call the jump segment number in the jump segment index arrangement sequence, perform structure comparison according to the repeating order of the number, merge the repeating structures into fingerprint segments according to the appearance direction, construct the merged fingerprint set according to the segment arrangement, and generate the AI ​​decision system data processing scheme.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a unified field index structure is used to achieve sequential fusion of image and behavior fields. Cross-dimensional mapping relationships are constructed through field value intervals to enhance semantic association capabilities. Similar fields are selected through entropy difference calculation to perform merging and compression, thereby enhancing structural expression efficiency. Behavioral trajectory recognition capabilities are strengthened through segment selection and rearrangement. Multi-level merging and sequence compression are achieved through repeated structure comparison to extract fingerprint fragments, thereby improving the data organization hierarchy and structural response flexibility. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Please see Figure 1 This invention provides a data processing method for an AI decision-making system based on privacy computing, comprising the following steps: S1: Obtain the image end texture fragment and behavior end structure field, call the corresponding value of texture position point and behavior record item to segment and split, perform position normalization on the two types of information according to the field appearance order, rearrange all fields according to index continuity and group them into the same sequence to generate a normalized field index structure. S2: Call the field items in the normalized field index structure to obtain the texture field value and behavior field value. Perform a comparison based on the interval between the occurrence of the two types of values. Reorganize the parallel positions of the fields in the index according to the comparison results, construct the correspondence between the fields, and generate the field corresponding union set. S3: Call the pathological texture field and behavior index field in the union set corresponding to the field, perform entropy difference calculation based on the field value distribution, determine the difference between fields according to the calculation results, filter fields with similar differences and perform linear merging, and generate field compressed expression sequence; S4: Call the field compression expression sequence to express the sorting relationship, generate the flow trajectory based on the continuous position of the field in the sequence, obtain the interval field and perform jump segment filtering according to the interval size, rearrange the jump segment positions in order, and generate a jump segment index arrangement sequence. S5: Call all the segment numbers in the segment index arrangement sequence, perform structural comparison according to the order of repeated numbers, merge the repeated structures into fingerprint fragments according to the direction of appearance, construct the merged fingerprint set according to the arrangement relationship of the fragments in the sequence, and generate the AI ​​decision system data processing scheme.

[0023] The normalized field index structure includes texture location point index, behavior record item number, field order arrangement and unified sequence label. The field corresponding union set includes texture field mapping group, behavior field mapping group and field parallel relationship identifier. The field compressed expression sequence includes difference close field pairs, merged field label and compressed field sorting information. The jump segment index arrangement sequence includes field jump segment number, jump segment interval identifier and jump segment order mark. The AI ​​decision system data processing scheme includes repeated structure fragment number, fragment merging index and AI decision reference unit.

[0024] Please see Figure 2 The specific steps of S1 are as follows: S101: Based on the image-end texture fragments and behavior-end structure fields, extract the index values ​​of texture location points and behavior record items, call the two types of index values ​​to perform segmentation and splitting, and reposition the fields according to their order in the original sequence to generate a field position mapping table; First, the original image data is read frame by frame and processed into a grid. Image frames are divided into several texture unit regions of uniform size; for example, each frame can be divided into 64×64 pixel regions, forming consecutively numbered texture fragment regions. Then, the grayscale distribution, edge gradient direction, and edge intensity value are extracted from each texture fragment, and each texture fragment is numbered T1, T2, etc., and its spatial location information is recorded. Behavioral structure fields are extracted from user interaction logs, such as operation time, location, and interaction target identifier fields from click events, and start and end time, trajectory path, and target area fields from swiping behaviors. These fields are numbered F1, F2, etc., according to the order of behavior occurrence. Then, two sets of index values ​​are established for the texture end and the behavioral end respectively. Next, the texture location point index value and the corresponding behavior record item index value are combined according to the event occurrence sequence, such as when the user is in 1.2 If a click event F5 occurs at a certain time and texture fragment T27 is captured in the image frame at that moment, the mapping relationship is recorded as [T27, F5]. The matching and recording of texture fragments corresponding to all events is completed in sequence. Then, the original data is segmented according to the trigger time order in the behavior field. The segmentation standard is to divide the image sequence into multiple time periods based on the behavior event time point. Only the texture fragments corresponding to the triggered behavior in each time period are retained. By matching the texture and behavior of each data segment field by field, a preliminary field positioning relationship is constructed. Finally, a well-structured field position mapping table is generated by combining the order of field triggering. For example, if fields F5, F7, and F8 correspond to texture fragments T27, T35, and T41 respectively, and are arranged in the sequence as the 1st, 2nd, and 3rd positions, the table entries can be constructed as F5→1, F7→2, F8→3, forming a field position mapping table.

