A data mining intelligent processing analysis method

Abstract

The present application relates to the technical field of data processing, and particularly relates to a data mining intelligent processing and analysis method, comprising the following steps: identifying a responsive section in the behavior sequence which is rewritten by a treatment action, and outputting a corresponding treatment cut set; performing chain breaking separation on the behavior sequence to obtain a native evolution section which does not cross the treatment cut set; taking the native evolution section as input to generate a corresponding candidate mode chain, and performing back-filling exclusion on the candidate mode chain in combination with the treatment sequence to output a stable mode chain; and taking the stable mode chain as input to generate a trusted mining result of the object to be mined. The present application can effectively block back-filling pollution generated by system or artificial treatment behavior on subsequent data, so that the subsequent mining process no longer depends on the data track which has been intervened, and the pseudo-law shaped by treatment is avoided to be misidentified as a real mode from the source, thereby significantly improving the objectivity and authenticity of the data mining result.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a data mining intelligent processing and analysis method. Background Technology

[0002] With the widespread application of information systems, various business systems continuously generate a large amount of behavioral data during operation, such as user operation records, equipment operation logs, business process trajectories, and status change records. Mining and analyzing this data to extract behavioral patterns and evolutionary rules has become an important technical means to support risk identification, intelligent decision-making, and system optimization.

[0003] Existing data mining methods typically rely on behavioral sequences, identifying statistically significant behavioral paths from historical data through frequent pattern mining, sequence pattern analysis, or association rule extraction. These methods generally assume that behavioral data can objectively reflect the true evolutionary process of an object, thus directly modeling and analyzing the entire dataset. However, in real-world business systems, behavioral data is often not entirely naturally generated but is continuously influenced by automatic system interventions or manual processing.

[0004] Specifically, when a system detects anomalies, risks, or specific states, it typically triggers corresponding actions, such as restricting operations, adjusting states, supplementing information, or implementing corrective strategies. These actions not only change the current state but also have a lasting impact on subsequent behaviors, causing the sequence of subsequent behaviors to deviate from the original evolutionary trajectory. In this case, subsequent behaviors are actually "responsive behaviors," rather than the natural behaviors of the object under uninterrupted conditions. However, existing data mining methods often fail to distinguish between natural behaviors and responsive behaviors, incorporating them uniformly into the mining process. This easily leads to extracted patterns being mixed with the effects of responsive actions. Some behavioral patterns, although not directly falling within the timeframe of the responsive actions, are essentially driven by responsive motivations, such as compensatory or avoidance behaviors arising to satisfy responsive strategies. These patterns may statistically exhibit high frequency and stability, but their essence is not the object's inherent law but rather the result of reverse shaping by external responsive mechanisms. Existing technologies typically lack the ability to identify such responsive-induced patterns, leading to hidden distortions in the mining results. Summary of the Invention

[0005] This invention provides an intelligent data mining processing and analysis method that can effectively identify and eliminate interference caused by actions and their motivations during the data mining process, thereby extracting the true patterns reflecting the original behavioral evolution of the object to be mined.

[0006] A data mining intelligent processing and analysis method includes the following steps: S1. Taking the behavior sequence of the object to be mined and the corresponding disposal sequence as input, identify the responsive segments in the behavior sequence that are rewritten by the disposal action, and output the corresponding disposal cut-off set; S2. Using the set of disposal cutoffs as input, perform chain breaking separation on the behavior sequence to obtain the original evolution segment that does not cross the set of disposal cutoffs; S3. Using the original evolution segment as input, generate a corresponding candidate pattern chain, and combine the treatment sequence to perform backfeeding and elimination on the candidate pattern chain, and output a stable pattern chain. S4. Using the stable pattern chain as input, generate reliable mining results for the object to be mined.

[0007] Optionally, S1 further includes traversing each action in the action sequence, extracting the timestamp attribute and scope label of each action; the action sequence is expanded into continuous action tuples along the time axis, and the action tuples that have temporal intersection with the action are located based on the timestamp attribute and scope label, and the action tuples that have temporal intersection are marked as disturbed tuples.

[0008] Optionally, the consecutive disturbed tuples are merged into intervals to generate responsive segments; for each responsive segment, the start boundary time and the end boundary time of the responsive segment in time are taken to form a processing cut, and all the processing cuts constitute the processing cut set.

[0009] Optionally, S2 further includes obtaining the start boundary time and end boundary time corresponding to each treatment cut in the treatment cut set; expanding the behavior sequence into a behavior tuple sequence according to its own temporal order; and segmenting the behavior tuple sequence using the start boundary time and the end boundary time as cutting reference points.

[0010] Optionally, after the behavior tuple sequence is segmented and truncated, any segmented sequence fragment does not contain any complete internal interval of the treatment cut within its temporal range.

