An artificial intelligence-based multi-source data labeling method and system

Abstract

The application discloses a multi-source data labeling method and system based on artificial intelligence, and relates to the technical field of artificial intelligence, and comprises the following steps: collecting multi-source data in a continuous time range, and recording the change amplitude, change direction and occurrence sequence of each source at the same time position in the collecting process to form a change record corresponding to the time position. The application records the changes of multi-source data in a continuous time range, splits the superimposed changes at the same time position, realizes the source differentiation of mixed changes, and improves the labeling accuracy. Meanwhile, by constructing a continuous change process and establishing a change corresponding relationship, the change sources are divided and labeled in time sequence, so that the labeling result has continuity and consistency, and can accurately reflect the change conditions of the sources.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and specifically to a multi-source data annotation method and system based on artificial intelligence. Background Technology

[0002] AI-based multi-source data annotation refers to a data processing method that utilizes artificial intelligence technology to uniformly identify, understand, and automatically label data from different sources, types, and structures. Specifically, this technology collaboratively analyzes multiple types of data, including images, text, speech, and sensor signals, extracting key features and inherent relationships from each type. Based on preset rules or learned discriminative abilities, it then categorizes, labels attributes, or assigns semantic values ​​to the data content. During processing, it not only automatically labels individual data points but also performs cross-correction and consistency constraints based on the correspondences between multiple data sources, thereby improving the accuracy and completeness of the annotation results. This method is widely used in scenarios such as training dataset construction, intelligent model optimization, and data-driven decision-making, and is a crucial foundational technology supporting the performance improvement of AI systems.

[0003] The existing technology has the following shortcomings: In existing technologies, the annotation of multi-source data typically relies on direct analysis and feature extraction of changes in each source at the same time point. However, in actual dynamic processes, data from different sources can influence each other and overlap at certain time points, resulting in a mixed state of multiple changes in the received output. In this case, the processing can easily misinterpret this overlapping change as originating from a single data feature, leading to misjudgment of the source of change and biased identification of the labeled object. This problem is particularly prominent in scenarios with complex changes or high correlation between multiple sources, and can adversely affect subsequent data analysis and model training results.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-source data annotation method and system based on artificial intelligence to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multi-source data annotation method based on artificial intelligence, comprising the following steps: Multi-source data is collected over a continuous time period, and the magnitude, direction, and order of change of each source at the same time position are recorded during the collection process to form a change record for the corresponding time position. Based on the change records, the changes at the same time location are split and processed. The superimposed comprehensive changes are expanded according to their sources to obtain multi-source change records at the corresponding time location, and the original changes of each source at the corresponding time location are preserved. For multi-source change records, the changes of each source in adjacent time positions are continuously organized, and the changes of the same source in consecutive time positions are sequentially connected to form the continuous change process corresponding to each source. By combining continuous change processes, overlapping changes at the same time point are re-marked, and the overlapping parts are assigned to the corresponding continuous change processes and arranged according to the change order of each source to form a change correspondence relationship; Based on the correspondence of changes, the sources of change at the same time location are divided, mixed changes are assigned to their corresponding sources, and the labels are completed point by point in chronological order to obtain a labeling result with clear sources.

[0007] Preferably, the processing steps for constructing multi-source data change records, focusing on the unified expression and correlation organization of change information from multiple sources within a continuous time range, are as follows: Simultaneously collect data from multiple sources and establish a unified time scale. Divide the continuous time range into multiple time positions and record each data source at each time position to form a set of original change data containing time position identifiers and source identifiers. Based on the original set of change data, the source data within each time point are compared and processed to calculate the corresponding change magnitude of each source, mark the direction of change, and determine the order of occurrence according to the time of change to form a multi-dimensional change information set; By combining multidimensional change information sets, change information from various sources within the same time location is collected and organized, arranged according to the magnitude and direction of change, and change units are constructed according to the order of appearance. By connecting the corresponding change units at each time position and linking them in chronological order to form a change sequence, and using time position identifiers to achieve continuous expression, a change record covering a continuous time range is obtained.

[0008] Preferably, the processing steps for splitting and expanding the sources of superimposed changes at the same time location, focusing on the decomposition, matching, and recombination of comprehensive changes in the change records, are as follows: Extract the changes corresponding to each time position in the change record, identify the comprehensive changes within the same time position, and break down the comprehensive changes into change segments based on the magnitude, direction, and order of occurrence of the changes. Matching changed segments involves mapping the changed segments to their corresponding original source changes, determining their attribution based on the magnitude and direction of the changes, and forming a set of changed segments. The collection of variation fragments is integrated, and variation fragments corresponding to the same source are gathered and arranged in chronological order of appearance to form the variation expression corresponding to the source; Record the changes corresponding to the source, and store the magnitude, direction, and order of change in relation to the original changes to form a multi-source change record.