[0025] S102: Based on the field order in the field position mapping table, combine the texture and behavior fields to form a field fragment sequence, extract the index values, and generate a set of field rearrangement indexes; Extract the texture fragments corresponding to each behavior field sequentially, and combine and concatenate the texture fragments with the behavior fields according to the sequence number in the record to construct a field fragment sequence. The combination method is to take the behavior field F as the main element and use the corresponding texture fragment T as an auxiliary element for unified structuring. For example, if the behavior field F5 corresponds to texture T27 and the field F7 corresponds to texture T35, then the fragment items can be [F5, T27] and [F7, T35]. After completion, an initial field fragment sequence is formed. Next, the position value P0 of each field fragment in the sequence is read sequentially in the original field position sequence, and according to the target position value P1 specified in the field position mapping table, each fragment is... The current position of a segment is correlated with the target position, that is, the field rearrangement index value is recorded. For example, if the 3rd field in the original sequence needs to be moved to the 1st position, an index value pair (3, 1) is formed. This process is repeated to generate a complete field rearrangement index set for all field segments. This set can be represented as several position mapping pairs such as [(3, 1), (1, 2), (2, 3)], which means that the field that was originally in the 3rd position should be placed in the 1st position, the field that was in the 1st position should be moved to the 2nd position, and so on. The generation of this set depends on the position information in the mapping table rather than the original input order. Therefore, the rearrangement index set has a stable order control capability and can be used for subsequent unified field structure reorganization processing.

[0026] S103: Call the index values ​​in the field rearrangement index set, perform continuous rearrangement, combine the fields into a unified sequence structure, and generate a normalized field index structure; The process involves sequentially reordering the field fragments. This is done by creating an empty array of equal length to the number of field fragments. Each original field fragment is then inserted into the array according to its target position in its rearranged index pair. For example, the field fragment originally at index 3 is moved to index 1, and the fragment originally at index 2 is moved to index 0. This process is repeated until all field fragments are reordered, creating a unified sequence structure. The resulting field group sequence has a consistent arrangement logic and follows the field order requirements defined in the field position mapping table. After sorting, each field fragment is recorded... The structured data, such as the new position number, field number, and texture fragment number in the unified sequence, are recorded to form a unified normalized field index structure table. For example, the unified structure table records: the sequence number 0 corresponds to the field F5 and the texture fragment T27, the sequence number 1 corresponds to the field F7 and the texture fragment T35, etc. The structure table is represented in the form of key-value pairs as {0: [F5, T27], 1: [F7, T35], 2: [F8, T41]}. This structure table constitutes the final unified normalized field index structure, which facilitates field identification and order restoration in the subsequent multi-field fusion processing.

[0027] Please see Figure 3 The specific steps of S2 are as follows: S201: Call the field items in the normalized field index structure, retrieve the texture field value and behavior field value respectively, locate the position of the two types of fields in the sequence based on the field index, establish a correspondence table between field value and position index, and generate a field value index mapping table; First, the records in the structure table need to be traversed item by item. Each item contains a texture field number, an action field number, and the sequence position of the combination in the normalized structure. For each record, the texture field value is located from the texture dataset. Using the texture field number provided in the index structure, the texture fragment region is located within the image frame data, and the pixel values ​​of this region are read. Based on the read grayscale information, edge gradient direction, color principal components, etc., a complete texture field value information item is constructed. For example, for a texture field with the number T301, the corresponding region is in the 4th row and 6th column of the 12th frame image. After reading the pixels, the average grayscale value of this texture is extracted to be 130, the edge direction is horizontal, and the color principal component is blue. Therefore, the texture field value is set to [130, horizontal, blue]. Then, the corresponding action field number is read, and the event corresponding to this number is found in the user action record table. The field values ​​are extracted to include information such as behavior type, interaction target ID, and trigger timestamp. For example, if the behavior field record for ID F45 is "slide, target ID: Z23, time: 5.2 seconds", then the behavior field value is [slide, Z23, 5.2 seconds]. After extracting the texture field value and the behavior field value, a correspondence between the field value and the index position provided in the normalized structure is established to form a one-to-one mapping structure. For example, if the texture field value is in the 8th position and its corresponding behavior field value is in the 9th position, then the record is {[130, horizontal, blue]: 8, [slide, Z23, 5.2 seconds]: 9}. In this way, the retrieval and extraction process of all field items is completed, and a mapping relationship is established between all field values ​​and their index positions. Finally, a field value index mapping table is generated for subsequent position information comparison and relationship matching operations between fields.

