An AI semantic self-discovery and periodic state memory-based cross-source task responsibility closed loop system

Abstract

The application provides a cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory, and relates to the technical field of computer information processing. The cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory comprises a task entry access module, which is used for receiving a periodic task synchronization request and reading a business object record in a main task entry data source; a data source discovery module, which is used for determining a target data source according to link information, interface description information or data source meta-information in the business object record; and a task bearing structure positioning module. By receiving the periodic task synchronization request and establishing a unified processing link around the business object, the target data source range is determined actively according to the business object association information in the initial stage of task circulation, the data coverage limitation generated by relying on fixed data configuration is avoided, and the task acquisition process has stronger scene adaptability.

Description

Technical Field

[0001] This invention relates to the field of computer information processing technology, specifically to a cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory. Background Technology

[0002] The field of computer information processing technology is a comprehensive technical field that studies the use of computer systems to collect, store, transmit, organize, analyze, process, retrieve, share, and apply data, information, and knowledge. It covers multiple directions, including database technology, data integration technology, information retrieval technology, software engineering technology, artificial intelligence technology, natural language processing technology, knowledge graph technology, big data processing technology, cloud computing technology, and business process management technology. Among them, a cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory refers to an intelligent task management system that integrates artificial intelligence semantic understanding, cross-source data discovery, task identification, responsibility subject analysis, state tracking, and periodic memory management technologies. It can automatically discover task information from multiple heterogeneous data sources, identify task content and responsibility attribution, and continuously track the changes of tasks in different time periods through a periodic state memory mechanism, realizing the creation, updating, circulation, supervision, archiving, and closed-loop management of tasks.

[0003] Existing task management models, when faced with multi-source business data, typically focus on aggregating and displaying existing information. Different business systems often maintain their own independent data structures and representations. The same task may use different field names, record formats, and status description rules on different platforms. In actual processing, it is necessary to rely on human experience for understanding and correspondence, making it difficult to maintain consistent data interpretation standards. As the number of task sources continues to increase, managers need to frequently switch between multiple systems to check, which can easily lead to task omissions, biases in responsibility identification, and inaccurate status judgments. For example, a rectification task may be shown as being processed on one platform but recorded as pending confirmation on another platform. The lack of a unified judgment basis can easily cause discrepancies in execution perception. At the same time, during the cross-cycle progress of tasks, there is a lack of a stable correlation mechanism between historical records and current records. When task names are adjusted, record locations change, or data sources are migrated, the original execution trajectory is difficult to accurately continue, leading to problems such as duplicate filing, duplicate tracking, or misjudgment of completion status. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory. This solves the problem that existing task management models, when faced with multi-source business data, typically focus on collecting and displaying existing information. Different business systems often maintain their own independent data structures and expressions, and the same task may use different field names, record formats, and state description rules on different platforms. In actual processing, human experience is required for understanding and correspondence, making it difficult to maintain consistent data interpretation standards.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory, comprising the following modules: The task entry access module is used to receive periodic task synchronization requests and read business object records from the main task entry data source; The data source discovery module is used to determine the target data source based on the link information, interface description information, or data source metadata in the business object record. The task carrier structure localization module is used to identify the task carrier structure that carries task records from the target data source; The AI ​​field semantic recognition module is used to identify task description fields, task status fields, responsible entity fields, deadline fields, priority fields, and task record identifier fields based on field names, field descriptions, field types, sample values, and historical mapping records, and generate field semantic mapping results. The task intent recognition module is used to read task records based on the semantic mapping results of the fields, and generate a unified task status by combining structured status information and task text semantics. The responsibility entity parsing module is used to parse the responsibility entity information corresponding to the task record; The task aggregation module is used to aggregate tasks to be processed according to the responsible party. The task fingerprint generation module is used to generate task fingerprints based on task records; The periodic state memory module is used to store information about historical periodic task execution objects and their corresponding task fingerprints, and to read the state memory of the previous period. The task object management module is used to perform creation, update, reuse, archive, cancellation, postponement or completion processing on task execution objects based on the matching results of the task fingerprint of the current cycle and the task fingerprint of the previous cycle. The AI ​​orchestration module is used to generate task execution objects for the corresponding responsible parties. The execution report module is used to output task synchronization results and exception handling results, thereby forming a closed loop of cross-source task responsibility.