[0009] Preferably, the set of change segments is further subdivided, and the uncompleted change segments are divided according to the direction of change. The change segments are then matched again based on the magnitude of change. The subdivided change segments are then assigned to their corresponding sources, while maintaining consistency in the magnitude of change, direction of change, and order of appearance, thus forming a change expression with a clear source correspondence.

[0010] Preferably, for the temporal continuity of changes from various sources in multi-source change records, the following steps are taken for connecting and expanding adjacent time positions: Collect changes from the same source in multiple change records, sort them by time position to form a continuous change sequence, and record the magnitude, direction and order of change. Connecting change sequences involves linking changes in adjacent time positions to form continuous change segments through the continuous relationship between the magnitude and direction of change, while maintaining a consistent order of appearance. Extend the continuous change segments, connect adjacent continuous change segments one by one to form a continuous change chain covering multiple time positions, and record the change amplitude and direction at each time position; Organize the continuous change chain, arrange the changes at each time position in chronological order and record the order of occurrence synchronously to form the continuous change process corresponding to each source.

[0011] Preferably, in the process of connecting changes in adjacent time positions, by maintaining the continuous arrangement of change amplitudes and synchronously marking the continuity of change direction, the changes in each time position are embedded into a continuous change chain according to their order of appearance, forming a continuous change process with a time correlation.

[0012] Preferably, for the labeling and attribution of overlapping changes within the same time location, the steps for association adjustment and arrangement construction around the continuous change process are as follows: The continuous change process and the change records from multiple sources are organized accordingly. The change content within the same time position is extracted, and the overlapping change content is identified and a set of overlapping changes is formed by combining the change magnitude, change direction and the order of occurrence. Process overlapping sets of changes, add a source identifier to each change, and retain the magnitude, direction, and order of appearance of the change to form an independent change record; The independent change records are then inserted into the time nodes of the continuous change process. The change in time position and direction before and after the change are sequentially connected, and the order of occurrence is embedded. Arrange the changes in a continuous process, sort the changes from different sources at the same time point in chronological order, and associate them with the continuous process to form a correspondence between changes.

[0013] Preferably, in the process of recognizing overlapping changes, the changes are aggregated based on the consistency between the magnitude and direction of change. During the change marking process, the magnitude, direction, and order of change are kept consistent. When the changes are included in a continuous change process, they are inserted sequentially according to the time nodes, and the change order is kept continuous, thereby forming a stable change correspondence.

[0014] Preferably, for the classification and labeling of change sources within the same time location, the steps for splitting, connecting, and labeling the corresponding relationships of changes are as follows: Extract the changes at each time position in the change correspondence, decompose the changes at the same time position, break down the changes into change units and assign source identifiers, and record the change magnitude, change direction and order of occurrence. Organize the change units, arrange the change units within the same time position according to the order of their appearance, and distinguish different sources to form a structured change set; By connecting structured change sets, change units corresponding to the same source at different time locations are sequentially linked together, and a continuous change sequence is formed by combining the change magnitude and change direction, while maintaining the consistency of the order of appearance; The continuous change sequence is labeled, and the change units at each time position are recorded in chronological order. The source identifier, change range, change direction and order of appearance are associated to form a labeling result with a clear source.

[0015] An AI-based multi-source data annotation system includes a multi-source change information acquisition module, a superimposed change decoupling module, a single-source change trajectory generation module, a coincident change relationship reconstruction module, and a source determination and annotation module. The multi-source change information acquisition module collects multi-source data over a continuous time range and records the magnitude, direction, and order of change of each source at the same time location during the acquisition process, forming a change record for the corresponding time location. The overlay change decoupling module, based on change records, splits the change content at the same time position, expands the comprehensive changes presented by overlay according to their sources, obtains multi-source change records at the corresponding time position, and retains the original change situation of each source at the corresponding time position. The single-source change trajectory generation module continuously organizes the changes of each source in adjacent time positions for multi-source change records, and sequentially connects the changes of the same source in front and behind time positions to form the continuous change process corresponding to each source. The overlapping change relationship reconstruction module, in conjunction with the continuous change process, re-marks the overlapping change content at the same time position, assigns the overlapping parts to the corresponding continuous change process, and arranges them according to the change order of each source to form a change correspondence relationship; The source determination and labeling module classifies the sources of change at the same time location based on the change correspondence, assigns mixed changes to their corresponding sources, and completes the labeling point by point in chronological order to obtain labeling results with clear sources.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: Beneficial Effect 1: This invention records changes in multi-source data over a continuous time range, constructing change records by combining the magnitude, direction, and order of change. It then separates superimposed changes at the same time point, unfolding the combined changes according to their source, thereby distinguishing the sources of mixed changes. This method allows previously intertwined changes to be presented as independent sources, avoiding annotation bias caused by confusion of change sources, making the data annotation process more accurate, and providing a reliable data foundation for subsequent data analysis.