[0011] Optionally, the temporally continuous and semantically consistent sequence fragments after the segmentation are extracted as the original evolutionary segments, wherein the start and end times of any one of the original evolutionary segments do not fall between the start and end boundary times of the same processing cut.

[0012] Optionally, S3 further includes performing frequent pattern mining on each of the original evolutionary segments, extracting high-frequency and temporally ordered behavioral subsequences within the original evolutionary segments to form initial pattern fragments corresponding to each of the original evolutionary segments; and splicing together the initial pattern fragments that are temporally adjacent and semantically compatible to generate the candidate pattern chain.

[0013] Optionally, the effective time period of each treatment action in the treatment sequence is obtained, and each candidate pattern chain is projected onto the effective time period for matching and verification. If a candidate pattern chain overlaps with the effective time period of any treatment action within its temporal extension interval, the candidate pattern chain is marked as a reinjection pattern chain. The reinjection pattern chain is removed from all candidate pattern chains, and the candidate pattern chains that do not temporally overlap with the effective time period of the treatment sequence are retained and output as the stable pattern chain.

[0014] Optionally, S4 further includes obtaining the pattern support and pattern confidence of each stable pattern chain, filtering stable pattern chains that meet the conditions according to a preset reliability threshold to form a reliable pattern chain set; performing temporal merging and semantic compression on each stable pattern chain in the reliable pattern chain set to obtain a summary of the behavioral patterns of the object to be mined.

[0015] Optionally, the behavioral pattern summary is compared and verified with the treatment motivations recorded in the treatment sequence, and the remaining treatment-induced patterns are removed to obtain a cleaned behavioral pattern summary; the cleaned behavioral pattern summary is used as the reliable mining result of the object to be mined, wherein the reliable mining result represents the original behavioral pattern of the object to be mined before being disturbed by the treatment action.

[0016] The beneficial effects of this invention are: This invention differs from existing technologies that directly mine full behavioral data. First, it identifies the scope of action taken to construct responsive segments. Then, it extracts action entry points to break down behavioral sequences, retaining only the original evolutionary segments that do not cross the action's influence range. This mechanism effectively prevents the backflow of data contamination caused by systemic or manual actions, ensuring that subsequent mining processes no longer rely on manipulated data trajectories. It avoids misidentifying pseudo-patterns shaped by actions as genuine patterns, significantly improving the objectivity and authenticity of data mining results.

[0017] This invention addresses the less-discussed issue of how interventions shape behavioral patterns, a relatively unexplored technology in existing techniques. It constructs a complete closed-loop processing mechanism: frequent pattern mining and pattern chain concatenation are performed within the original evolutionary phase to form candidate pattern chains; then, temporal overlap verification with the action sequence's time period is performed to identify and eliminate reinjected pattern chains; further, semantic-level induced elimination of behavioral pattern summaries is conducted in conjunction with the action motivation. This multi-layered filtering mechanism not only eliminates the influence of interventions in the temporal dimension but also eliminates action-driving factors at the semantic causal level, thereby accurately suppressing hidden and distorted patterns that are difficult to identify using traditional methods and improving the reliability of the mining results.

[0018] This invention transforms complex behavioral sequences into compact and stable behavioral pattern summaries by performing support and confidence screening on stable pattern chains and further performing temporal merging and semantic compression. Simultaneously, through an induced verification mechanism, it ensures that the final output behavioral patterns originate solely from the object's natural evolutionary process, without relying on external intervention logic. The resulting reliable mining results not only possess statistical stability but also clear semantic expression and interpretability, facilitating direct application in scenarios such as risk identification, behavioral analysis, and strategy optimization, thereby reducing subsequent manual review and system adaptation costs. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the method execution logic in an embodiment of the present invention. Detailed Implementation

[0021] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. For some well-known technologies, those skilled in the art may also use other alternative methods to implement the invention. Moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0022] like Figure 1 As shown, a data mining intelligent processing and analysis method includes the following steps: S1. Taking the behavior sequence of the object to be mined and the corresponding disposal sequence as input, identify the responsive segments in the behavior sequence that are rewritten by the disposal action, and output the corresponding disposal cut-off set; Behavioral sequences refer to a series of continuous behavioral or state changes that occur over time in the natural business process of the object to be mined, reflecting the evolutionary trajectory of the object itself, excluding the results of human or system intervention.

[0023] The action sequence pointer is a record sequence formed by arranging various manual or system intervention actions triggered on the object to be explored in time. It is used to change, restrict or correct the state of the object, and the result will affect subsequent actions.

[0024] S11, traverse each action in the processing sequence and extract its timestamp attribute and scope label.