[0006] Preferably, the AI ​​field semantic recognition module constructs a field confidence scoring model by using field name similarity, field description similarity, field type matching degree, sample value feature matching degree, and historical mapping hit degree, and determines the field usage based on the scoring results.

[0007] Preferably, when there are multiple candidate task carrier structures in the target data source, the task carrier structure localization module generates a matching score based on the similarity of field sets, sample value features, and the number of historical tasks, and selects the task carrier structure with a score exceeding a preset threshold as the target task carrier structure.

[0008] Preferably, the task intent recognition module is used to uniformly map the task status in data from different sources to at least one of the following unified task statuses: pending execution, in execution, blocked, overdue, risky, completed, and no processing required, and to perform status correction or conflict marking based on confidence level when the structured status conflicts with the text semantics.

[0009] Preferably, the responsible entity parsing module determines the responsible entity in the order of personnel field parsing, project member relationship matching, organizational relationship matching, historical responsible entity mapping matching, and task text semantic extraction, and outputs the corresponding matching confidence score.

[0010] Preferably, when a stable task record identifier exists, the task fingerprint generation module generates a stable task fingerprint based on the data source identifier, task carrier structure identifier, task record identifier, and responsible entity identifier; when a stable task record identifier does not exist, it generates an approximate task fingerprint based on the project identifier, task text summary, task text vector features, deadline, and responsible entity identifier.

[0011] Preferably, the cycle state memory module stores the execution batch identifier, task execution object identifier, task fingerprint, source data source identifier, source task record identifier, responsible entity identifier, task execution object status, access channel type, and anomaly marker information, and calls the information in subsequent execution cycles to perform task object replacement processing.

[0012] Preferably, the task object management module identifies continued tasks, disappeared tasks, completed tasks, tasks with changed responsible parties, and tasks with conflicting fingerprints by comparing the task fingerprint set of the current cycle with the task fingerprint set of the previous cycle, and performs object reuse, status update, archiving, responsibility transfer, or review processing respectively.

[0013] Preferably, the AI ​​orchestration module generates a structured task execution object based on the responsible entity, project context, task priority, deadline, supervising entity, and unified task status. The task execution object includes an object title, task summary, responsible entity identifier, supervising entity identifier, source project identifier, source data link, and task fingerprint.

[0014] Preferably, it also includes an AI anomaly attribution module, which is used to classify and attribute the causes of data source parsing failure, task carrying structure identification failure, field mapping failure, responsible subject matching failure, task execution object creation failure, periodic status update failure, and task fingerprint conflict anomaly, and generate corresponding repair suggestions. The execution report module associates the anomaly attribution results with the task synchronization results and outputs them.

[0015] This invention provides a cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory. It has the following beneficial effects: This invention receives periodic task synchronization requests and establishes a unified processing link around business objects. At the initial stage of task flow, it proactively determines the target data source range based on the business object's association information, avoiding the data coverage limitations caused by relying on fixed data configurations and making the task acquisition process more adaptable to different scenarios. After identifying the task's carrying structure, it performs multi-dimensional semantic inference by combining field names, field descriptions, field types, sample values, and historical mapping relationships. This provides a unified interpretation basis for field meanings in different data environments, reducing the impact of heterogeneous structure differences on the accuracy of task parsing. During the task content reading stage, it jointly analyzes structured status information and textual semantic information, ensuring that task status judgment is no longer limited to a single field result but reflects the actual execution status of the task, improving the consistency and reliability of status identification. Based on the analysis of the responsible entity, it aggregates the responsibility dimensions of the matters to be processed, making the dispersed tasks more comprehensive. This system forms a centralized view oriented towards the executing entity, improving the efficiency of task distribution and management. By constructing fingerprint information that reflects the core characteristics of tasks and establishing a correlation with historical periodic data, cross-period tasks have continuous identification capabilities, reducing duplicate identification problems caused by record changes, data migration, or changes in source. By combining current task characteristics with historical task characteristics for matching analysis, processing actions such as creation, continuation, archiving, extension, and completion can be automatically determined based on actual evolutionary relationships, enhancing the continuity and accuracy of task lifecycle management. From the perspective of the responsible entity, corresponding execution objects are generated, and synchronous and abnormal results are output, forming a complete and interconnected link between task discovery, responsibility association, status tracking, historical continuation, and execution feedback. This helps improve task consistency, responsibility traceability, and periodic management stability in cross-source business environments, and enhances task collaboration and execution supervision capabilities in complex organizational environments. Attached Figure Description