[0017] Benefit 2: This invention constructs a continuous change process corresponding to each source, and re-labels and adjusts the attribution of overlapping changes based on this continuous change process. It establishes a change correspondence at the same time location, further classifies the change sources based on this correspondence, and completes the annotation point by point in chronological order, thus forming a clearly defined annotation result. This method enables the continuous expression of the change relationships of multi-source data in the time dimension, improves the consistency of the annotation results, and ensures that the annotation results fully reflect the actual situation of changes from each source. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a flowchart of a multi-source data annotation method based on artificial intelligence according to the present invention.

[0020] Figure 2 This is a schematic diagram of a multi-source data annotation system based on artificial intelligence according to the present invention. Detailed Implementation

[0021] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0022] This invention provides, for example Figure 1 The method for labeling multi-source data based on artificial intelligence, as shown, includes the following steps: Multi-source data is collected over a continuous time period, and the magnitude, direction, and order of change of each source at the same time position are recorded during the collection process to form a change record for the corresponding time position. In multi-source data processing, by meticulously recording the changes in data from various sources within a continuous timeframe, a change log with temporal correlation and source differentiation capabilities can be formed, providing a reliable basis for subsequent change breakdown and source classification. The specific implementation method for this process is as follows: Multi-source data is synchronously collected within a set continuous time range. By configuring a unified time base for each source of data, the data from different sources are aligned according to a unified time scale. During the collection process, the continuous time is divided into multiple interconnected time positions, and data content from different sources is acquired at each time position. At the same time, the status of each source of data at the current time position is recorded item by item, including recording the numerical change of the source at this time position relative to the previous time position, and the change is saved in the form of specific numerical values, thereby forming a set of original change data containing time position identifiers and source identifiers.

[0023] Throughout the data collection process, a consistent time division and recording method was used for all data sources to ensure that data from different sources have a corresponding relationship at the same time position, thus providing a unified data foundation for subsequent analysis of change characteristics. Based on the completion of multi-source data acquisition and time alignment, the change range of each source data within each time position is calculated and recorded one by one. By comparing the corresponding data values ​​of the current time position with those of the previous time position, the change value of that source at that time position is obtained, and the change value is stored as the change range. While recording the change range, the change direction is clearly distinguished. The increase of the value is marked as a positive change, and the decrease of the value is marked as a negative change. The change direction is then appended to the corresponding change range data in a clear identifier form.

[0024] Furthermore, for the change process of multiple sources of data within the same time location, by comparing the time points of change of each source one by one, the source with the earlier change time is placed in the first position and the source with the later change time is placed in the second position, and recorded in this order, thereby forming a multi-dimensional change information set that includes the magnitude of change, the direction of change, and the order of occurrence. After obtaining information on the magnitude, direction, and order of change of data from various sources within the same time location, the change information of all sources within that time location is centrally organized. The magnitude, direction, and order of change corresponding to each source are uniformly collected and grouped according to time location. All change information from sources belonging to the same time location is integrated into a complete change unit.

[0025] In this process, the magnitude of change from each source is arranged according to its numerical value. At the same time, the direction of change is used to distinguish different trends. The arrangement results are then sorted again according to the order of appearance, so that the change information from multiple sources at the same time position forms a combination structure with a clear sequential relationship. In this way, each change unit at a time position not only contains change information from a single source, but also reflects the arrangement relationship of changes from multiple sources at that time position, thus forming a change expression form with internal correlation. After constructing the change units at each time position, the change units at all time positions within the continuous time range are connected one by one in chronological order. The change units at the previous time position are sequentially connected with the change units at the next time position, and each change unit is assigned a corresponding time position identifier, so that the change information at each time position forms a continuously arranged change sequence.

[0026] In this change sequence, each time position corresponds to an information set containing the magnitude, direction, and order of change from multiple sources. At the same time, adjacent time positions maintain a temporal continuity, thus constructing a change record covering the entire continuous time range. This change record can fully present the change relationship of each source at the same time position and the evolution process of change in the time dimension, providing clear data support for subsequent decomposition of superimposed changes and classification of change sources.

