An all-scene based artificial intelligence (AI) assistance method

Abstract

The application relates to the technical field of digital asset management, and particularly discloses an artificial intelligence (AI) auxiliary method based on a full scene, which is used for solving the problems of wrong contact binding, single-round willingness misjudgment and key matters not being checked, which cause error pushing risks in existing digital asset management, and comprises the following steps: multi-source data access and dispute state graph generation, first gate processing to generate a reachable unified asset index candidate set, second gate processing to generate a hospital stable anchor point candidate set, third gate processing to establish conflict blocking variable and blocking result, and blocking result write-back and task access control; the first gate processing, the second gate processing and the third gate processing are set on the dispute state graph and are coordinated with each other, and the blocking result write-back is used to constrain the previous recalculation, so that the response pushing access of the guided question and answer mediation assistant is established on the basis of the converged stable state, and the accuracy of object merging, willingness identification and pushing distribution is improved.

Description

Technical Field

[0001] This invention relates to the field of digital asset management technology, and in particular to an AI-assisted method based on all scenarios. Background Technology

[0002] In digital asset management, when intelligent customer service uses artificial intelligence to assist in digital asset management, complaint tickets, after-sales dispute leads, negotiation follow-up targets, and existing dispute customers are usually included in the asset pool as dispute resolution digital assets. They then undergo further processing steps such as identity merging, contact information updating, status stratification, and external outreach. Although existing AI customer service can use AI technology to stratify dispute resolution digital assets and trigger subsequent pushes based on voice interaction results, historical records, and tag rules, there are often problems such as inconsistent contact information left by the same person on different channels, invalid historical contact information, single-round responses not representing stable intentions, and unclear dispute types, demands, amount ranges, processing time points, decision-making roles, or preferred communication channels directly related to the push content and acceptance path. If the object is written into the external push task queue based solely on the current round of identification results or general stratification tags, the unified asset merging error, short-term intention fluctuations, and key issue conflicts will be amplified into incorrect asset queuing, mis-activation of short-term consent, and external pushes without clear verification of sensitive matters. This will cause mismatch in digital asset outreach, decreased push task quality, and increased burden of subsequent manual correction. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide an AI-assisted method based on all scenarios. By setting up a first gating process, a second gating process, and a third gating process that are coordinated with each other on the dispute state graph, and by using the blocking result write-back constraint to recalculate the preceding order, the admission of the guided question-and-answer mediation assistant's response push is established on a stable state after convergence, thereby improving the accuracy of object merging, intention recognition, and push distribution.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A full-scenario AI-assisted method includes:

[0006] Step S1: Obtain the voice interaction data and associated asset records of the dispute resolution object, and construct a dispute status diagram including round nodes, contact information nodes, mediation willingness nodes, and activation of sensitive matters nodes;

[0007] Step S2: Perform the first gating process on the disputed state diagram to generate a candidate set of reachable unified asset indexes. The first gating process performs heterogeneous verification on the current round of contact information and SMS receipt results, key input results, historical valid contact information records and existing binding records according to the preset heterogeneous rebate judgment rules.

[0008] Step S3: Perform the second gating process. Generate a candidate set of stable anchor points based on the candidate set of reachable unified asset index. The second gating process filters valid rounds according to the preset confirmation round and clarification round judgment rules, and performs category merging and polarity merging of the mediation intentions in the target round according to the preset intention merging rules.

[0009] Step S4: Perform the third gating process, establish conflict blocking edges based on the candidate set of intentional stable anchor points and generate blocking results. The third gating process establishes, retains, deletes or eliminates conflict blocking edges according to the preset conflict blocking edge generation and elimination rules.

[0010] Step S5: Write the blocking result back to the dispute state graph, constrain the first gating process and the second gating process to recalculate. The writing back and recalculation are performed according to the preset blocking result writing back and recalculation process until the number of reachable unified asset indexes, intentional stable anchor points and unresolved conflict blocking edges in the dispute state graph meets the preset conditions. Then, generate a push access identifier and write it into the push task queue. Otherwise, generate a supplementary verification task or a manual review task and output the verification Q&A content.