[0028] S202: Based on the field position index in the field value index mapping table, calculate the interval between the texture field value and the behavior field value in the sequence, and match the interval value with the set field order comparison rules to establish a field combination group that conforms to the order relationship and generate a field order matching set; The position index between each pair of texture field values ​​and behavior field values ​​is read. The interval value within the unified field sequence is calculated by subtracting the sequence positions of the two fields. For example, if the index of the texture field value [130, horizontal, blue] is 8, and the index of its corresponding behavior field value [slide, Z23, 5.2 seconds] is 9, then the interval value is 1, recorded as the spacing parameter for this field combination. This operation is performed on all field value pairs to form an interval record table. Then, each pair of field interval values ​​is compared against a pre-defined field order comparison rule. This rule is preset based on statistical analysis of frequently occurring effective behavior-texture combinations in historical data, or is a rule library pre-defined by domain experts based on business logic. For example, if it is required that the interval between click-type behaviors and the preceding main texture field should be less than or equal to 2, then it is determined whether the current field pair meets the "texture field" rule. The condition "before the behavior field, after the behavior field, with an interval ≤ 2" requires simultaneous verification of whether the behavior type field value is "click" and confirmation of the field position order. For example, if the texture field is at index 6, the behavior field is at index 8, the interval is 2, and the behavior field value type is "click", then the field pair meets the comparison rules and is recorded as a successfully matched field pair. If the condition is not met, it is ignored. In this way, all field combinations that meet the preset order logic are filtered out, and they are combined into a field combination structure according to the field value pair, interval value, and matching rule number. Each structure is recorded as "field value pair + interval value + rule matching result", such as {([grayscale 120, vertical, green], [click, OBJ99, 3.1 seconds]), interval: 2, matching rule: R1}. All successfully matched field combinations are summarized, and finally a field order matching set is generated, recording the set of all field value pairs that conform to the set logic in terms of structural order.

[0029] S203: Call the field order matching set, rearrange and organize the fields according to their parallel relationship in the original index structure, combine and aggregate related fields into corresponding items, and generate a field-corresponding union set; For each successfully matched field value pair, its sequence position number in the normalized field index structure must first be obtained, and the corresponding field number information must be read. Then, the field grouping information recorded in the original field sequence structure is combined to determine whether these fields have a parallel relationship. Specifically, the determination method is to check whether the behavior event number or image frame number to which the field number belongs is consistent. If the texture field and the behavior field come from the same image frame or the same user interaction event number, then the field value pair is considered to have a parallel attribute and is marked as parallel. When performing the parallel judgment operation, the event source number of each pair of field values ​​is compared. For example, if the texture field T301 and the behavior field F45 are both identified as originating from event E17, then the field combination relationship under E17 is recorded, and then the field is processed. The aggregation operation concatenates all field values ​​belonging to the same event number or image frame number to form an aggregated field item. Each aggregated item uniformly records its event number, the list of field values ​​it contains, and its position identifier in the normalized field sequence. For example, the aggregated item record corresponding to event E17 is {E17: {[130, horizontal, blue], [slide, Z23, 5.2 seconds]}, index position: [8, 9]}. This aggregation operation is performed on all field combinations in the field order matching set. Finally, the set of all field value pairs belonging to the same event or image frame number is summarized to complete the generation of the field corresponding union set. The structure is that the event number index is mapped to the set of field value pairs, which is used as a structural support data source in subsequent structural reasoning or cross-modal behavior modeling scenarios.