[0027] Based on the change records, the changes at the same time location are split and processed. The superimposed comprehensive changes are expanded according to their sources to obtain multi-source change records at the corresponding time location, and the original changes of each source at the corresponding time location are preserved. After generating change records covering a continuous time range, to eliminate the mixed effects of multiple source changes overlapping at the same time location, it is necessary to split the change content at the same time location in the change records. By expanding the comprehensive changes at the source level, each source change is presented as an independent state again, thereby obtaining multi-source change records with clear sources, while maintaining the original change status of each source at the corresponding time location during the splitting process. The specific implementation method for this process is as follows: The process involves extracting the change content corresponding to each time position from the change records, centrally reading all source change information contained within that time position, identifying the composite changes that are superimposed, and breaking down the composite changes into multiple distinguishable change components. During the decomposition process, the change amplitude, change direction, and order of appearance already contained in the change records are used as the basis for analyzing the change content within the same time position item by item. The numerical changes corresponding to the change amplitude are distinguished according to the change direction, and the arrangement position of each change component is determined by combining the order of appearance. In this way, the originally mixed composite changes are divided into several change segments with independent source orientations, so that each change segment can be associated with a specific source, and a preliminary source separation result is formed. After obtaining the initial segmentation of change fragments, source matching is performed on each change fragment. The change fragment is matched one by one with the original change situation of the corresponding source in the change record at that time position. By comparing the consistency between the change amplitude and change direction of the change fragment and the original change situation, the change fragment with a high degree of matching is assigned to the corresponding source. The change fragments that do not match completely are further subdivided according to the change direction and change amplitude, so that each subdivided change fragment can form a corresponding relationship with the original change situation of a certain source.

[0028] Throughout this process, the relationship between the changed fragments and the original changes is maintained, thereby ensuring the consistency of the splitting results in terms of source attribution. After establishing the correspondence between the change segments and their sources, the change segments corresponding to each source within the same time position are integrated. Multiple change segments belonging to the same source are gathered together and arranged in chronological order of appearance, so that the changes of each source at that time position can be presented in a continuous and orderly form.

[0029] During the integration process, the magnitude and direction of change are kept consistent with the original changes, and the sequential relationship between multiple change segments is preserved, thereby forming a complete expression of change for each source at that time position. This ensures that the split change content can reflect both the independence of the source and the temporal sequence of the changes. After integrating all source change fragments within the same time location, the change content corresponding to each source is uniformly recorded to form a multi-source change record for that time location. During the recording process, the original change situation of each source at that time location is retained, and the original change magnitude, change direction, and order of appearance are stored in correspondence with the split change results, so that the change of each source includes both the independent change expression after splitting and the original change information.

[0030] By repeating the above processing steps at each time point within a continuous time range, a multi-source change record covering the entire time range is finally formed. This allows the superimposed comprehensive changes to be unfolded according to their respective sources, thus providing a clear data foundation for the construction of subsequent continuous change processes and source classification.

[0031] For multi-source change records, the changes of each source in adjacent time positions are continuously organized, and the changes of the same source in consecutive time positions are sequentially connected to form the continuous change process corresponding to each source. Given that multi-source change records have already been established, by continuously organizing the changes from each source at adjacent time points, change information scattered across different time points can be systematically connected, thus forming a continuous change process reflecting the continuity of changes. This processing method, by linking the changes from the same source point by point at consecutive time points, allows the change trajectories of each source to be fully presented in the time dimension. The specific implementation method for this process is as follows: The changes from each source within a continuous time range are extracted one by one from the multi-source change records. Change data belonging to the same source are grouped according to time position, and the grouping results are sorted according to the order of time position, so that the changes of each source at different time positions can form a continuous change sequence in chronological order.

[0032] During the classification process, the change magnitude corresponding to the source within each time point is read value by value, the direction of change is clearly marked, and the order in which the source appears at that time point is recorded, so that each time point contains complete change information; by uniformly collecting the change information of the same source in multiple time points, the change performance of the source in a continuous time range forms a traceable change sequence, providing continuous data support for the connection between subsequent adjacent time points; After forming a time-ordered sequence of changes, the changes between adjacent time positions are connected one by one. The change information from the same source at the previous time position is matched with the change information at the next time position. By continuously comparing the numerical changes in the magnitude of the changes at the previous and next time positions, and combining the changes in the direction of change at the two time positions, the changes at the previous and next time positions are connected.