[0011] It should be noted that the technical problem addressed by this invention is that existing technologies typically treat contact verification, intent assessment, and conflict checking as a series of sequential processing steps. This leads to the possibility that subsequent push admission results may still be based on erroneous merging or short-term intent expressions when there are deviations in the preceding assessments. This invention does not simply execute the three processes sequentially, but first forms a candidate set of reachable unified asset indexes and a candidate set of intent stability anchors on the same dispute state graph. Then, the third gating process generates blocking results and writes them back to the dispute state graph, thereby constraining the first and second gating processes to recalculate until the graph state meets the stability condition. Thus, this invention enables push admission determination to be based on the converged graph state, which can simultaneously suppress the problems of erroneous object queuing, false triggering of short-term consent, and extrapolation without clear verification of sensitive matters.

[0012] As a further embodiment of the present invention, in step S1, the contact information node of the dispute status diagram records the contact information value, source round identifier, value confidence status, and verification status, wherein the verification status is determined according to the preset heterogeneous rebate judgment rule; the mediation willingness node records the willingness category, willingness polarity, source round identifier, and merging status, wherein the merging status is determined according to the preset willingness merging rule; and the activated sensitive matter node records the matter category, matter value, source round identifier, and conflict status, wherein the conflict status is determined according to the preset conflict blocking edge generation and elimination rule.

[0013] The edges in the dispute state diagram include round-related edges, unified asset merging edges, intention anchoring edges, and conflict blocking edges. The dispute state diagram also records blocking result write-back flags and recalculation trigger flags, which can be called according to the preset blocking result write-back and recalculation process.

[0014] It should be noted that in existing technologies, voice interaction results, historical asset records, and subsequent verification relationships are mostly stored in a scattered manner, making it difficult to support the completion of state inheritance, result write-back, and edge relationship updates on the same data carrier for the front and rear gating processes. This invention organizes contact information, intentions, and sensitive matters into nodes, and establishes explicit graph relationships through round-related edges, unified asset merging edges, intention anchoring edges, and conflict blocking edges. This enables the candidate results, blocking results, and recalculation constraints of the three gating processes to be expressed on the same disputed state graph, thereby providing a unified data foundation for closed-loop gating.

[0015] As a further aspect of the present invention, in step S2, the first gating process includes:

[0016] The contact information value corresponding to the current round is taken as the first contact information value. The second contact information value is obtained from the SMS receipt result, key input result, historical valid contact information record or existing binding record. The contact information values ​​from different sources are compared in priority, determined in consistency and resolved in accordance with the preset heterogeneous rebate judgment rules. The first contact information value and the second contact information value are compared in consistency.

[0017] When the consistency comparison passes, write the unified asset merging edge into the disputed state graph and generate a candidate set of reachable unified asset indexes.

[0018] When the consistency comparison fails, the corresponding contact information node is marked as pending verification in the dispute status diagram, and a contact information supplementary verification task or contact information confirmation task is generated.

[0019] It should be noted that existing technologies often directly merge objects based on contact information obtained from a single round of speech recognition, or only perform static historical matching, which can easily lead to mis-binding of objects, invalid historical records, or amplification of single-round recognition errors. In this invention, the first gating process does not directly generate a unique merging conclusion, but rather compares the current round's extracted value with data from multiple heterogeneous contact methods for consistency, writes a unified asset merging edge into the disputed state graph, and forms a candidate set of reachable unified asset indexes. This invention first restricts the reachability judgment to the range of contact methods that have passed heterogeneous verification, providing a more reliable object basis for subsequent intention merging and event clearance.

[0020] As a further aspect of the present invention, in the first gating process of step S2, the consistency comparison includes at least one of the following: full value consistency comparison of the number, consistency comparison of the preset number digits, consistency comparison of the identifier, and consistency comparison of the contact information with the existing binding relationship. When at least two different sources simultaneously support the same contact information value, the contact information value is determined as the target value that passes the verification. When there is a conflict between different sources, the verification status of the contact information node in the current round is determined according to the source priority of the preset different source rebate judgment rule. The blocking result written back by the third gating process is used to delete, freeze, or reduce the validity of the unified asset merging edge associated with the blocked item.

[0021] It should be noted that even if existing technologies perform contact information matching, they usually do not change the merging result after the matching is completed, lacking a reverse correction mechanism for subsequent conflicts on the previous merging result. This invention not only specifies multiple consistency comparison methods, but also further specifies that the blocking result of the third gating process write-back can be applied to the validity of the unified asset merging edge. This invention makes the object merging result no longer a one-time static conclusion, but can be constrained and corrected by subsequent conflict events, thereby reducing the risk of erroneous merging results remaining in the push judgment link for a long time.