[0016] Figure 1 This is a system block diagram of the present invention; Figure 2 This is a flowchart of the AI ​​semantic self-discovery process of the present invention; Figure 3This is the periodic state memory and task replacement process of the present invention; Figure 4 This is a flowchart illustrating the identification of the responsible party and the aggregation of tasks in this invention. Figure 5 This is a flowchart of the closed-loop processing of task responsibilities in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example: like Figure 1-5 As shown, this embodiment of the invention provides a cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory, including the following modules: The task entry access module is used to receive periodic task synchronization requests and read business object records from the main task entry data source; The interface layer first reads the object set in the main business database, then generates a pending index according to the records. For each object, it extracts the association identifier, time tag, source tag, and status fragment to build a temporary processing unit. For example, after reading a record of a certain matter from the project management platform, it extracts the object number, update time, and associated address. It calculates the completeness coefficient C = (number of fields - number of empty fields) / number of fields according to the record completeness. When C is in the preset qualified range, it enters the subsequent process. When C is below the preset range, it registers the exception and forms the corresponding processing queue. Finally, it obtains a standardized object set for subsequent synchronization.

[0019] The data source discovery module is used to determine the target data source based on the link information, interface description information, or data source metadata in the business object record. The associated address, interface description, and source attribute are parsed sequentially, and a source candidate list is built. A matching score is calculated for each candidate. Where A represents address matching degree, B represents interface feature matching degree, and C represents source attribute consistency degree. Based on historical statistical results, a proportional parameter with a total of one is set. For example, when an object is associated with multiple business platforms, each candidate source is scored, and the source with the score in the preferred range is selected as the subsequent reading object, and the corresponding source location result is output.

[0020] The task carrier structure localization module is used to identify the task carrier structure that carries task records from the target data source; After the target source is determined, the data tables, view structures, and record containers within the source are traversed to extract the structure name, number of fields, record size, and update time information. The structure relevance R is calculated as (number of matching structural features) / (total number of features). For example, when there are multiple data tables in the collaborative system, the number of task description fields, status fields, and responsibility fields in each data table are counted and sorted according to relevance. When R enters the preset valid range, it is included in the candidate set. After round-by-round comparison, the actual record location is determined.

[0021] The AI ​​field semantic recognition module is used to identify task description fields, task status fields, responsible entity fields, deadline fields, priority fields, and task record identifier fields based on field names, field descriptions, field types, sample values, and historical mapping records, and generate field semantic mapping results. Each field in the candidate structure is parsed, including the field name text, field remarks, data type characteristics, and sample content. A semantic scoring matrix is ​​generated by combining this with historical correspondences, and a comprehensive field score is set. M1 represents name similarity, M2 represents note similarity, M3 represents type consistency, and M4 represents sample feature matching. For example, if a field sample presents a set of personnel names, then the corresponding responsibility attribute score will be improved. After sorting all field scores, a set of field role correspondence relationships will be formed.

[0022] The task intent recognition module is used to read task records based on the semantic mapping results of the fields, and generate a unified task status by combining structured status information and task text semantics. The structured state information and text content are processed simultaneously. First, state labels are extracted, then action words, time words, and processing words in the descriptive statements are extracted. Proportional parameters are assigned to each feature to form a state score. Where Wi represents the corresponding feature weight and Pi represents the feature matching result. For example, when the recorded text contains semantic fragments such as "to be followed up" or "to be confirmed", the corresponding state score changes, and finally a unified state identifier is generated and written into the processing intermediate table.