[0030] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the field-corresponding union set, call the pathological texture field and the behavior index field, count the frequency distribution range of the field values, calculate the dispersion of the field based on the distribution range, and obtain the information on the distribution of the two types of fields to generate a field frequency difference matrix. First, extract the field pairs one by one according to the event number. Then, read the specific values ​​of the pathological texture field and the behavior index field in each group. For the pathological texture field, count the frequency of discrete attributes such as color category, texture shape, and edge direction in all field pairs, and divide them into fixed intervals. For example, if there are four color categories: red, purple, blue, and yellow, count the occurrence frequency of each color category in the field union set. Assume red appears 22 times, purple 18 times, blue 30 times, and yellow 30 times, with a total of 1 field pair. If 0 is found, the frequencies of each color in the overall set are calculated to be 0.22, 0.18, 0.30, and 0.30, respectively. For behavior fields, the behavior types such as click, swipe, and zoom are extracted, and their occurrence counts in the union set are counted. For example, if clicks occur 28 times, swipes 42 times, and zooms 30 times, the frequencies are calculated to be 0.28, 0.42, and 0.30, respectively. After the frequency statistics are completed, the above statistical data are divided into pre-defined distribution intervals, such as 0–0.2, 0.2–0.4, and 0.4–0.6. Next... This method measures the dispersion of the frequency distribution of each field across different intervals. Specifically, it compares the frequency differences of each field category within its respective interval, obtains the magnitude of frequency changes, and calculates the uniformity of its distribution across intervals. For example, if the color frequency distribution of a texture field is highly concentrated in a certain interval, it indicates low dispersion; if multiple values ​​have a wide frequency distribution, the dispersion is high. Following this logic, the dispersion of pathological texture fields and behavioral fields is measured separately. The dispersion is determined by statistically analyzing the absolute value of the frequency difference between various values. For example, for the color field, its maximum frequency is 0.30 and its minimum frequency is 0.18, so the frequency difference is 0.12. Similarly, the maximum and minimum frequency differences are calculated for behavioral fields, and the frequency difference value of each type of field is recorded. Finally, the frequency difference of the pathological texture field and the frequency difference of the behavioral field in the same field pair are subtracted, and the result is recorded as the information distribution difference value of the field pair. The results of all field pairs are organized into a matrix structure, where each row of the matrix represents a field pair, and each column records its entropy difference value, forming a complete field frequency difference matrix.

[0031] S302: Based on the entropy difference of field pairs in the field frequency difference matrix, determine whether the difference magnitude is less than the preset entropy difference proximity threshold, filter all field pairs that meet the conditions, mark the original field index, and generate a set of field pairs with close entropy differences. First, iterate through each row of the matrix, reading the field pair information and its entropy difference value. Compare this value with a pre-set entropy difference proximity threshold. The threshold can be determined by testing different thresholds on the validation set (e.g., incrementing by 0.01 from 0.01 to 0.2) and selecting the threshold that maximizes the accuracy of the final model, keeping it below 0.1. For example, in this step, the proximity threshold is set to 0.08, meaning that two fields with a distribution difference within 0.08 are considered structurally similar. Compare the entropy difference value of each record with the threshold to determine if it meets the condition of "entropy difference less than or equal to the threshold." If it does, the field pair is added to the result set; otherwise, it is not added. For example, if the entropy difference of field pair T7-F14 is 0.065, which is less than 0.08, then this field pair is a valid close field pair. Its corresponding field index information in the original normalized structure is then marked. For instance, if the index of T7 is 11 and the index of F14 is 23, then the record is {T7-F14, entropy difference: 0.065, index: 11-23}. After performing this judgment process on all field pairs, all field pairs that meet the conditions are filtered out and marked with field number, entropy difference value, and original index. This is then summarized to form a set of field pairs with close entropy differences. This set can be used as the input basis for subsequent field fusion operations, and only field pair combinations with highly correlated distribution characteristics are retained.

[0032] S303: Call the field pairs in the entropy difference close field pair set, perform linear mapping fusion operation according to the field index order, and serialize and arrange the fusion result to generate a field compressed expression sequence; Extract the field number and its original index position in the normalized field structure for each field pair, and arrange them in ascending order of the original index to ensure the consistency of the fusion operation results. When performing linear mapping fusion on each field pair, first read the original values ​​of the pathological texture field and the behavior field in that group. For example, if the texture field is color "red" and texture density "medium", and the behavior field is "click", convert these values ​​into numerical vectors through standardized mapping. The mapping values ​​can be set based on the frequency statistics of historical data samples: if the color "red" appears 30% of the training data, the corresponding mapping value is 0.6; if the density "medium" appears 20% of the data, the corresponding mapping value is 0.4; and if the click behavior frequency is 40%, the corresponding mapping value is 0.8. To achieve field fusion, a clear fusion rule needs to be set, and a linear weighting operation is performed on the two types of field values. The weights in this weighting operation can be set according to the stability of the field (such as the degree of fluctuation in the samples) or by the experience of domain experts. For example, if the fields have similar volatility, they are assigned an equal weight of 0.5. During the fusion process, each item is assigned an equal weight or weighted according to the field stability parameter. Then, the fusion results are serialized and sorted according to the original index order of each field in the normalized structure, that is, the fusion items with smaller index values ​​are placed first, and the items with larger index values ​​are placed later, so as to keep their relative positions in the overall structure unchanged. The fusion value is calculated as 0.5 × texture value vector + 0.5 × behavior value vector. The fusion result is a three-dimensional field combination vector such as [0.6, 0.4, 0.8]. The fusion process is performed on all field pairs in this way to obtain a complete list of fusion results. Then, according to the original index order of the field pairs in the normalized structure, the fused field values ​​are serialized and arranged, that is, the field fusion items with smaller indices are placed first, and the items with larger indices are placed later, so as to keep their positions in the overall structure unchanged. In this way, a field compression expression sequence is constructed, and finally an expression structure containing fusion values ​​and with a stable order is formed, which can be used for subsequent field dimensionality reduction or structural feature compression scenarios.