[0025] During iteration, the following is executed: a. Sort the disposal sequence in chronological order; b. Starting from the first action, read the current actions one by one; c. Extract the timestamp attribute and scope tag from the current action; d. Register the extracted results in a structured record, move to the next processing action, and repeat the above process until the traversal is complete.

[0026] Therefore, the disposal sequence is represented as: ; Then each action Represented as: ; in, For the first The timestamp attribute of each action indicates the point in time when the action occurred. This indicates the total number of actions in the action sequence. For the first The scope tag of a disposal action indicates the scope of objects or constraints to which the disposal action applies. The scope tag is obtained as follows: a. Analyze the target object identifier associated with the handling action; b. Match the target object to its associated business entity or object set based on the target object identifier; c. Read the rule definition or strategy configuration corresponding to the action to determine the scope of impact; d. Merge and map the scope of the object set with the scope of the rule's influence; e. Output a uniform scope label to characterize the scope of the action.

[0027] S12 expands the action sequence into continuous action tuples along the time axis.

[0028] Let the sequence of actions be represented as: ; Then each action tuple Represented as: ; in, Indicates the first The timestamp of each action tuple. This represents the total number of tuples representing the number of rows in the sequence of rows. Indicates the first Each behavior tuple corresponds to a behavior label or status identifier. Behavior labels include access, submission, trigger, and change, while status identifiers include normal, abnormal, pending, and processed. By reading the original record content corresponding to the behavior tuple, the behavior type field or status field is parsed from the original record. If the original fields are not consistent, they are standardized according to the preset mapping rules, and the unified result after the transformation is registered as the behavior label or status identifier of the behavior tuple.

[0029] Based on this, it is determined whether there is a temporal intersection between the behavior tuple and the action, while simultaneously satisfying the scope constraint, that is, whether a certain behavior has been affected by an action. The system has two timelines: a behavior sequence and a action sequence. For each behavior tuple in the behavior sequence, a judgment must be made: whether an action is affecting it when the current behavior occurs. This judgment is done in two layers: The first level is whether the time overlaps: if the time of a certain behavior falls within the time range of a certain action, it means that the time may be affected. The second layer is whether the object is within the scope of influence: that is, whether the object or label corresponding to the behavior is within the scope of this action. Only when both of these conditions are met is the behavior considered to have been intervened, i.e., a disturbed tuple; otherwise, the behavior is considered to have occurred naturally and is undisturbed.

[0030] Specifically, define behavioral tuples The function to determine whether a tuple is perturbed: ; This formula means that if there is at least one action that causes the behavior to occur within the time range of the action, and the label of the behavior is within the range of the action, then the behavior is determined to be disturbed, and the result is 1; otherwise, the behavior is determined to be undisturbed, and the result is 0.

[0031] in, This indicates a disturbance flag for behavioral tuples; a value of 1 indicates a disturbed tuple, and a value of 0 indicates an undisturbed tuple. Not the first The effective time range of each action. This indicates that the label of the behavior tuple falls within the scope label range of the action. This indicates that at least one action meets the condition. Indicates the first The duration of the impact of each intervention action is used to characterize the sustained range of its influence on subsequent behavior. Based on adaptive determination of behavioral recovery characteristics, the duration is determined by observing the changing trends of the subsequent behavioral sequence, starting from the moment the intervention action occurs. The impact ends when the behavioral pattern returns to a stable state consistent with the pre-intervention state, and this time interval is used as the threshold. Because the impact of the action will essentially change the distribution of behavior, once the behavior returns to stability, it means that the impact has subsided.

[0032] when When this happens, the tuple containing the behavior is marked as a disturbed tuple.

[0033] S13 merges consecutive disturbed tuples into intervals to generate responsive segments.

[0034] The set of disturbed tuples is: ; According to time sequence After sorting, disturbed elements that satisfy the condition that the adjacent time interval is less than a preset continuous threshold are grouped into the same segment.

[0035] Define the criteria for determining whether adjacent disturbed tuples are consecutive: ;in, The continuous judgment threshold is used to limit the maximum allowable time interval between adjacent disturbed tuples. Under normal sampling density, there will be no obvious time break between consecutive actions. Exceeding this range usually indicates that the action chain has been interrupted. Therefore, the average sampling interval of the action sequence is obtained. Set the value to be within the range of 1.5 to 2 times the sampling interval. This represents the timestamps of two adjacent disturbed tuples.

[0036] Based on the above determination, consecutive disturbed elements are combined to generate a set of responsive segments: ; Indicates the total number of responsive segments.

[0037] Each responsive segment By the The start time of the first responsive segment and the first The termination time of each responsive segment is constituted.