[0033] During the connection process, the change amplitude of the previous time position and the change amplitude of the next time position are arranged sequentially, and the change direction of the previous time position and the change direction of the next time position are continuously marked. At the same time, the order of appearance of the source in the two time positions is kept unchanged, so that the change information in adjacent time positions can form a continuous change segment, which can reflect the continuity relationship of the change between the two time positions. After completing the connection processing between adjacent time positions, the formed continuous change segments are expanded and integrated. The continuous change segments between multiple adjacent time positions are connected segment by segment in chronological order. By connecting the end of the previous continuous change segment with the beginning of the next continuous change segment, multiple continuous change segments can be combined to form a continuous change chain covering multiple time positions.

[0034] During the expansion process, the magnitude of change at each time point is recorded point by point, and the direction of change is continuously marked in the entire continuous change chain. At the same time, the order of appearance of the source at each time point is kept consistent with the original records, so that the formed continuous change chain can fully reflect the change path of the source within a continuous time range. Through this segment-by-segment connection method, the discrete changes originally distributed in multiple time points are gradually formed into a holistic change expression with temporal continuity. After constructing the continuous change chain for each source, the continuous change chain corresponding to each source is organized and expressed. The change information at each time position in the continuous change chain is arranged in a unified manner according to time order, and the change magnitude, change direction and occurrence order of the source are recorded synchronously at each time position, so that each node in the continuous change chain has a complete change description.

[0035] During the data processing, the entire chain of continuous changes is recorded as a continuous process of change for that source. This continuous process of change not only reflects the change status at a single time point, but also the connection relationship between multiple time points, thus forming a continuous process of change corresponding to each source. This continuous process of change can provide a continuous basis for re-marking overlapping changes and re-dividing the sources of change.

[0036] By combining continuous change processes, overlapping changes at the same time point are re-marked, and the overlapping parts are assigned to the corresponding continuous change processes and arranged according to the change order of each source to form a change correspondence relationship; Given that continuous change processes corresponding to each source have been established, by combining the changes at the same time point with the continuous change processes, overlapping changes from different sources at the same time point can be identified. Through relabeling and attribution adjustments, overlapping changes can be accurately mapped to the continuous change processes of their respective sources, thereby establishing a clear change correspondence. The specific implementation method for this process is as follows: A one-to-one correspondence is established between continuous change processes and time positions in multi-source change records. The change content within each time position is compared and organized with the corresponding continuous change processes from each source. By extracting the change amplitude, change direction, and order of occurrence of each source within that time position, and matching them with the change nodes of the continuous change process at that time position, overlapping change content formed by the combined effects of multiple sources at the same time position can be identified. During the identification process, change content with similar change amplitude values ​​and consistent change directions is grouped together, and similar changes generated by multiple sources at the same time position are merged and identified, so that overlapping changes can be clearly distinguished in combination and form a set of overlapping changes containing source information. After identifying the overlapping change sets, each change in the overlapping change sets is re-marked. Each change is assigned a corresponding source identifier, and the magnitude, direction, and order of change at that time position are fully preserved. During the marking process, changes from different sources in the same overlapping change set are split and recorded one by one, so that the change from each source at that time position can exist as an independent entry, while keeping its original change information unchanged.

[0037] By marking each item in the overlapping change set, the change content that originally existed in the form of superposition can be transformed back into multiple change records with clear source attribution, providing a clear basis for subsequent classification into the continuous change process; After re-marking the overlapping changes, the marked changes are assigned to the corresponding continuous change processes of the source. By inserting the changes from each source at that time position into the corresponding time node of the continuous change process, the change information of the continuous change process at that time position is supplemented and corrected.

[0038] During the incorporation process, the magnitude and direction of change are sequentially connected with the changes in the time positions before and after the continuous change process, so that the changed content after incorporation can maintain a consistent trend with the changes in the time positions before and after. At the same time, the order of appearance of the time position is embedded into the time sequence of the continuous change process, so that the continuous change process forms a complete and continuous change expression at that time position. After the overlapping changes are incorporated into the continuous change process, the changes from each source at the same time position are uniformly arranged. The changes from each source are sorted according to their order of appearance, and the arrangement results are organized in conjunction with the time sequence in the continuous change process, so that the changes from different sources at the same time position can be presented in an orderly form.

[0039] During the arrangement process, the magnitude, direction, and order of change of each source at that time position are associated with and recorded with its corresponding continuous change process. This ensures that the arrangement result not only reflects the sequential relationship between sources but also the positional relationship of each source during the continuous change process, thus forming a complete change correspondence. This change correspondence can clearly describe the correspondence between changes of different sources at the same time position and provide a direct basis for the subsequent reclassification of change sources.