[0033] Please see Figure 5 The specific steps of S4 are as follows: S401: Call the field items in the field compression expression sequence, construct the continuous flow path of the fields in the structure according to the index sorting relationship in the sequence, and generate the field flow trajectory sequence based on the direction and magnitude of index change; First, based on the index numbers recorded in the sequence, the order in which each field item appears is read sequentially. The field items are then sorted in ascending order of index number, and during this sorting process, the absolute position number of each field item in the sequence and the distance change between adjacent field items are recorded. When comparing field items pairwise, the index p of the previous field item and the index q of the next field item are read, and the operation q minus p is performed to obtain the index change magnitude. The direction indicator is determined based on the sign of the change result: if q is greater than p, it is determined to flow forward; if q is less than p, it is determined to flow backward. Simultaneously, the field numbers corresponding to the starting and ending points of the flow are recorded. All field items are connected into a chain structure according to the reading order. Each chain segment records the "starting field index, ending field index..." The sequence consists of "index, direction identifier, and amplitude value". For example, if the index of field F3 is 4 and the index of the next field F7 is 10, then the record segment is "4→10, direction is forward, amplitude is 6". If the next segment is 10→6, then the record is "direction is backward, amplitude is 4". After each segment is generated, it is immediately structurally spliced ​​with the previous segment so that all segments are connected in sequence to form a complete flow path. Then, all path segments are integrated as a whole to form a continuous flow line according to the order of the fields in the sequence. The direction changes, amplitude values ​​and field numbers of all records are combined in sequence into a single sequence to obtain the field flow trajectory sequence. This sequence is composed of multiple structural segments arranged in index order, which can completely express the movement trajectory information of the fields in the compressed sequence.

[0034] S402: Based on the field flow trajectory sequence, extract the index difference between field positions, extract the interval field index that is greater than the preset jump threshold, filter and judge according to the interval size, and mark the original sequence number of the jump field to generate a jump field index set; The system reads the start and end indices of each segment of the flow trajectory segment by segment. The index difference is obtained by subtracting the start index from the end index, and this difference is recorded as the value of the interval field index change. Based on this, a judgment is made on whether the value is "greater than the jump segment threshold". This jump segment threshold needs to be set in conjunction with the total number of fields, field uniformity, and the field span characteristics of the target scene. For example, with 120 fields, a jump segment threshold of 10 is set, meaning that an index interval exceeding 10 is considered a jump segment. For each trajectory segment, the interval value is compared with the threshold. If the interval value is greater than 10, it is marked as a jump segment, and the start field index is recorded as the jump segment. The original sequence number of the field is used. For example, if the index of a trajectory segment jumps from 15 to 32, the interval is 17, which is greater than the threshold of 10, so the jump segment field is recorded as field index 15. If a segment is from 20 to 26, the interval is 6, which is less than the threshold, so it is not recorded. When performing filtering, the same judgment logic is applied to each segment. A temporary set of jump segment fields is built for all field items that meet the conditions, and the original position index of the field is retained in the record so that the order of jump segment fields can be restored later. This operation is performed cyclically for each trajectory segment until all trajectories are processed, and finally a jump segment field index set is formed. Each item records the original sequence number, interval value and corresponding trajectory segment information of the jump segment field, thus completing the jump segment field identification process.

[0035] S403: Call the field indexes in the jump segment field index set, rearrange them according to the order of appearance in the compressed sequence, and build a continuous structure for all jump segment field indexes to generate a jump segment index arrangement sequence; First, the skip segment field index set needs to be sorted sequentially according to the position of the fields in the compressed expression sequence. The sorting action is based on the actual numbering order of the fields in the compressed sequence. For example, if the field indices in the skip segment set are 15, 4, 23, and 8, their position numbers in the compressed expression sequence need to be queried in sequence. If their positions are 2, 0, 3, and 1 respectively, then the field indices are rearranged in the order of 0, 1, 2, and 3 to 4, 8, 15, and 23. After sorting, these skip segment fields are linearly connected in sequence. That is, each field index after the sorting is regarded as a node, and the nodes are connected end to end in sequence to form a continuous structural chain. The structural chain is based on the field number, such as "4→8→15→23". The new sorting number of each field index in the skip segment structure is recorded. Finally, all skip segment fields are organized into a continuous sequence according to the updated order. All field indexes are connected by the sorted structure as the main line. This sequence expresses the overall order and continuity of the skip segment fields in the compressed structure, and finally forms the skip segment index arrangement sequence.