[0023] The responsibility entity parsing module is used to parse the responsibility entity information corresponding to the task record; The system parses the personnel, organization, and role information in the records, establishes association mappings for the name, account, and department fields, and calculates the identity consistency K = (number of matching features) / (total number of features). For example, when the same matter involves both a person's name and an account identifier, cross-verification is performed through the association database. If multiple candidate entities exist, they are sorted according to consistency and the candidate with the higher priority is retained. Subsequently, the corresponding responsibility attribution record is generated.

[0024] The task aggregation module is used to aggregate tasks to be processed according to the responsible party. After the responsibility attribution record is formed, a task set is established according to the subject identifier. For each subject, the number of related matters, status distribution, and time distribution are statistically analyzed, and aggregation indicators are set. The weight of an item comes from the item level information, and the status coefficient comes from the current status classification. For example, when a person in charge is associated with multiple items to be processed, they are grouped into the same responsibility set and an aggregated result record is generated for subsequent object generation processes to call.

[0025] The task fingerprint generation module is used to generate task fingerprints based on task records; For each item, core features are extracted, including title fragments, responsibility identifiers, time tags, and key attributes. These features are then standardized to generate a unique feature string. The feature value H is set as Hash (feature set). For example, two item records from different sources but with highly consistent content will have the same standardized feature set. After encoding, they will obtain the same or similar recognition results and form corresponding identification data.

[0026] The periodic state memory module is used to store information on historical periodic task execution objects and corresponding task fingerprints, and to read the previous periodic state memory. When there is no previous periodic state memory, the system initializes the previous periodic task fingerprint set to empty and directly enters the current round of task execution object generation process. After the system completes the creation, updating and replacement of the current round of task execution objects and the previous round of object replacement, it writes a new periodic state memory snapshot. Establish a cycle history database to store object records generated from each processing cycle, and retain the corresponding identification results, time tags, and processing trajectories. At the beginning of a new cycle, read the historical set and calculate the historical matching rate P = (number of matched records) / (total number of historical records). For example, if the current item and a historical item have the same identification result, include them in the scope of association analysis, retrieve relevant historical data, and form an association set between the current cycle and the historical cycles.

[0027] The task object management module is used to perform creation, update, reuse, archive, cancellation, postponement or completion processing on task execution objects based on the matching results between the task fingerprint of the current cycle and the task fingerprint of the previous cycle. The current period's identification results are compared item by item with the historical set to calculate the similarity. When Q is in different preset intervals, different processing actions are corresponding to it. For example, when it is in the continuous interval, the existing object is used; when it is in the change interval, the object attributes are updated; when it is in the termination interval, the end status is registered; new items are created for unmatched records; archive checks are performed on historical records that are currently missing; and the object processing result set is output.

[0028] The AI ​​orchestration module is used to generate task execution objects for the corresponding responsible parties. Based on the responsibility set and the object processing result set, an execution unit is constructed. First, the information of the responsible entity, the information of the matter set, and the time information are organized. Then, the content is organized according to the preset template, and the content completeness rate U = (number of information items filled) / (number of information items to be filled) is calculated. For example, when a responsible entity is associated with multiple matters, the relevant content is combined into a unified execution unit, and the matter content is arranged in chronological order to form the corresponding execution object record.

[0029] The execution report module is used to output task synchronization results and exception handling results, thereby forming a closed loop of cross-source task responsibility.

[0030] The processing records throughout the entire cycle are summarized, and statistics are compiled on source retrieval, object processing, and anomaly registration, with an anomaly percentage set. The data is categorized and statistically analyzed according to source type, responsible entity type, and processing result type. For example, when a source undergoes structural changes in this cycle, the corresponding anomaly is written into the statistical record, and the processing trajectory and result information are recorded. Finally, synchronized result data and anomaly result data are generated.

[0031] The AI ​​field semantic recognition module constructs a field confidence scoring model based on field name similarity, field description similarity, field type matching degree, sample value feature matching degree, and historical mapping hit degree, and determines the field's purpose based on the scoring results.