[0038] S1, for each responsive segment, extract its start and end boundaries to form a processing entry point; specifically, this includes: traversing all responsive segments in chronological order; for the current responsive segment, reading its start time. This serves as the initial boundary for entering the disturbance; the end time of this responsive segment is read. As the termination boundary for exiting the disturbance; the starting boundary and the termination boundary are combined into a set of boundary structures to generate corresponding disposal cuts; a single disposal cut is defined as: Unlike responsive segments, which describe a continuous interval of a phenomenon, disposal cuts are structured extractions and controlled expressions of the boundaries of that interval. Specifically, a responsive segment is a time interval formed by combining continuously disturbed behavioral elements to describe where a sustained impact occurs; while a disposal cut is the starting and ending boundaries extracted from that interval, transforming the original interval information into a boundary control unit that can be used for subsequent chain-breaking operations to guide the subsequent segmentation of the behavioral sequence. Therefore, responsive segments focus on expressing the range of influence itself, while disposal cuts focus on providing the operational basis for cutting off and isolating that range.

[0039] The treatment incision set consists of all treatment incisions: ; in, Indicates the first Each treatment incision is used to identify a time interval affected by the treatment action. This represents the set of cut points, providing boundary constraints for subsequent chain breakage and separation steps.

[0040] S2. Using the set of disposal cutoffs as input, perform chain breaking separation on the behavior sequence to obtain the original evolution segment that does not cross the set of disposal cutoffs.

[0041] S21. Obtain the starting boundary time corresponding to each treatment cut in the treatment cut set. and termination boundary time Since the behavioral sequence needs to be segmented only at the moments of entering and exiting the perturbation, and each processing segment has already fully recorded two key time points: Initial boundary moment: The behavior begins to be affected by the action; Termination boundary moment: The action ends and is affected by the action.

[0042] When all these two types of boundary moments of all the processing cuts are summarized, they cover all time positions that need to be broken or segmented. Therefore, the set of these start boundary moments and end boundary moments constitutes the complete set of time boundaries required to segment the behavior sequence, that is, the complete set of time boundaries used for behavior sequence segmentation is obtained by extraction.

[0043] S22. Segmented truncation based on the treatment incision: Obtain the sequence of behavioral tuples in S1 according to their own temporal order, and segment and truncate the sequence of behavioral tuples using the start boundary time and the end boundary time as cutting reference points.

[0044] In S1, the action sequence is represented as: ;and: Therefore, the set of all treatment incision boundaries is uniformly represented as: ;in, Represents the ordered set of all boundary moments for handling the incision boundary; Specifically, segmented truncation involves traversing each processing cut in the processing cut set; reading the start and end boundary times of the current processing cut; adding the start boundary time to the boundary time set; adding the end boundary time to the boundary time set; deduplicating all the added boundary times; and sorting the boundary time set in chronological order to obtain an ordered boundary time set. Based on the ordered set of boundary moments, the sequence of behavioral tuples is segmented to obtain several sequence fragments: ; Each sequence segment satisfy: ; in, , For the first The start and end time boundaries of each sequence segment are determined by the boundary times of adjacent treatment cuts. This represents the number of sequence segments obtained after segmentation.

[0045] At the same time, each sequence segment is constrained to satisfy: ;in, The expression indicates that no such expression exists. The above expression means that any sequence segment does not contain any complete internal region of any treatment cut.

[0046] Specifically, this involves: a sequence of behavioral tuples unfolded chronologically based on the behavioral sequence; using an ordered set of boundary moments as a cutting reference, dividing the behavioral tuple sequence into multiple continuous time intervals at each boundary moment; for each time interval, extracting behavioral tuples whose timestamps fall within that interval to form corresponding sequence segments; when the timestamp of a behavioral tuple equals a boundary moment, treating that behavioral tuple as an interval boundary point, ensuring it belongs to only a single sequence segment; repeating the above extraction process for all time intervals to obtain multiple sequence segments divided chronologically; checking each sequence segment to ensure its time range does not cross the complete interval between the start and end boundaries of any treatment cut; if a crossover occurs, performing further subdivision and truncation at the corresponding boundary; and outputting all truncated sequence segments as input for subsequent original evolution segment selection.