[0040] Based on the correspondence of changes, the sources of change at the same time location are divided, mixed changes are assigned to their corresponding sources, and the annotations are completed point by point in chronological order to obtain annotation results with clear sources; Once a correspondence of changes has been established, by meticulously dividing the sources of change at the same time point and assigning mixed changes to their corresponding sources one by one, the originally intertwined changes can be restored to a clear expression of changes with distinct sources. Simultaneously, by combining this with point-by-point annotation in chronological order, changes throughout the entire continuous time range have a clear source indication. The specific implementation method for this process is as follows: For each time position recorded in the change correspondence, the change content with established correspondence within that time position is extracted item by item. Each change content is decomposed according to the source direction in the change correspondence, and the change content that originally existed in combination at the same time position is split into multiple individual change units, and each change unit is assigned a unique source identifier. During the splitting process, the change amplitude corresponding to each change unit is read value by value, the change amplitude is preserved in the original numerical form, the change direction is clearly marked, and the change direction and change amplitude are matched one-to-one. At the same time, the order of appearance of the change unit at that time position is completely recorded, so that each split change unit contains four types of information: source identifier, change amplitude, change direction, and order of appearance.

[0041] By splitting each item and assigning source identifiers, mixed changes at the same time location can be transformed into a set of changes from multiple independent sources; Based on the set of changes that have already been categorized by source, the change units belonging to different sources within the same time position are sequentially organized. All change units are arranged according to their order of appearance, with earlier appearing change units placed in the first position and later appearing change units placed in the last position. During the arrangement process, the magnitude and direction of change for each change unit remain unchanged. In this sequential organization, multiple change units within the same source are grouped together, so that changes from the same source at that time position appear as a continuous sequence. Source identifiers are added to the arrangement results to clearly distinguish changes from different sources within the structure. Through this sequential organization, the change units within the same time position not only complete source categorization but also form a structured expression with internal sequential relationships. By combining the results of the changes at each time position within a continuous time range, the change units of each source at different time positions are connected point by point. The change units belonging to a certain source in the previous time position are sequentially connected with the change units of the same source in the next time position. The change amplitudes of the two time positions are continuously arranged in chronological order, and the change direction is continuously marked in the preceding and following time positions. At the same time, the consistency of the order of appearance in the time dimension is maintained, so that the changes of the same source at different time positions can form a change sequence that extends point by point.

[0042] During the connection process, each time position change unit is processed one by one, so that each source forms a change sequence composed of multiple time positions connected throughout the continuous time range. This change sequence can fully reflect the change continuity of the source in the time dimension, thus providing continuous support for the generation of the final annotation results. After constructing the change sequences for each source, the changes for each source within a continuous time range are labeled point by point. The change units at each time position are labeled and recorded sequentially according to time order. During the labeling process, the source identifier, change magnitude, change direction, and order of appearance are recorded synchronously, so that the change at each time position corresponds to clear source information. At the same time, the labeling results of different sources at the same time position are recorded side by side, so that the changes from each source at the same time position form independent and unmixed labeling expressions.

[0043] By labeling all time locations within a continuous time range point by point, a labeling result with a clear source is finally formed. This labeling result can fully present the distribution of changes of each source in the time dimension and accurately reflect the specific changes of each source in each time location, thereby realizing the complete process of source division and labeling output for mixed changes.

[0044] This invention records changes in multi-source data over a continuous time range, constructing a change record by combining the magnitude, direction, and order of change. It then separates overlapping changes at the same time point, unfolding the combined changes according to their source, thereby distinguishing the sources of mixed changes. This method allows previously intertwined changes to be presented as independent sources, avoiding annotation bias caused by confusion of change sources, making the data annotation process more accurate, and providing a reliable data foundation for subsequent data analysis.

[0045] This invention constructs a continuous change process corresponding to each source, and re-labels and adjusts the attribution of overlapping changes based on this continuous change process. It establishes a change correspondence at the same time location, further classifies the change sources based on this correspondence, and completes the annotation point by point in chronological order, thus forming a clearly source-defined annotation result. This method enables the continuous expression of the change relationships of multi-source data in the time dimension, improves the consistency of the annotation results, and ensures that the annotation results fully reflect the actual situation of changes from each source.