[0022] As a further aspect of the present invention, the second gating process in step S3 includes:

[0023] Based on the candidate set of reachable unified asset index, multiple round nodes associated with the same asset node through unified asset merging edges are extracted in the dispute state graph. At least two non-adjacent target round nodes that contain a confirmation round or a clarification round are selected from the multiple round nodes. The confirmation round and the clarification round are determined according to the preset confirmation round and clarification round judgment rules. The mediation intention value associated with the target round node is extracted and intention category merging and intention polarity merging are performed.

[0024] When the result of merging the willingness categories and the result of merging the willingness polarities are consistent, write willingness anchor edges between the mediation willingness nodes and generate a candidate set of willingness stable anchor points;

[0025] When any merge result is inconsistent, generate a supplementary verification task and output clarification verification questions and answers.

[0026] It should be noted that existing technologies often directly use single-round responses or responses from adjacent rounds as the basis for judging willingness, which cannot effectively distinguish between short-term consent, polite responses, and genuine stable willingness. In this invention, the second gating process does not average all historical rounds, but rather, based on the candidate set of reachable unified asset indexes given by the first gating process, it merges willingness only from non-adjacent target round nodes that contain confirmation or clarification rounds in between, and writes willingness anchor edges on the graph. This invention establishes willingness judgment based on the results after cross-round confirmation and clarification, improving the authenticity and resilience of the willingness stability anchor point.

[0027] As a further embodiment of the present invention, in step S5, the blocking result written back by the third gating process is used to remove the round nodes associated with the blocking event, freeze the corresponding intention anchor edge, or trigger the recalculation of intention category merging and intention polarity merging.

[0028] Based on the call duration field, call quality score field, and hang-up party label field, the historical round nodes are filtered for valid rounds, and round nodes that meet the preset conditions are marked as confirmation rounds or clarification rounds. The preset rules for determining confirmation rounds and clarification rounds include at least the following: the call duration reaches a preset threshold, the call quality score reaches a preset threshold, the hang-up party label meets the preset end source condition, and the current round of dialogue hits at least one of the contact confirmation, intention confirmation, or matter clarification modes.

[0029] It should be noted that existing technologies typically do not provide clear input criteria for confirmation and clarification rounds, leading to unclear scope of intent merging and arbitrary round selection. In this invention, quantifiable fields such as call duration, call quality score, and hang-up party label are introduced into the effective round selection, and the results are used to mark the confirmation or clarification round. This invention provides a technical basis for repeatable determination of the input rounds in the second gating process, reducing intent bias caused by invalid rounds, abnormally interrupted rounds, or low-quality rounds participating in the merging.

[0030] As a further aspect of the present invention, in step S4, the third gating process includes:

[0031] Based on the candidate set of stable anchor points of intent, the active sensitive item nodes associated with the candidate set of stable anchor points of intent are extracted in the dispute state graph, and the differences of active sensitive item nodes belonging to the same item category are compared.

[0032] When the difference comparison results are inconsistent, write conflict blocking edges between the nodes that activate sensitive items, and generate blocking results based on the conflict blocking edges;

[0033] Among them, activating sensitive matters includes at least one of the following: dispute type, demand direction, amount range, processing time, decision-making role, and preferred communication channel; difference comparison, establishment of conflict blocking edges, and elimination of conflict blocking edges are carried out in accordance with the preset conflict blocking edge generation and elimination rules.

[0034] It should be noted that existing technologies for handling conflicts typically only involve statistical analysis of inconsistencies in ordinary fields, making it difficult to separate the issues that truly affect the push content and delivery path from general information. In this invention, the third gating process only establishes conflict blocking edges for active sensitive issue nodes associated with the candidate set of stable anchor points of intent, and generates blocking results based on the conflict blocking edges. This invention enables subsequent blocking to no longer treat all fields equally, but instead performs targeted blocking on key issues that directly affect the external push link, improving the targeting of conflict identification.

[0035] As a further aspect of the present invention, the stability conditions include at least two of the following: the reachable unified asset index remains unchanged in two consecutive recalculations, the intentional stable anchor point remains unchanged in two consecutive recalculations, and the number of unresolved conflict-blocking edges remains unchanged or is less than or equal to a preset threshold in two consecutive recalculations.