[0047] S23, Extract the original evolutionary segment: In the segmented and truncated sequence fragments, those that are temporally continuous and semantically consistent are extracted as the original evolutionary segments. A temporally continuous and semantically consistent sequence fragment can serve as an original evolutionary segment because, although a sequence fragment has been segmented by the processing cut, not every remaining segment can be directly used as an original evolutionary segment. Further evaluation is needed to determine whether the fragment still maintains the continuity and consistency expected of the object's natural evolution. Temporally continuous means that the time interval between adjacent behavioral tuples within the sequence fragment cannot be too large, and there should be no obvious time breaks. If the interval between adjacent behaviors is too long, it often indicates that this segment is not the same natural evolutionary process, but rather a process that has been interrupted and restarted, and therefore is not suitable as a complete original evolutionary segment. Semantically consistent means that the various behavioral tuples within the sequence fragment should be coordinated in terms of business meaning, and there should be no obvious conflicts, cross-state jumps, or abrupt changes in semantic direction. For example, if the preceding action is still expressing a normal progress, but the following action suddenly jumps to a state of reversal or cancellation, or restarting after completion—a state that does not conform to the natural evolutionary pattern of the same segment—it indicates that although this segment is continuous in time, it is semantically inconsistent and cannot be considered a native evolutionary segment. Only when both of these conditions are met can it be said that this segment of data has neither been covered by the processing cut-off nor has it experienced any abnormal interruption or semantic break within it. Therefore, it can represent a natural evolutionary process of the object to be mined under conditions of no processing backfeeding interference, and only such a sequence segment can be extracted as a native evolutionary segment.

[0048] Define candidate fragments The temporal continuity determination is as follows: ;in, The timestamps of adjacent row tuples. The sequence continuity threshold is used to determine whether the time intervals within a segment are continuous. It involves statistically analyzing the common time intervals between adjacent behavioral tuples of similar objects under natural business conditions. These common time intervals serve as a basic reference, and redundancy is then added to this base as the sequence continuity threshold. In real-world applications where behaviors occur frequently, the essence of the sequence continuity threshold is not simply limiting the duration, but rather defining whether two adjacent behaviors still belong to the same natural evolutionary chain. If the threshold is too small, it will excessively fragment sequence segments that belong to the same natural evolutionary process; if the threshold is too large, it will incorrectly reassemble two independent processes that have already been interrupted. Therefore, it is based on the concentrated distribution range of adjacent behavior intervals in historical normal samples, and then adjusted according to the specific business rhythm.

[0049] Define the behavior semantic autonomy determination function: This indicates that the sequence segment has not experienced any cross-state conflict or abnormal transition in terms of behavioral semantics.

[0050] The behavioral semantic consistency determination function is executed as follows: a. Read all the line tuples in the current sequence segment and arrange them in chronological order; b. Extract the behavior label or status identifier corresponding to each behavior tuple one by one; c. Check the semantic connection between adjacent behavior tuples to determine whether the latter behavior can be naturally deduced from the former behavior; if there is a forward progression, similarity maintenance or reasonable fluctuation between two adjacent behaviors that are allowed by the business process, then the semantic connection is considered to be valid. d. Perform a full check on the entire sequence segment to determine if there are any abnormalities such as conflicting states, stage reversals, sudden semantic changes, or restarts without basis after the segment has ended. If no such abnormalities exist, the sequence segment is considered to have consistent behavior and semantics. If such abnormalities exist, the sequence segment is considered to have inconsistent behavior and semantics. Finally, the determination results are output to filter sequence segments that can be used as original evolutionary segments.

[0051] A sequence segment is identified as a primary evolutionary segment when all of the following conditions are met: ; in, This indicates the identifier for the original evolutionary segment.

[0052] The final set of original evolutionary segments is obtained: ;in, This is a set of original evolutionary segments.

[0053] Further constraints require that any primary evolutionary segment satisfy: ; This formula indicates that the time range of any sequence segment identified as a primary evolutionary segment cannot completely fall within the time interval corresponding to any treatment cut. Wherein, , These represent the start and end times of the original evolutionary segment. This indicates the time interval of the impact of the treatment, corresponding to the incision made. This means for all.

[0054] S3. Using the original evolutionary segment as input, generate a corresponding candidate pattern chain, and perform a backfeeding and elimination process on the candidate pattern chain in conjunction with the treatment sequence to output a stable pattern chain.

[0055] S31, Initial Pattern Fragment Extraction: In S2, the data affected by the treatment was removed, and only the data fragments of natural evolution, i.e. the original evolution segment, were retained. However, the original evolution segment is still just the original behavior flow and no pattern has been extracted yet. Therefore, the purpose of S31 is to find the behavior patterns that frequently recur.

[0056] Frequent pattern mining is performed on each of the original evolutionary segments to extract high-frequency and temporally ordered behavioral subsequences within the segment, forming initial pattern fragments. The set of original evolutionary segments can be represented as: ;in, Indicates the first One original evolutionary segment, This indicates the total number of original evolutionary segments.

[0057] For each original evolutionary segment , is represented as a sequence of row tuples: ;in, Represents a behavior tuple. For timestamps, Indicates a behavior label or status identifier. This indicates the number of behavioral tuples in the original evolutionary segment.

[0058] Extracting frequent behavioral subsequences from the original evolutionary segment: ; satisfy: ; and: ; in, Indicates the first The first of the original evolutionary stages An initial pattern fragment, Indicates the length of the subsequence. This indicates the frequency of occurrence of the subsequence within the original evolutionary segment. The threshold for frequent determination is 10% to 15% of the length of the original evolutionary segment.