[0046] This invention provides, for example Figure 2 The system shown is an AI-based multi-source data annotation system, comprising a multi-source change information acquisition module, a superimposed change decoupling module, a single-source change trajectory generation module, a coincident change relationship reconstruction module, and a source determination and annotation module. The multi-source change information acquisition module collects multi-source data over a continuous time range and records the magnitude, direction, and order of change of each source at the same time location during the acquisition process, forming a change record for the corresponding time location. The overlay change decoupling module, based on change records, splits the change content at the same time position, expands the comprehensive changes presented by overlay according to their sources, obtains multi-source change records at the corresponding time position, and retains the original change situation of each source at the corresponding time position. The single-source change trajectory generation module continuously organizes the changes of each source in adjacent time positions for multi-source change records, and sequentially connects the changes of the same source in front and behind time positions to form the continuous change process corresponding to each source. The overlapping change relationship reconstruction module, in conjunction with the continuous change process, re-marks the overlapping change content at the same time position, assigns the overlapping parts to the corresponding continuous change process, and arranges them according to the change order of each source to form a change correspondence relationship; The source determination and labeling module classifies the sources of change at the same time location based on the change correspondence, assigns mixed changes to their corresponding sources, and completes the labeling point by point in chronological order to obtain labeling results with clear sources.

[0047] The present invention provides an artificial intelligence-based multi-source data annotation method, which is implemented through the aforementioned artificial intelligence-based multi-source data annotation system. For details of the specific method and process of the artificial intelligence-based multi-source data annotation system, please refer to the aforementioned embodiment of the artificial intelligence-based multi-source data annotation method, which will not be repeated here.

[0048] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A multi-source data annotation method based on artificial intelligence, characterized in that, Includes the following steps: Multi-source data is collected over a continuous time period, and the magnitude, direction, and order of change of each source at the same time position are recorded during the collection process to form a change record for the corresponding time position. Based on the change records, the changes at the same time location are split and processed. The superimposed comprehensive changes are expanded according to their sources to obtain multi-source change records at the corresponding time location, and the original changes of each source at the corresponding time location are preserved. For multi-source change records, the changes of each source in adjacent time positions are continuously organized, and the changes of the same source in consecutive time positions are sequentially connected to form the continuous change process corresponding to each source. By combining continuous change processes, overlapping changes at the same time point are re-marked, and the overlapping parts are assigned to the corresponding continuous change processes and arranged according to the change order of each source to form a change correspondence relationship; Based on the correspondence of changes, the sources of change at the same time location are divided, mixed changes are assigned to their corresponding sources, and the labels are completed point by point in chronological order to obtain a labeling result with clear sources.

2. The multi-source data annotation method based on artificial intelligence according to claim 1, characterized in that, The following steps are taken to construct multi-source data change records, focusing on the unified expression and correlation organization of change information from multiple sources within a continuous time range: Simultaneously collect data from multiple sources and establish a unified time scale. Divide the continuous time range into multiple time positions and record each data source at each time position to form a set of original change data containing time position identifiers and source identifiers. Based on the original set of change data, the source data within each time point are compared and processed to calculate the corresponding change magnitude of each source, mark the direction of change, and determine the order of occurrence according to the time of change to form a multi-dimensional change information set; By combining multidimensional change information sets, change information from various sources within the same time location is collected and organized, arranged according to the magnitude and direction of change, and change units are constructed according to the order of appearance. By connecting the corresponding change units at each time position and linking them in chronological order to form a change sequence, and using time position identifiers to achieve continuous expression, a change record covering a continuous time range is obtained.

3. The multi-source data annotation method based on artificial intelligence according to claim 2, characterized in that, The following steps are taken to decompose, match, and reassemble the sources of superimposed changes at the same time location, focusing on the decomposition, matching, and recombination of comprehensive changes in the change records: Extract the changes corresponding to each time position in the change record, identify the comprehensive changes within the same time position, and break down the comprehensive changes into change segments based on the magnitude, direction, and order of occurrence of the changes. Match the changed segments, associate the changed segments with the corresponding original changes, determine the attribution by the magnitude and direction of the change, and form a set of changed segments; The collection of variation fragments is integrated, and variation fragments corresponding to the same source are gathered and arranged in chronological order of appearance to form the variation expression corresponding to the source; Record the changes corresponding to the source, and store the magnitude, direction, and order of change in relation to the original changes to form a multi-source change record.

4. The multi-source data annotation method based on artificial intelligence according to claim 3, characterized in that, The set of change segments is further subdivided. Unfinished change segments are divided according to the direction of change and matched again based on the magnitude of change. The subdivided change segments are then assigned to their corresponding sources, while maintaining consistency in the magnitude of change, direction of change, and order of appearance, thus forming a change expression with a clear source correspondence.