[0036] Please see Figure 6 The specific steps of S5 are as follows: S501: Call all the jump segment numbers in the jump segment index arrangement sequence, identify the repeated field arrangement structure based on the order of the numbers in the structure, compare the structural form of the corresponding field values ​​of the same number in the sequence, extract the structural combination with consistent features, and generate a repeated structure comparison set. First, the order of all field indices is read, and a mapping table of the order's position information in the structure is constructed. Each hop segment field number records its segment number, its position in the fragment structure, and its original index position. During a full traversal of the sequence, the occurrence of multiple field numbers is extracted. Field numbers appearing two or more times are marked as "candidate duplicate numbers." For each candidate number, its corresponding field value at different positions is located. The field value includes its texture feature combination, behavior identifier content, and the connection state of the preceding and following structures. Subsequently, a structural comparison is performed on the field values ​​at each occurrence position of the number to determine if it is a homogeneous structure. Specifically, the comparison method involves reading the color information, edge features, and behavior action type from the field, and performing a comparison for each dimension. The degree is compared sequentially to see if they are equal. For example, if number 27 appears in the 4th and 11th positions of the sequence, the first time it is the color "blue", the edge "smooth", and the behavior "sliding". If the second time it is the field value, then the field is considered a repeating item in the structural form. For all field combinations that meet this feature, its number, the position of the structure, and the content of each dimension of the field value are recorded and stored in the comparison set. Structural consistency requires that the field value has some overlap or attribute substitution in each attribute dimension. If any attribute dimension is inconsistent, the field number will not be included in the comparison set. Finally, all field combinations that pass the judgment are combined into a comparison table according to their field number and structural arrangement. The structure is "field number → field value → position 1 → position 2". The output after integration is the repeating structure comparison set.

[0037] S502: Based on the field arrangement direction in the duplicate structure comparison set, perform field-level aggregation on the structure groups with the same direction, call the start and end positions of each group of structures in the sequence to complete the aggregation process, and construct the fragment structure to generate a merged fingerprint fragment set. The process reads the directional relationship of the fields corresponding to each structural combination. By comparing the relative index positions of the field numbers in the structure's first and second occurrences, if the second position is higher than the first, it is marked as "forward"; otherwise, it is marked as "reverse". All field combinations marked with the same direction are aggregated and grouped, with forward structural combinations in one group and reverse structural combinations in another. Within each group, the start and end indexes of its fields in the original sequence are read as the aggregation boundary for that group. Then, all field items within the aggregation boundary are called, and field-level aggregation operations are performed on their structural value content. The aggregation method is to read the attribute value of each field and merge identical attribute values. If multiple fields have completely identical attribute values, the aggregation result is a single item; if a certain field has an identical attribute value, the aggregation result is a single item. If a property has multiple values, a multi-value set structure is formed according to the field dimension. For example, if the field color attribute has "red", "blue", and "red", the final aggregated value is "red" and "blue". If the field edge attribute has "sharp", "sharp", and "smooth", the aggregated value is "sharp" and "smooth". After the aggregation operation is completed, a fragment structure is generated for the entire aggregation region. A new fragment item is formed for each combination direction group. The structure records information such as the starting index, ending index, field number set, aggregated field value set, and direction marker. After all the structure items are summarized, they are output as a merged fingerprint fragment set. Each item represents the field aggregation result formed in the same direction. The merged fingerprint fragment set provides continuous input data for subsequent structure reconstruction and logical mapping.

[0038] S503: Call the fingerprint fragments in the merged fingerprint fragment set, perform structural continuation and sequence reconstruction according to the arrangement relationship in the original fragment index sequence, integrate all fragments and map them to the decision process, and generate an AI decision system data processing scheme. The system sequentially reads the start index position, end index position, field number combination information, and aggregated field value set of each segment. Based on the segment's arrangement in the original segment index sequence, all segments are sorted in ascending order of their start index positions. After sorting, adjacent segments are connected sequentially according to the sorting results. A direct connection marker is established between the end position of the previous segment and the start position of the next segment. During the structural continuation process, overlapping field numbers are identified and processed. If the end and beginning field numbers of two adjacent segments are the same, the field items are merged into a single structure, and their field attribute values ​​are fused, retaining all appearing attribute content to form an aggregated field value list. Then, according to the sorting results, the content of each segment is connected one by one to complete the sequence reconstruction. The reconstructed sequence is arranged in order according to the field number. Each field retains its fusion attribute set to form a compressed expression with a clear logical structure and clear field relationships. Finally, after the structure is built, it is mapped to the data input module of the AI ​​decision-making process. According to the behavior / texture attribute represented by each field number, the field rule judgment conditions are set in the decision node. The field number, field aggregation attribute and its context relationship in the structure are converted into rule input items. The relationship table between field-rule-node is constructed. This relationship table serves as the input template for the decision engine. After integration, it becomes the data processing solution for the AI ​​decision-making system.