[0059] The final set of initial pattern fragments is obtained: .

[0060] The execution of frequent pattern mining is as follows: 1. Read all behavior tuples from a native evolutionary segment and arrange them in chronological order; 2. Extract the behavior label or status identifier corresponding to each behavior tuple to form an ordered behavior sequence; 3. In the ordered sequence of behaviors, generate multiple candidate subsequences of behaviors according to a preset length range; 4. Count the number of times each candidate behavioral subsequence appears in the original evolutionary segment; 5. Filter out subsequences of behaviors that appear multiple times, reaching a preset frequent judgment threshold; 6. Retain the subsequences of behaviors that satisfy temporal order consistency as the initial pattern fragments.

[0061] S32, Candidate pattern chain generation: Initial pattern fragments that are temporally adjacent and semantically compatible are concatenated to generate a candidate pattern chain. Let any two initial pattern fragments... , Their time intervals are as follows: ; in, Representation pattern fragment The start and end times, Representation pattern fragment The start and end times.

[0062] Condition 1: Define the timing adjacency condition: ;in, Indicates the time threshold for splicing pattern fragments. The interval should not exceed the maximum reasonable interval for continuous occurrence of normal behavior. Therefore, by statistically analyzing the time interval distribution of adjacent behavioral tuples in each primary evolutionary segment, the main concentrated distribution interval of the time interval distribution is extracted, and the upper bound of the concentrated interval is taken as... The initial value of .

[0063] Condition 2: Define a semantic compatibility determination function: The semicolon indicates that the two pattern fragments are semantically compatible. The semantic compatibility determination function is used to determine whether the two pattern fragments are semantically compatible. The specific execution steps are as follows: 1. Read pattern fragments The last action label or status identifier; 2. Read pattern fragments The initial behavior label or status identifier; 3. Determine whether the final action and the starting action satisfy a preset semantic succession relationship rule. The semantic succession relationship includes forward progression relationship, state preservation relationship, and reasonable transfer relationship. 4. If there is no conflict between the ending behavior and the starting behavior, it is determined to be semantically compatible. Conflict relationships include state reversal, logical rollback, and restarting without basis after termination. 5. Output the judgment result. If the semantics are compatible, set it to 1; otherwise, set it to 0.

[0064] If both conditions 1 and 2 are met, then the splicing operation is performed: ;in, This represents the candidate pattern chain obtained after concatenation. This indicates a sequence concatenation operation.

[0065] By iteratively concatenating the candidate pattern chains, a set of candidate pattern chains is obtained: ;in, Indicates the first A candidate pattern chain, This indicates the number of candidate pattern chains. Iterative splicing involves using the spliced ​​result in the next round of splicing, repeating this process until no further splicing is possible. Specifically, first, two splicable pattern fragments are combined into a longer fragment; then, the new fragment is used to further determine with other pattern fragments whether they can be spliced; if they can still be spliced, the splicing continues; this process is repeated until no new splicable combinations remain.

[0066] S33, Reinforcement Pattern Chain Identification: The purpose is to determine whether a candidate pattern chain was formed during the intervention period, identify it, and exclude it. Specifically, first, obtain the effective time periods of all intervention actions; then observe the time range of each candidate pattern chain itself, that is, the time from the start to the end of the pattern chain; project the time range of the pattern chain onto the intervention period for comparison. If it is found that the time range of the pattern chain overlaps with the effective time period of a certain intervention action, it indicates that part of the formation process of the pattern chain occurred within the time of the intervention. This pattern is not a naturally occurring law of the object, but a result shaped by the intervention behavior, and needs to be marked.

[0067] Therefore, the effective time period of each action in the action sequence is obtained, and the candidate pattern chain is projected to the effective time period for matching and verification; the action sequence in S1 is represented as follows: Each action is represented as ; It is the start time of the handling action. It refers to the duration of the impact of the response action.

[0068] Set up a candidate pattern chain The time interval is: ;in, The start time of the candidate pattern chain. This is the termination time of the candidate pattern chain.

[0069] Define timing overlap criteria: ; in, This serves as an indicator for determining the overlap between candidate pattern chains and handling actions. This represents the empty set.

[0070] If the following conditions are met: Then mark the candidate pattern chain as a recharge pattern chain.

[0071] S34, Remove the backfeeding pattern chain from all candidate pattern chains, retain the candidate pattern chains that have not experienced temporal overlap, and define the stable pattern chain set: ;in, This represents a set of stable mode chains.

[0072] S4. Using the stable pattern chain as input, generate reliable mining results for the object to be mined.