5. The multi-source data annotation method based on artificial intelligence according to claim 3, characterized in that, For the temporal continuation of changes from various sources in multi-source change records, the following steps are taken to connect and extend adjacent time positions: Collect changes from the same source in multiple change records, sort them by time position to form a continuous change sequence, and record the magnitude, direction and order of change. Connecting change sequences involves linking changes in adjacent time positions to form continuous change segments through the continuous relationship between the magnitude and direction of change, while maintaining a consistent order of appearance. Extend the continuous change segments, connect adjacent continuous change segments one by one to form a continuous change chain covering multiple time positions, and record the change amplitude and direction at each time position; Organize the continuous change chain, arrange the changes at each time position in chronological order and record the order of occurrence synchronously to form the continuous change process corresponding to each source.

6. The multi-source data annotation method based on artificial intelligence according to claim 5, characterized in that, In connecting the content of changes in adjacent time positions, by maintaining the continuous arrangement of the change amplitude and synchronously marking the continuity of the change direction, the changes in each time position are embedded into a continuous change chain according to the order of their appearance, forming a continuous change process with a time relationship.

7. The multi-source data annotation method based on artificial intelligence according to claim 5, characterized in that, The steps for labeling and assigning overlapping changes within the same time location, and for constructing the association adjustment and arrangement around the continuous change process, are as follows: The continuous change process and the change records from multiple sources are organized accordingly. The change content within the same time position is extracted, and the overlapping change content is identified and a set of overlapping changes is formed by combining the change magnitude, change direction and the order of occurrence. Process overlapping sets of changes, add a source identifier to each change, and retain the magnitude, direction, and order of occurrence of the change to form an independent change record; The independent change records are then inserted into the time nodes of the continuous change process. The change in time position and direction before and after the change are sequentially connected, and the order of occurrence is embedded. Arrange the changes in a continuous process, sort the changes from different sources at the same time point in chronological order, and associate them with the continuous process to form a correspondence between changes.

8. The multi-source data annotation method based on artificial intelligence according to claim 7, characterized in that, In the process of recognizing overlapping changes, the changes are grouped based on the consistency between the magnitude and direction of change. During the change marking process, the magnitude, direction, and order of change are kept consistent. When the changes are included in a continuous process, they are inserted sequentially according to the time nodes, and the change order is kept continuous, thus forming a stable change correspondence.

9. The multi-source data annotation method based on artificial intelligence according to claim 7, characterized in that, For the classification and annotation of change sources within the same time location, the steps for splitting, connecting, and annotating the corresponding change relationships are as follows: Extract the changes at each time position in the change correspondence, decompose the changes at the same time position, break down the changes into change units and assign source identifiers, and record the change magnitude, change direction and order of occurrence. Organize the change units, arrange the change units within the same time position according to the order of their appearance, and distinguish different sources to form a structured change set; By connecting structured change sets, change units corresponding to the same source at different time locations are sequentially linked together, and a continuous change sequence is formed by combining the change magnitude and change direction, while maintaining the consistency of the order of appearance; The continuous change sequence is labeled, and the change units at each time position are recorded in chronological order. The source identifier, change range, change direction and order of appearance are associated to form a labeling result with a clear source.

10. A multi-source data annotation system based on artificial intelligence, used to implement the multi-source data annotation method based on artificial intelligence as described in any one of claims 1-9, characterized in that, It includes a multi-source change information acquisition module, a superimposed change decoupling module, a single-source change trajectory generation module, a coincident change relationship reconstruction module, and a source determination and labeling module: The multi-source change information acquisition module collects multi-source data over a continuous time range and records the magnitude, direction, and order of change of each source at the same time location during the acquisition process, forming a change record for the corresponding time location. The overlay change decoupling module, based on change records, splits the change content at the same time position, expands the comprehensive changes presented by overlay according to their sources, obtains multi-source change records at the corresponding time position, and retains the original change situation of each source at the corresponding time position. The single-source change trajectory generation module continuously organizes the changes of each source in adjacent time positions for multi-source change records, and sequentially connects the changes of the same source in front and behind time positions to form the continuous change process corresponding to each source. The overlapping change relationship reconstruction module, in conjunction with the continuous change process, re-marks the overlapping change content at the same time position, assigns the overlapping parts to the corresponding continuous change process, and arranges them according to the change order of each source to form a change correspondence relationship; The source determination and labeling module classifies the sources of change at the same time location based on the change correspondence, assigns mixed changes to their corresponding sources, and completes the labeling point by point in chronological order to obtain labeling results with clear sources.

Images