[0039] Please see Figure 7 AI decision-making system data processing system based on privacy computing, including: The field index normalization module is used to implement S1: extract texture fragments from the image end and structural fields from the behavior end, call the corresponding texture location points and behavior record items to segment and split, normalize the positions according to the order of field appearance, rearrange according to index continuity, and generate a normalized field index structure. The field union matching module is used to implement S2: Based on the normalized field index structure, it extracts the texture field value and behavior field value, compares them in order of their occurrence interval, reorganizes the field positions according to the comparison results, and generates the corresponding union set of the fields. The field entropy difference merging module is used to implement S3: based on the field corresponding union set, it calls the pathological texture field and the behavior index field, calculates the entropy difference of the field value distribution, filters similar field pairs according to the difference magnitude and linearly merges them to generate a field compressed expression sequence; The jump segment trajectory rearrangement module is used to implement S4: based on the field sorting relationship of the field compressed expression sequence, generate the flow trajectory of continuous field positions, extract the interval field and filter the jump segments according to the interval size, rearrange the jump segment positions in order, and form a jump segment index arrangement sequence; The fingerprint structure generation module is used to implement S5: call the jump segment number in the jump segment index arrangement sequence, perform structure comparison according to the repetition order of the number, merge the repetitive structures into fingerprint segments according to the appearance direction, construct the merged fingerprint set according to the segment arrangement, and generate the AI ​​decision system data processing scheme.

[0040] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A data processing method for an AI decision-making system based on privacy computing, characterized in that, Includes the following steps: S1: Extract texture fragments from the image and structural fields from the behavior, call the corresponding texture location points and behavior record items to segment and split them, organize the positions according to the order of field appearance, rearrange them according to the index continuity, and generate a normalized field index structure. S2: Based on the normalized field index structure, extract the texture field value and behavior field value, compare them sequentially according to the interval between their occurrences, reorganize the field positions according to the comparison results, and generate the corresponding union set of the fields; S3: Based on the joint set corresponding to the fields, call the pathological texture field and the behavior index field, calculate the entropy difference of the field value distribution, filter similar field pairs according to the difference magnitude and linearly merge them to generate a field compressed expression sequence; S4: Based on the field sorting relationship of the compressed expression sequence, generate the flow trajectory of continuous positions of the fields, extract the interval fields and perform jump segment filtering according to the interval size, rearrange the jump segment positions in order, and form a jump segment index arrangement sequence; S5: Call the segment number in the segment index arrangement sequence, perform structural comparison according to the repetition order of the number, merge the repetitive structures into fingerprint segments according to the appearance direction, construct the merged fingerprint set according to the segment arrangement, and generate the AI ​​decision system data processing scheme.

2. The data processing method for an AI decision-making system based on privacy computing according to claim 1, characterized in that, The normalized field index structure includes a texture location point index, behavior record item number, field order arrangement, and unified sequence label. The field corresponding union set includes a texture field mapping group, a behavior field mapping group, and a field parallel relationship identifier. The field compressed expression sequence includes difference close field pairs, merged field labels, and compressed field sorting information. The jump segment index arrangement sequence includes a field jump segment number, a jump segment interval identifier, and a jump segment order mark. The AI ​​decision system data processing scheme includes a repeating structure fragment number, a fragment merging index, and an AI decision reference unit.

3. The data processing method for an AI decision-making system based on privacy computing according to claim 1, characterized in that, The step of filtering similar field pairs based on the magnitude of difference and linearly merging them refers to filtering out the different field pairs based on the entropy difference of the field value distribution and then linearly merging them according to the value trend.

4. The data processing method for an AI decision-making system based on privacy computing according to claim 1, characterized in that, The index continuity refers to the fields being arranged in the original order of appearance in the structure without any gaps, verifying a continuous index number sequence.

5. The data processing method for an AI decision-making system based on privacy computing according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Based on the image-end texture fragments and behavior-end structure fields, extract the index values ​​of texture location points and behavior record items, call the two types of index values ​​to perform segmentation and splitting, and reposition the fields according to their order in the original sequence to generate a field position mapping table; S102: Based on the field order in the field position mapping table, combine the texture and behavior fields to form a field fragment sequence, extract the index value, and generate a field rearrangement index set; S103: Call the index values ​​in the field rearrangement index set, perform continuous rearrangement, combine the fields into a unified sequence structure, and generate a normalized field index structure.