[0036] As a further aspect of the present invention, in the third gating process of step S4, when the stability condition is not met and the number of recalculations reaches the preset upper limit, a supplementary verification task or a manual review task is generated; wherein, the blocking result write-back and recalculation process includes at least: writing the blocking result into the blocking status field of the node activating the sensitive matter, deleting, freezing or reducing the validity of the unified asset merging edge associated with the blocked matter, removing the round node associated with the blocked matter or freezing the corresponding intention anchoring edge, and triggering the first gating process and the second gating process to perform local recalculation or global recalculation.

[0037] It should be noted that although existing technologies introduce write-back and recalculation, the lack of clear termination criteria may lead to non-convergence or infinite loops in the processing link. In this invention, at least two of the following conditions are explicitly defined as remaining unchanged after continuous recalculation: reachable unified asset index, intentional stable anchor point, and number of unresolved conflict blocking edges. An upper limit on the number of recalculations is also set. This invention provides a termination-based convergence criterion for closed-loop gating, ensuring that the gating processes before and after undergo sufficient constraint correction while avoiding the system from falling into an unlimited recalculation loop.

[0038] As a further aspect of the present invention, the method also includes generating and writing structured records based on the dispute status diagram. The structured records include a gender field, a call duration field, a rating field, a call quality rating field, a hang-up party label field, an intention label field, a transfer to human agent status field, a contact information field, a mediation willingness field, a sensitive activation field, a round identifier field, and a task type field.

[0039] The scoring field is generated based on at least two of the following: the first gating processing result, the second gating processing result, the third gating processing result, the blocking result, the number of supplementary verifications, the call quality scoring field, the intention tag field, and the transfer to human agent status field.

[0040] It should be noted that in existing technologies, interface fields are often only used for display and are difficult to integrate with core gating results, leading to a disconnect between subsequent filtering, review, and statistics and the core access logic. In this invention, gating results, blocking results, call quality, intention tags, and transfer to human operator status fields are uniformly written into a structured record, and a scoring field is generated based on multiple fields. This allows the interface-displayed data to be directly connected with the core dispute status diagram and gating results, facilitating subsequent filtering, sorting, review, and tracking according to the same technical link.

[0041] As a further aspect of the present invention, the method further includes pre-storing a preset filtering scheme, wherein the preset filtering scheme is a filtering condition combination template consisting of one or more filtering fields, value conditions corresponding to each filtering field, and logical relationships between multiple value conditions.

[0042] The filtering fields include at least one of the following: gender, call duration, rating, call quality rating, hang-up party label, intention label, transfer to human agent status, gating result, blocking result, and task type.

[0043] In response to the call of the preset filtering scheme, condition matching is performed on the structured record set, the push task queue or the manual review queue, the target record set is output, and the target record set is displayed, distributed to the corresponding queue or exported as a record file.

[0044] It should be noted that although existing technologies form structured records, they usually rely on manual temporary combination of conditions for filtering, making it difficult to integrate gating results, blocking results, and queue processing into a reusable automatic filtering template. In this invention, a preset filtering scheme consisting of filtering fields, value conditions, and logical relationships is pre-stored, and the gating processing result field and blocking result field are included in the filtering field set. The resulting technical effect is that results directly related to the disputed state diagram and closed-loop gating can be repeatedly called, automatically filtered, and targeted, thereby improving the organizational efficiency and consistency of the push task queue and the manual review queue.

[0045] As a further aspect of this invention, the AI-assisted method based on all scenarios can be deployed in the unified business processing platform or intelligent collaborative service platform of an industrial internet platform. Relying on the platform's support capabilities for multi-source data access, structured record management, task queue scheduling, and rule engine control, it unifies the aggregation, parsing, and scheduling of voice interaction data, historical asset records, SMS receipts, key input results, existing binding records, and manual review results. Furthermore, it establishes a full-process industrial internet security management mechanism on the platform side, centered around first, second, and third gated processing. This invention can provide an industrial APP platform, further enhancing the security, accuracy, and controllability of dispute resolution object merging, task access control, and external push distribution in the industrial internet platform environment.