[0032] First, the name composition of the attribute to be processed is broken down, extracting phrases, abbreviations, industry terms, and historical aliases to form multiple comparison items. Then, a semantic description set is constructed by combining the field annotation content, and the data categories of the attribute are classified and registered. Based on this, feature information such as character length distribution, numerical range distribution, and enumeration frequency in the sample records is extracted. Subsequently, according to a pre-set reference ratio, weights are assigned to name relevance, description relevance, category consistency, sample feature similarity, and historical corresponding record coverage. Each weight is allocated into low, medium, and high intervals based on past statistical results. For example, when the name relevance is high... A higher participation ratio is assigned when the correlation is in a high range, and a lower participation ratio is assigned when it is in a low range. After each indicator is quantified, a weighted sum is used to generate a comprehensive evaluation value. The comprehensive evaluation value can be expressed as the cumulative result of the product of each individual evaluation result and its corresponding weight. Each parameter involved in the calculation comes from the corresponding comparison process between the current attribute and the reference attribute. For example, if the attribute name in a business record has a strong semantic overlap with the task title description, and the sample content presents a similar text structure, then its corresponding evaluation item will obtain a higher grade value. Finally, the attribute category is determined based on the range in which the comprehensive evaluation value is located, and the corresponding attribute usage result is formed.

[0033] When multiple candidate task carrier structures exist in the target data source, the task carrier structure localization module generates a matching score based on the similarity of field sets, sample value features, and the number of historical tasks, and selects the task carrier structure with a score exceeding a preset threshold as the target task carrier structure.

[0034] First, the attributes contained within each candidate object are extracted, and an attribute list is established. Then, text features, status features, and business identifier features are obtained from the sample records to form a feature set. At the same time, the number and level of historical carrying records for each candidate object are counted as participation parameters. During the execution process, the attribute list is matched item by item, and the number of overlapping attributes is compared with the total number of attributes to form a matching ratio. Then, the feature similarity is calculated by combining the keyword distribution in the sample content. Subsequently, the scale of historical carrying records is divided into multiple level intervals, with different levels corresponding to different participation coefficients. For example, candidates in the middle level interval use the middle level participation coefficient, and candidates in the high level interval use the high level participation coefficient. The evaluation results are combined and calculated according to the preset ratio to form the overall evaluation result. The overall evaluation result can be expressed as the cumulative value of attribute matching results, sample feature matching results, and historical record evaluation results. For example, if a candidate object contains most of the corresponding attributes, and the business identifier in the sample content shows a high degree of similarity with the target record, and the historical carrying scale is at a high level, then its overall evaluation result is in a high interval. Afterward, the overall evaluation results of each candidate object are sorted and compared, and the target object is confirmed according to the preset judgment interval to form the corresponding business container identification result.

[0035] The task intent recognition module is used to uniformly map the task status in data from different sources to at least one of the following unified task statuses: pending execution, in execution, blocked, overdue, risky, completed, and no processing required. When the structured status conflicts with the text semantics, the module performs status correction or conflict marking based on confidence level.

[0036] First, structured status content and text description content are extracted to form a dual-status dataset. A status correspondence table is established for the structured status content, and keyword splitting, phrase recognition, and semantic classification are performed on the text description content. Then, status evaluation results are generated according to preset mapping rules. The structured status evaluation results are derived from the standard classification relationship corresponding to the status field content, while the text status evaluation results are derived from semantic features such as progress descriptions, abnormal descriptions, and completion descriptions appearing in the text. When the two types of evaluation results are consistent, the corresponding status is directly recorded. When the two types of evaluation results differ, the credibility level is further calculated. The credibility level is formed by the combined participation of data integrity level, text matching level, and historical consistency level to form a comprehensive evaluation value. Each participating item is accumulated according to a preset ratio. For example, if the status field of a record shows "processing in progress," while the text content shows descriptions of long-term stagnation and waiting for processing, different evaluation results are generated. Then, the comprehensive evaluation values ​​corresponding to the two types of evaluation results are compared. When one evaluation value is in the high-level range and is significantly higher than the other, that evaluation result is adopted. When both evaluation values ​​are in the same range, it is recorded as a status information to be verified. Finally, a unified status classification result or a status difference identification result is formed.

[0037] The responsible entity parsing module determines the responsible entity in the following order: personnel field parsing, project member relationship matching, organizational relationship matching, historical responsible entity mapping matching, and task text semantic extraction, and outputs the corresponding matching confidence score.