6. The data processing method for an AI decision-making system based on privacy computing according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Call the field items in the normalized field index structure, retrieve the texture field value and behavior field value respectively, locate the position of the two types of fields in the sequence based on the field index, establish a correspondence table between field value and position index, and generate a field value index mapping table; S202: Based on the field position index in the field value index mapping table, calculate the interval between the texture field value and the behavior field value in the sequence, and match the interval value with the set field order comparison rules to establish a field combination group that conforms to the order relationship and generate a field order matching set; S203: Call the field order matching set, rearrange and organize the fields according to their parallel relationship in the original index structure, combine and aggregate the related fields into corresponding items, and generate a joint set of fields.

7. The data processing method for an AI decision-making system based on privacy computing according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Based on the union set corresponding to the fields, call the pathological texture field and the behavior index field, count the frequency distribution range of the field values, calculate the dispersion of the field based on the distribution range, and obtain the information on the distribution of the two types of fields to generate a field frequency difference matrix. S302: Based on the entropy difference value of the field pairs in the field frequency difference matrix, determine whether the difference magnitude is less than the preset entropy difference proximity threshold, filter all field pairs that meet the conditions, mark the original field index, and generate a set of field pairs with close entropy differences. S303: Call the field pairs in the entropy difference close field pair set, perform linear mapping fusion operation according to the field index order, and serialize and arrange the fusion result to generate a field compressed expression sequence.

8. The data processing method for an AI decision-making system based on privacy computing according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Call the field items in the field compression expression sequence, construct the continuous flow path of the fields in the structure according to the index sorting relationship in the sequence, and generate the field flow trajectory sequence based on the direction and magnitude of index change; S402: Based on the field flow trajectory sequence, extract the index difference between field positions, extract the interval field index that is greater than the preset jump threshold, filter and judge according to the interval size, and mark the original sequence number of the jump field to generate a jump field index set; S403: Call the field indexes in the jump field index set, rearrange them according to the order of appearance in the compressed sequence, and build a continuous structure for all jump field indexes to generate a jump index arrangement sequence.

9. The data processing method for an AI decision-making system based on privacy computing according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Call all the jump segment numbers in the jump segment index arrangement sequence, identify the repeated field arrangement structure based on the order of the numbers in the structure, compare the structural form of the corresponding field values ​​of the same number in the sequence, and extract the structural combination with consistent features to generate a repeated structure comparison set. S502: Based on the field arrangement direction of the repeated structure comparison set, perform field-level aggregation on the structure groups with the same direction, call the start and end positions of each group of structures in the sequence to complete the aggregation process, and construct the fragment structure to generate a merged fingerprint fragment set. S503: Call the fingerprint fragments in the merged fingerprint fragment set, perform structural continuation and sequence reconstruction according to the arrangement relationship in the original fragment index sequence, integrate all fragments and map them to the decision process to generate an AI decision system data processing scheme.

10. A data processing system for an AI decision-making system based on privacy computing, characterized in that, The system is used to implement the data processing method for an AI decision-making system based on privacy computing as described in any one of claims 1-9, and the system comprises: The field index normalization module is used to implement S1: extract texture fragments from the image end and structural fields from the behavior end, call the corresponding texture location points and behavior record items to segment and split, normalize the positions according to the order of field appearance, rearrange according to index continuity, and generate a normalized field index structure. The field union matching module is used to implement S2: Based on the normalized field index structure, extract the texture field value and behavior field value, compare them in order according to the interval between their occurrence, reorganize the field positions according to the comparison results, and generate the field corresponding union set; The field entropy difference merging module is used to implement S3: based on the joint set corresponding to the fields, the pathological texture field and the behavior index field are called to calculate the entropy difference of the field value distribution, and similar field pairs are selected and linearly merged according to the difference magnitude to generate a field compressed expression sequence; The jump segment trajectory rearrangement module is used to implement S4: based on the field sorting relationship of the field compressed expression sequence, generate the flow trajectory of continuous field positions, extract the interval field and perform jump segment filtering according to the interval size, rearrange the jump segment positions in order, and form a jump segment index arrangement sequence; The fingerprint structure generation module is used to implement S5: call the jump segment number in the jump segment index arrangement sequence, perform structure comparison according to the repeating order of the number, merge the repeating structures into fingerprint segments according to the appearance direction, construct the merged fingerprint set according to the segment arrangement, and generate the AI ​​decision system data processing scheme.

Images