[0046] Compared with the prior art, the present invention has at least the following technical effects:

[0047] This invention uses a dispute state graph as a unified processing carrier. Instead of directly triggering external pushes based on the current round's identification results or general hierarchical labels, it first performs heterogeneous rebates on multi-source contact methods through a first gating process, forming a candidate set of reachable unified asset indexes. Then, through a second gating process, it merges and mediates intentions only among valid rounds that are not adjacent and contain confirmation or clarification rounds, forming a candidate set of intention stability anchors. Finally, through a third gating process, it establishes conflict blocking edges on activation sensitive items associated with the intention stability anchor candidate set and generates blocking results. The blocking results are written back to the dispute state graph to constrain the recalculation of the first two gating processes until the graph state meets the stability conditions before generating a push access identifier. This can simultaneously reduce the amplification effect of unified asset merging errors, invalid historical contact methods, short-term intention fluctuations, and unclear key items on external pushes by AI customer service. It improves the accuracy of digital asset merging in dispute processing, the accuracy of stable intention identification, and the reliability of push access, and reduces problems such as erroneous asset queuing, accidental activation of short-term consent, and external pushes without clearing of sensitive items. Attached Figure Description

[0048] Figure 1 This is a flowchart of the method of the present invention;

[0049] Figure 2 This is a schematic diagram of the AI-powered customer service call record and result display interface in the digital asset management system of this invention. Detailed Implementation

[0050] 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.

[0051] like Figure 1 As shown, the present invention proposes an AI-assisted method based on all scenarios, comprising:

[0052] Step S1: Obtain the voice interaction data and associated asset records of the dispute resolution object, and construct a dispute status diagram including round nodes, contact information nodes, mediation willingness nodes, and activation of sensitive matters nodes;

[0053] Step S2: Perform the first gating process on the disputed state graph to generate a candidate set of reachable unified asset indexes;

[0054] Step S3: Perform the second gating process to generate a candidate set of intentional stable anchor points based on the candidate set of reachable unified asset index;

[0055] Step S4: Perform the third gating process, establish conflict blocking edges based on the candidate set of intentional stable anchor points, and generate blocking results;

[0056] Step S5: Write the blocking result back to the dispute state graph, constrain the first gating process and the second gating process to recalculate until the number of reachable unified asset indexes, intentional stable anchor points and unresolved conflict blocking edges in the dispute state graph meets the preset conditions. Then, generate a push access identifier and write it into the push task queue. Otherwise, generate a supplementary verification task or a manual review task and output the verification Q&A content.

[0057] Specifically, in step S1, the contact information node of the dispute status diagram records the contact information value, source round identifier, value confidence status and verification status; the mediation willingness node records the willingness category, willingness polarity, source round identifier and merging status; and the sensitive matter activation node records the matter category, matter value, source round identifier and conflict status.

[0058] The edges in the disputed state diagram include round-related edges, unified asset merging edges, intention-anchored edges, and conflict-blocking edges.

[0059] Specifically, in step S2, the first gating process includes:

[0060] The contact information value corresponding to the current round is taken as the first contact information value. The second contact information value is obtained from the SMS receipt result, key input result, historical valid contact information record or existing binding record. A consistency comparison is performed between the first contact information value and the second contact information value.

[0061] When the consistency comparison passes, write the unified asset merging edge into the disputed state graph and generate a candidate set of reachable unified asset indexes.

[0062] When the consistency comparison fails, the corresponding contact information node is marked as pending verification in the dispute status diagram, and a contact information supplementary verification task or contact information confirmation task is generated.

[0063] Specifically, in the first gating process of step S2, the consistency comparison includes at least one of the following: full value consistency comparison of the number, consistency comparison of the preset number of digits, consistency comparison of the identifier, and consistency comparison of the contact information with the existing binding relationship; the blocking result written back by the third gating process is used to delete, freeze, or reduce the validity of the unified asset merging edge associated with the blocking event.

[0064] Specifically, the second gating process in step S3 includes:

[0065] Based on the candidate set of reachable unified asset index, multiple round nodes associated with the same asset node through unified asset merging edges are extracted in the dispute state graph. At least two non-adjacent target round nodes that contain a confirmation round or a clarification round are selected from the multiple round nodes. The mediation intention values ​​associated with the target round nodes are extracted and intention category merging and intention polarity merging are performed.

[0066] When the result of merging the willingness categories and the result of merging the willingness polarities are consistent, write willingness anchor edges between the mediation willingness nodes and generate a candidate set of willingness stable anchor points;

[0067] When any merge result is inconsistent, generate a supplementary verification task and output clarification verification questions and answers.

[0068] Specifically, in step S5, the blocking result written back by the third gating process is used to remove the round nodes associated with the blocking items, freeze the corresponding intention anchor edges, or trigger the recalculation of intention category merging and intention polarity merging;

[0069] Based on the call duration field, call quality score field, and hang-up party label field, the historical round nodes are filtered for valid rounds, and round nodes that meet the preset conditions are marked as confirmation rounds or clarification rounds.