[0038] During the execution of personnel field parsing, project member relationship matching, organizational relationship matching, historical responsibility entity mapping matching, and task text semantic extraction, the personnel field of the responsible entity in the task record is first decomposed, including name, employee number, and job code. The frequency and timestamp of the responsible entity's appearance in the task log are then extracted using a simple counting formula. ,in Indicates the responsible party In the mission The number of times it appears in To identify the occurrence time, a threshold for the number of occurrences is set. (Example value range 1-5) Filter out the main responsible parties, then match the filtered personnel information with the project member database, and calculate the matching degree. ,in To the number of successfully matched entries, The total number of entries is used. A match is considered successful if the match score is higher than 0.7. Then, based on the organizational relationship map, the hierarchical distance between the department to which the responsible entity belongs and the department to which the task belongs is extracted, and a hierarchical distance coefficient is used. ,in To organize the node spacing in the graph, coefficients are weighted and added to the matching score. Simultaneously, the historical responsibility entity mapping is invoked, comparing the responsible entity for the current task with entities that frequently undertook this type of task in the past. The text matching score is then calculated using cosine similarity. It also performs semantic extraction on the task text, word segmentation, and vector feature generation. Assign weight to important keywords (Example range: 0.1-0.5), and finally, the matching confidence score is output by combining the scores of the above items.

[0039] When a stable task record identifier exists, the task fingerprint generation module generates a stable task fingerprint based on the data source identifier, task carrier structure identifier, task record identifier, and responsible entity identifier. When a stable task record identifier does not exist, an approximate task fingerprint is generated based on the project identifier, task text summary, task text vector features, deadline, and responsible entity identifier.

[0040] If a stable task record identifier exists, then a combined encoding is performed based on the data source identifier, task hosting structure identifier, task record identifier, and responsible entity identifier, where the data source identifier... and task-bearing structure identifier The task record identifier is assigned an integer value between 0 and 100 through the mapping table. A unique serial number, identifying the responsible party. For employee ID or code, use the weighted summation formula. ,in , , , Examples of possible values ​​are 0.2, 0.3, 0.3, and 0.2. A stable task fingerprint is generated. If no stable task record identifier exists, the project identifier is extracted. Task text summary Task text vector features Deadline With the identification of the responsible entity Through the similarity combination formula Fingerprint calculation, where similarity The example values ​​are 0.6-0.9 obtained by dot product of word vectors, and the weights are... The example range is 0.2-0.4, and the final output is an approximate task fingerprint.

[0041] The cycle state memory module stores the execution batch identifier, task execution object identifier, task fingerprint, source data source identifier, source task record identifier, responsible entity identifier, task execution object status, access channel type, and anomaly marker information, and calls the information to perform task object replacement processing in subsequent execution cycles.

[0042] First, record the execution batch identifier. With task execution object identifier and the task fingerprint Source data source identifier Source task record identifier Responsible entity identification Perform associated storage, and then record the state of the task execution object. The status value can be in the range of 0-2, representing not started, in progress, or completed, and also indicates the type of channel reached. If an error occurs via email, system notification, or push notification, an error flag should be set. ,otherwise After storing all records, the status of historical task objects is compared with that of current periodic task objects through batch iteration, and a simple replacement algorithm is used. Complete the task object replacement process, among which For the backup objects of the previous cycle, This serves as an exception marker, and the iteration ensures that the state of the task execution object is invoked and updated in the new cycle.

[0043] The task object management module compares the current task fingerprint set with the previous cycle's task fingerprint set to identify continued tasks, disappeared tasks, completed tasks, tasks with changed responsible parties, and tasks with conflicting fingerprints, and then performs object reuse, status update, archiving, responsibility transfer, or review processing respectively.