[0073] S41, Let the set of stable mode chains in S34 be represented as: ;in, This represents the u-th stable mode chain. This represents the total number of stable mode chains. Retrieves the set of stable mode chains. The pattern support and pattern confidence of each stable pattern chain are used to select stable pattern chains that meet the conditions based on a preset reliability threshold, thus forming a reliable pattern chain set.

[0074] Define pattern support: ;in, Representation of pattern chain Number of times it appears in all the original evolutionary segments This represents the total number of original evolutionary segments.

[0075] Define pattern confidence: ;in, Representation of pattern chain The prefix subsequence is the subsequence obtained by sequentially extracting the preceding part from the beginning of the pattern chain. This indicates the number of times the prefix subsequence appears.

[0076] Define reliability determination function ; in, This represents the support threshold; a pattern is considered valid only if it appears in at least a certain percentage of the original evolutionary segments. Therefore, it is taken as 10% to 15% of the total number of original evolutionary segments. This represents the confidence threshold. For high stability requirements, a value of 0.8–0.9 is used; for high coverage requirements, a value of 0.6–0.7 is used. This is a reliability assessment identifier.

[0077] The reliable pattern chain set is obtained through filtering: ;in, It is a set of reliable pattern chains.

[0078] S42, Generate a summary of behavioral patterns: Perform temporal merging and semantic compression on each stable pattern chain in the reliable pattern chain set.

[0079] Let any stable mode chain be represented as: ;in, This represents the length of the pattern chain.

[0080] Define the timing merge operation: This formula represents: removing consecutively repeating behavior labels from the pattern chain, retaining only the unique sequences; where, This represents the pattern chain after removing consecutive repetitive behaviors.

[0081] Define a semantic compression function: ;in, This represents a compressed tag obtained by merging multiple semantically similar behavioral tags. The semantic compression function executes as follows: 1. Read the stable pattern chain to be processed and obtain its behavior label sequence arranged in chronological order; 2. Traverse each behavior label in the behavior label sequence and extract its corresponding semantic feature information or semantic category identifier; 3. Based on preset semantic mapping rules or semantic similarity relationships, the behavior tags are grouped, and behavior tags with similar semantics or belonging to the same business category are grouped into the same semantic group; 4. For each semantic group, determine a unified representative label to characterize the overall semantics of that group; 5. Following the original chronological order, multiple behavior tags belonging to the same semantic group and appearing adjacently are merged into the representative tag to generate a compressed behavior tag sequence; 6. Perform a complete traversal of the compressed behavior label sequence to ensure that there are no semantically duplicated or conflicting adjacent labels; output the compressed behavior label sequence as the semantically compressed pattern chain.

[0082] Each stable pattern chain, after temporal merging and semantic compression, is simplified into an abstract expression. This simplified result is defined as a behavioral pattern summary. Then, the same process is applied to all stable pattern chains, resulting in multiple behavioral pattern summaries. All summaries are then aggregated to form a behavioral pattern summary set. The behavioral pattern summary set is represented as: ;in, Indicates the first Summary of behavioral patterns This refers to the number of abstracts. Simply put, it's as follows: Input: A set of stable mode chains S; Processing: For each s∈S, perform temporal merging to remove redundant structures and semantic compression to merge similar semantics; Output: Each s becomes an h, and the set of all h equals the set of behavioral pattern summaries H.

[0083] S43, Handling Induced Pattern Removal: The behavioral pattern summary is compared and verified with the treatment motivations recorded in the treatment sequence to eliminate residual treatment-inducing patterns.

[0084] Let the set of motivations for action be represented as: ;in, Indicates the first One of the motivations for handling the matter. The number of driving factors to be dealt with.

[0085] Define the induced criterion function: ; in, This indicates that the summary of behavioral patterns can trigger or explain the corresponding motivations for action. If the following conditions are met: If the behavior pattern summary is determined to be a disposition-inducing pattern, it will be removed. This formula indicates that in the set of disposition motivations, there exists at least one disposition motivation such that there is a correspondence or triggering relationship between the behavior pattern summary and the disposition motivation.

[0086] The final set of behavioral patterns after cleaning is obtained: ;in, This represents the set of behavioral pattern summaries after cleaning. The formula states: for each behavioral pattern summary... Examine each one in relation to all the motivations for the action. The relationship, if for all All satisfied If there is no triggering or interpreting relationship, then the [retained value] is retained. The final set It is a collection of pure, original behavioral patterns that are not affected by any motives or actions.

[0087] Furthermore, a behavioral pattern summary that can trigger or explain the corresponding action motivation means that the behavioral pattern described in a behavioral pattern summary has a direct correlation with the cause of a certain action in business logic. That is, when the behavioral pattern occurs, it will trigger the corresponding action, or the behavioral pattern itself is a process manifestation of the cause of the action. In other words, if a behavioral pattern summary can semantically explain why the action occurs, or temporally show that the action is triggered immediately after the behavior occurs, then it is believed that there is a causal or explanatory relationship between the behavioral pattern summary and the action motivation. Such patterns are often the result of being induced or reinforced by the action mechanism, rather than the natural evolutionary law of the object itself.