[0070] Specifically, in step S4, the third gating process includes:

[0071] Based on the candidate set of stable anchor points of intent, the active sensitive item nodes associated with the candidate set of stable anchor points of intent are extracted in the dispute state graph, and the differences of active sensitive item nodes belonging to the same item category are compared.

[0072] When the difference comparison results are inconsistent, write conflict blocking edges between the nodes that activate sensitive items, and generate blocking results based on the conflict blocking edges;

[0073] Among them, activating sensitive matters includes at least one of the following: dispute type, demand direction, amount range, processing time, decision-making role, and preferred communication channel.

[0074] The stability conditions include at least two of the following: the reachable unified asset index remains unchanged in two consecutive recalculations, the intended stable anchor remains unchanged in two consecutive recalculations, and the number of unresolved conflict-blocked edges remains unchanged or is less than or equal to a preset threshold in two consecutive recalculations.

[0075] Specifically, in the third gating process of step S4, when the stability condition is not met and the number of recalculations reaches the preset upper limit, a supplementary verification task or a manual review task is generated.

[0076] Specifically, the method also includes generating and writing structured records based on the dispute status diagram. The structured records include fields such as gender, call duration, rating, call quality rating, hang-up party label, intention label, transfer to human agent status, contact information, mediation willingness, activation of sensitive matters, round identifier, and task type.

[0077] The scoring field is generated based on at least two of the following: the first gating processing result, the second gating processing result, the third gating processing result, the blocking result, the number of supplementary verifications, the call quality scoring field, the intention tag field, and the transfer to human agent status field.

[0078] Specifically, the method further includes pre-storing a preset filtering scheme, which is a filtering condition combination template consisting of one or more filtering fields, value conditions corresponding to each filtering field, and logical relationships between multiple value conditions.

[0079] The filtering fields include at least one of the following: gender, call duration, rating, call quality rating, hang-up party label, intention label, transfer to human agent status, gating result, blocking result, and task type.

[0080] In response to the call of the preset filtering scheme, condition matching is performed on the structured record set, the push task queue or the manual review queue, the target record set is output, and the target record set is displayed, distributed to the corresponding queue or exported as a record file.

[0081] It should be noted that, in one specific embodiment of the present invention, the A, B, and C level willingness grades are not pre-assigned, nor are they directly given by single-round speech recognition results or general labeling rules. Instead, they are layered output results after the "three gating processes and their interrelationships" in the present invention are applied to the dispute state diagram. Specifically, the first gating process first generates a candidate set of reachable unified asset indexes around the "unified asset accessibility" of the dispute resolution digital assets, which is used to determine the boundaries of objects participating in the subsequent merging calculation. The second gating process then extracts mediation willingness only within the boundaries of these objects from valid rounds that are not adjacent and contain confirmation or clarification rounds, generating a candidate set of willingness stabilization anchor points. The third gating process further establishes conflict blocking edges only for activation sensitive matters associated with the candidate set of willingness stabilization anchor points and generates blocking results. Subsequently, the blocking results do not stop at the final judgment, but are written back to the dispute state diagram, which inversely constrains the first and second gating processes to recalculate until the number of reachable unified asset indexes, willingness stabilization anchor points, and unresolved conflict blocking edges in the diagram meets the stability condition. Therefore, Level A willingness corresponds to objects where the three gating processes have achieved stable convergence and all push admission conditions are met; Level B willingness corresponds to objects that have passed the first gating process but are still in a pending confirmation state in the second or third gating process, requiring further verification before further judgment; and Level C willingness corresponds to objects that have consistently failed to meet at least one of the three gating processes, thus not entering the push admission chain. In other words, the emergence of Levels A, B, and C is essentially a specific hierarchical implementation of the closed-loop gating mechanism designed by this invention to address the technical problems of "unified asset consolidation error, short-term willingness fluctuations, and extrapolation without clear verification of key matters," rather than a static classification result in a conventional customer tagging system.