[0044] First, the task identification information extracted within the current period is broken down. This task identification information can be formed by combining responsibility markers, task descriptions, source items, and time features. Then, corresponding feature strings are generated according to a unified character rule. The current period's feature string set is then associated with the historical period's feature string set item by item. During the comparison process, a similarity calculation method is used to obtain the matching value. The similarity can be expressed as the ratio of the number of matching fields to the total number of fields. When the matching value is in a preset high matching range, it is considered that the same task exists continuously. When the matching value is in the middle range, the responsibility markers are further retrieved for cross-verification. When the matching value is in the low matching range, the process proceeds to the review stage. For example, in a project management scenario... If the same item contains the same project source and item description in both periods, and only the responsibility marker changes, then the identity code corresponding to the responsibility marker is extracted first, and the differences between the old and new identity codes are compared. The responsibility transfer trajectory is recorded. For items that existed in the past but have not appeared now, the status record information is retrieved and classified. For items that are newly added now but did not exist in the past, a new management record is created. If multiple items correspond to the same feature string, the keywords of the description item, the features of the source item, and the time feature are extracted for secondary verification and a review record is formed. Then, according to the various identification results, the existing records, the updated status records, the transferred responsibility records, or the review records are called respectively, and finally the corresponding management results are formed.

[0045] The AI ​​orchestration module generates structured task execution objects based on the responsible party, project context, task priority, deadline, supervising party, and unified task status. Each task execution object includes an object title, task summary, responsible party identifier, supervising party identifier, source project identifier, source data link, and task fingerprint. Based on the set of reusable task execution object identifiers, the AI ​​orchestration module generates new task execution objects only for unreused tasks, and updates the object content for reusable tasks without recreating them.

[0046] First, the responsibility and identity information of the participating items is parsed to obtain the corresponding identity code set. Then, the project-related information is retrieved to form context data. Next, the urgency information, planned completion time information, and supervisor identity information are read. All data enters the structured processing process. The urgency level can be divided according to the range of the number of items to be processed. When the number of items is in the higher range, a higher ranking value is assigned. When the number of items is in the middle range, a medium ranking value is assigned. The planned completion time is obtained by obtaining the remaining cycle parameter through the time difference between the current time and the planned time. The shorter the remaining cycle parameter, the higher the ranking value. After unified normalization, all parameters are processed to form a comprehensive ranking parameter. The comprehensive ranking parameter can be expressed as the sum of the products of each normalized parameter and the corresponding configuration coefficient. The configuration coefficient is set according to the historical project management rules and stored in the configuration library. For example, if a project has several items to be executed, a short remaining cycle, and supervisor relationship records, the comprehensive ranking parameter is higher than that of ordinary items. Then, the fields are mapped according to the sorted identity code, supervisor code, project source information, item summary information, and source link information. The data is then packaged into a standard execution carrier according to a unified template, and finally, an information object containing complete related attributes is formed.

[0047] It also includes an AI anomaly attribution module, which is used to classify and attribute the causes of data source parsing failure, task carrying structure identification failure, field mapping failure, responsible subject matching failure, task execution object creation failure, periodic status update failure, and task fingerprint conflict anomaly, and generate corresponding repair suggestions. The execution report module associates the anomaly attribution results with the task synchronization results and outputs them.

[0048] First, an anomaly classification rule base is established, and the runtime records generated during the collection, parsing, mapping, matching, creation, and update phases are imported into the analysis process. For each type of runtime record, anomaly identifiers, trigger times, associated items, and execution log content are extracted. Then, the records are categorized based on log field completeness, field matching rate, and processing result status. The field matching rate can be expressed as the ratio of successfully mapped fields to the number of fields that should be mapped. Records with a matching rate in a preset low range are classified as mapping anomalies; records with missing key structural information are classified as structural recognition anomalies; and records with an identity information that cannot be associated with the identity database are classified as other anomalies. When an anomaly is identified, it is categorized into the subject matching anomaly category. For example, in a project collaboration scenario, if the source record of a certain matter is missing identity information, the system first searches the identity database for associated names, organization information, and historical records for matching. If no valid correspondence is formed, a subject matching anomaly record is generated. At the same time, the corresponding repair rules are retrieved according to the anomaly category. The repair rules include suggestions for field supplementation, mapping relationship adjustment, structure reconstruction, and manual review. Subsequently, the anomaly record is associated and bound with the synchronous processing record, and the results are summarized and output according to a unified report template to form an execution result record containing the anomaly source, processing trajectory, and repair suggestions.