[0088] S44. The cleaned behavioral pattern summary has simultaneously eliminated two types of interference: interference from handling effects and rules induced by handling. The final retained behavioral pattern summary is neither affected by handling time nor driven by handling logic. Therefore, it is believed that these rules only originate from the natural behavioral evolution of the object to be mined. Thus, the cleaned behavioral pattern summary is used as the reliable mining result of the object to be mined.

[0089] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0090] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A data mining intelligent processing and analysis method, characterized in that, Includes the following steps: S1. Taking the behavior sequence of the object to be mined and the corresponding disposal sequence as input, identify the responsive segments in the behavior sequence that are rewritten by the disposal action, and output the corresponding disposal cut-off set; S2. Using the set of disposal cutoffs as input, perform chain breaking separation on the behavior sequence to obtain the original evolution segment that does not cross the set of disposal cutoffs; S3. Using the original evolution segment as input, generate a corresponding candidate pattern chain, and combine the treatment sequence to perform backfeeding and elimination on the candidate pattern chain, and output a stable pattern chain. S4. Using the stable pattern chain as input, generate reliable mining results for the object to be mined.

2. The data mining intelligent processing and analysis method according to claim 1, characterized in that, S1 further includes traversing each action in the action sequence, extracting the timestamp attribute and scope label of each action; the action sequence is expanded into continuous action tuples along the time axis, and the action tuples with temporal intersection with the action are located based on the timestamp attribute and scope label, and the action tuples with temporal intersection are marked as disturbed tuples.

3. The data mining intelligent processing and analysis method according to claim 2, characterized in that, The continuous disturbed tuples are merged into intervals to generate responsive segments; for each responsive segment, the start boundary time and the end boundary time of the responsive segment in time sequence are taken to form a processing cut, and all the processing cuts constitute the processing cut set.

4. The data mining intelligent processing and analysis method according to claim 1, characterized in that, S2 further includes obtaining the start boundary time and end boundary time corresponding to each treatment cut in the treatment cut set; expanding the behavior sequence into a behavior tuple sequence according to its own temporal order; and segmenting the behavior tuple sequence using the start boundary time and the end boundary time as cutting reference points.

5. The data mining intelligent processing and analysis method according to claim 4, characterized in that, After the behavior tuple sequence is segmented and truncated, any segmented sequence fragment does not contain any complete internal interval of the treatment cut within its temporal range.

6. The data mining intelligent processing and analysis method according to claim 5, characterized in that, Extract the temporally continuous and semantically consistent sequence segments after the segmentation and truncation as the original evolutionary segments, wherein the start and end times of any one of the original evolutionary segments do not fall between the start and end boundary times of the same processing cut.

7. The data mining intelligent processing and analysis method according to claim 1, characterized in that, S3 further includes performing frequent pattern mining on each of the original evolutionary segments, extracting high-frequency and temporally ordered behavioral subsequences within the original evolutionary segments to form initial pattern fragments corresponding to each of the original evolutionary segments; and splicing together the initial pattern fragments that are temporally adjacent and semantically compatible to generate the candidate pattern chain.

8. The data mining intelligent processing and analysis method according to claim 7, characterized in that, The effective time period of each treatment action in the treatment sequence is obtained, and each candidate pattern chain is projected onto the effective time period for matching and verification. If a candidate pattern chain overlaps with the effective time period of any treatment action in its time extension interval, the candidate pattern chain is marked as a reinjection pattern chain. The reinjection pattern chain is removed from all the candidate pattern chains, and the candidate pattern chains that do not overlap with the action period of the treatment sequence are retained and output as the stable pattern chain.

9. The data mining intelligent processing and analysis method according to claim 1, characterized in that, S4 further includes obtaining the pattern support and pattern confidence of each stable pattern chain, filtering stable pattern chains that meet the conditions according to a preset reliability threshold to form a reliable pattern chain set; performing temporal merging and semantic compression on each stable pattern chain in the reliable pattern chain set to obtain a summary of the behavioral patterns of the object to be mined.

10. The data mining intelligent processing and analysis method according to claim 9, characterized in that, The behavioral pattern summary is compared and verified with the treatment motivations recorded in the treatment sequence, and the remaining treatment-induced patterns are removed to obtain a cleaned behavioral pattern summary. The cleaned behavioral pattern summary is used as the reliable mining result of the object to be mined, wherein the reliable mining result represents the original behavioral pattern of the object to be mined before being disturbed by the treatment action.

Images