[0082] Furthermore, in the present specific embodiment of the present invention, Figure 2The interface displayed for applying the method proposed in this invention within a digital asset management system includes fields such as gender, call duration, score, call quality score, hang-up party tag, intention tag, and transfer to human assistant. These fields not only serve for display but also directly contribute to the improvements made in this invention. Specifically, the gender field is written into the structured record as a basic profile field; the call duration, call quality score, and hang-up party tag fields jointly participate in the effective round filtering and confirmation / clarification round determination; the intention tag field represents the hierarchical results after the convergence of the three gating processes; the score field is generated based on at least two of the following: the first gating result, the second gating result, the third gating result, the blocking result, the number of supplementary verifications, the call quality score field, the intention tag field, and the transfer to human assistant status field. This field is used to sort, filter, and prioritize disputed digital assets; the transfer to human assistant status field represents the execution status of the human review or human takeover task. After these fields are uniformly written into the structured record, they can be further matched according to a preset filtering scheme to form a target record set, which is then distributed to the push task queue, the human review queue, or exported as a result file.

[0083] In one specific embodiment of the present invention, the following description, in conjunction with parameterized examples, further illustrates the preset heterogeneous rebate determination rules, preset confirmation and clarification wheel determination rules, preset intention merging rules, preset conflict blocking edge generation and elimination rules, preset blocking result write-back and recalculation process, scoring field generation rules, and preset screening scheme. The content is used to form the unified implementation basis of Embodiment 1, Embodiment 2, and Embodiment 3.

[0084] In this embodiment, the electronic device sets the following priority for the source of contact information:

[0085] SMS receipt results > Key input results > Existing binding records > Historical valid contact information records > Current round of voice recognition results.

[0086] In this embodiment, the rule for determining off-source rebates preferably includes at least one of the following:

[0087] When the current round of speech recognition results are consistent with all values ​​of the existing bound records, the current round of contact information node is determined to be in a verified state.

[0088] When the current round of speech recognition results are consistent with all the historical valid contact information records, and the SMS receipt shows that the message was successfully delivered, the current round of contact information node is determined to be in a verified state.

[0089] If the last eight digits of the current round of voice recognition result match the last eight digits of the key input result, and there is no conflict with the existing binding records, then the current round of contact information node is determined to be in a verified state.

[0090] When the current round of speech recognition results is consistent with at least two heterogeneous sources, the value corresponding to the higher priority source is given priority as the target value of the current round's connection method node;

[0091] When there is a conflict between different heterogeneous sources, and the high-priority source is inconsistent with the current round of speech recognition results, the current round of contact information node is marked as pending verification, and a contact information supplement verification task or contact information confirmation task is generated.

[0092] In this embodiment, the confirmation wheel determination preferably satisfies at least three of the following conditions:

[0093] The call duration shall be no less than 30 seconds;

[0094] Call quality score no lower than 80 points;

[0095] The hang-up party label indicates whether the user side or the manual side has terminated the call normally, or the system side has terminated the call normally.

[0096] The current round of conversations hits at least one of the following modes: contact information confirmation mode, willingness confirmation mode, or solution confirmation mode;

[0097] The current round does not contain an exception interruption flag.

[0098] In this embodiment, the clarifier preferably satisfies at least three of the following conditions:

[0099] The call duration shall be no less than 20 seconds;

[0100] Call quality score no lower than 70 points;

[0101] At least one of the following: current round of dialogue hits the issue clarification mode, the processing time clarification mode, and the role relationship clarification mode;

[0102] The current round has at least one item whose value is inconsistent with the previous round;

[0103] The current round's disconnection tag is not an abnormal interruption.

[0104] In this embodiment, the call quality score is preferably divided into the following ranges:

[0105] A call quality score of 80 or higher is considered a high-quality round.

[0106] Call quality scores of 60 or higher but lower than 80 are eligible to participate in the verification round.

[0107] A call quality score below 60 indicates a low-quality round.

[0108] In this embodiment, the intention category merging rule can be set as follows:

[0109] "We can continue to communicate", "We can learn about it first", and "We can receive subsequent content" are grouped into the category of "Continue to communicate";

[0110] "Willing to continue mediation", "Can handle", and "Can advance the plan" are grouped into the category of "Continue mediation";

[0111] "We'll talk about it later," "We'll contact you later," and "Not processing for the time being" are grouped into the category of "temporarily suspended processing."

[0112] "Not needed", "Do not contact", and "Refuse to process" are grouped into the category of "Refusal to communicate".

[0113] In this embodiment, the intention polarity merging rule can be set as follows:

[0114] Expressions containing explicit acceptance or explicit agreement are grouped into the affirmative polarity;

[0115] Expressions containing explicit rejection or explicit negation are grouped into the negative polarity;

[0116] Statements containing terms that are pending confirmation, vaguely accepted, or conditionally accepted are grouped into the polarity of pending confirmation.