[0049] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory, characterized in that, Includes the following modules: The task entry access module is used to receive periodic task synchronization requests and read business object records from the main task entry data source; The data source discovery module is used to determine the target data source based on the link information, interface description information, or data source metadata in the business object record. The task carrier structure localization module is used to identify the task carrier structure that carries task records from the target data source; The AI ​​field semantic recognition module is used to identify task description fields, task status fields, responsible entity fields, deadline fields, priority fields, and task record identifier fields based on field names, field descriptions, field types, sample values, and historical mapping records, and generate field semantic mapping results. The task intent recognition module is used to read task records based on the semantic mapping results of the fields, and generate a unified task status by combining structured status information and task text semantics. The responsibility entity parsing module is used to parse the responsibility entity information corresponding to the task record; The task aggregation module is used to aggregate tasks to be processed according to the responsible party. The task fingerprint generation module is used to generate task fingerprints based on task records; The periodic state memory module is used to store information about historical periodic task execution objects and their corresponding task fingerprints, and to read the state memory of the previous period. The task object management module is used to perform creation, update, reuse, archive, cancellation, postponement or completion processing on task execution objects based on the matching results of the task fingerprint of the current cycle and the task fingerprint of the previous cycle. The AI ​​orchestration module is used to generate task execution objects for the corresponding responsible parties. The execution report module is used to output task synchronization results and exception handling results, thereby forming a closed loop of cross-source task responsibility.

2. The cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory as described in claim 1, characterized in that: The AI ​​field semantic recognition module constructs a field confidence scoring model based on field name similarity, field description similarity, field type matching degree, sample value feature matching degree, and historical mapping hit degree, and determines the field's purpose based on the scoring results.

3. The cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory as described in claim 1, characterized in that: When multiple candidate task carrier structures exist in the target data source, the task carrier structure localization module generates a matching score based on the similarity of field sets, sample value features, and the number of historical tasks, and selects the task carrier structure with a score exceeding a preset threshold as the target task carrier structure.

4. The cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory as described in claim 1, characterized in that: The task intent recognition module is used to uniformly map the task status in data from different sources to at least one of the following unified task statuses: pending execution, in execution, blocked, overdue, risky, completed, and no processing required. When the structured status conflicts with the text semantics, the module performs status correction or conflict marking based on confidence level.

5. A cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory as described in claim 1, characterized in that: The responsible entity parsing module determines the responsible entity in the order of personnel field parsing, project member relationship matching, organizational relationship matching, historical responsible entity mapping matching, and task text semantic extraction, and outputs the corresponding matching confidence score.

6. The cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory as described in claim 1, characterized in that: When a stable task record identifier exists, the task fingerprint generation module generates a stable task fingerprint based on the data source identifier, task carrier structure identifier, task record identifier, and responsible entity identifier. When no stable task record identifier exists, an approximate task fingerprint is generated based on the project identifier, task text summary, task text vector features, deadline, and responsible entity identifier.

7. A cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory as described in claim 1, characterized in that: The cycle state memory module stores the execution batch identifier, task execution object identifier, task fingerprint, source data source identifier, source task record identifier, responsible entity identifier, task execution object status, access channel type, and anomaly marker information, and calls the information to perform task object replacement processing in subsequent execution cycles.

8. A cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory as described in claim 1, characterized in that: The task object management module compares the current task fingerprint set with the previous cycle's task fingerprint set to identify continued tasks, disappeared tasks, completed tasks, tasks with changed responsible parties, and tasks with conflicting fingerprints, and then performs object reuse, status update, archiving, responsibility transfer, or review processing respectively.

9. A cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory as described in claim 1, characterized in that: The AI ​​orchestration module generates structured task execution objects based on the responsible entity, project context, task priority, deadline, supervising entity, and unified task status. The task execution objects include object title, task summary, responsible entity identifier, supervising entity identifier, source project identifier, source data link, and task fingerprint.

10. A cross-source task responsibility closed-loop system based on AI semantic self-discovery and periodic state memory as described in claim 1, characterized in that: It also includes an AI anomaly attribution module, which is used to classify and attribute the causes of data source parsing failure, task carrying structure identification failure, field mapping failure, responsible subject matching failure, task execution object creation failure, periodic status update failure, and task fingerprint conflict anomaly, and generate corresponding repair suggestions. The execution report module associates the anomaly attribution results with the task synchronization results and outputs them.

Images