[0117] In this embodiment, the electronic device generates a stable anchor point only when the following conditions are met simultaneously:

[0118] At least two non-adjacent target rounds have completed the merging of intention categories;

[0119] The category merging results for the target rounds are consistent;

[0120] The polarity merging results for the target rounds are consistent;

[0121] At least one of the target rounds is a confirmation round or at least one is a clarification round.

[0122] If only the categories are consistent but the polarities are inconsistent, then a stable anchor point of intent will not be generated, but a supplementary verification task of intent will be generated instead.

[0123] In this embodiment, the preferred criteria for activating sensitive matters include: dispute type, claim direction, amount range, processing time, decision-making role, and preferred communication channel.

[0124] For the above-mentioned issues, it is preferable to perform difference comparisons separately for electronic devices by category.

[0125] For example:

[0126] If the same object is processed at "contact within 24 hours" and "contact later" in two valid rounds respectively, then a conflict blocking edge is established for the processing time category;

[0127] When the amount falls into two different preset ranges, "below 1000 yuan" and "above 5000 yuan", a conflict blocking edge is established for the amount range category.

[0128] When the preferred communication channels are "receive push notifications" and "human contact only", establish a conflict blocking edge for the preferred communication channel category.

[0129] In this embodiment, the conflict blocking edge elimination rule preferably includes at least one of the following:

[0130] Subsequent clarification rounds provide clear and consistent values ​​for the corresponding matters;

[0131] At least two confirmation rounds must maintain consistent values ​​for the same item;

[0132] Based on the priority rules, the high-confidence value is determined to be the final valid value;

[0133] The results of manual review are written back as the final valid values.

[0134] If any of the above conditions are met, the corresponding conflict-blocking edge can be deleted or marked as eliminated.

[0135] In this embodiment, after the third gating process generates the blocking result, the following write-back and recalculation process is preferably executed:

[0136] Write the blocking result to the blocking status field of the corresponding activated sensitive item node;

[0137] If the blocking event is related to the object merging boundary, then delete, freeze, or reduce the validity of the unified asset merging boundary associated with it;

[0138] If the blocking matter is related to the boundary of intent confirmation, then freeze the corresponding intent anchor edge, or remove the round node associated with the blocking matter;

[0139] After the write-back is completed, the second gating process is triggered first to handle local recalculation;

[0140] If the blockage cannot be eliminated after local recalculation, then a joint recalculation of the first gating process and the second gating process is triggered.

[0141] When the reachable unified asset index remains unchanged, the intention stable anchor point remains unchanged, and the number of unresolved conflict blocking edges remains unchanged or is less than or equal to a preset threshold in two consecutive recalculations, the disputed state graph is determined to have reached the stability condition.

[0142] In this embodiment, the maximum number of recalculations is preferably 2 to 3 times; the preset threshold is preferably 0 or 1.

[0143] When the maximum number of recalculations is reached and the stability condition is still not met, the electronic device generates a supplementary verification task or a manual review task.

[0144] It should be noted that the above parameter values ​​are only a preferred embodiment of the present invention, intended to enable those skilled in the art to implement the present invention, and do not constitute a limitation on the scope of protection of the present invention.

[0145] In one specific embodiment of the present invention, the scoring field is not simply based on the displayed value of the call result, but rather a comprehensive evaluation value used to characterize the current processing status of the dispute resolution digital asset. The scoring field can be calculated by at least two of the following: first gating processing result, second gating processing result, third gating processing result, blocking result, number of supplementary verifications, call quality scoring field, intention tag field, and transfer to human agent status field, according to preset weights.

[0146] For example, the following internal comprehensive evaluation rules may be adopted:

[0147] First gate control processing passed: +40 points;

[0148] The second gating process establishes a clear and stable anchor point of intent: +40 points;

[0149] The second gating process forms a stable anchor point for the unconfirmed intention: +20 points;

[0150] Third gate control handling of unresolved conflict-blocked edges: +40 points;

[0151] The third gating system has unresolved conflict blocking edges: -30 points;

[0152] Call quality score above 80: +20 points;

[0153] Call quality score below 60: 0 points;

[0154] For each additional verification attempt: -10 points;

[0155] Is the transfer to human operator status not available? +10 points;

[0156] The status of being transferred to a human operator is: 0 points.

[0157] Following the example above:

[0158] If the object passes the first gate control process, forms a clear and stable anchor point in the second gate control process, has no unresolved conflict blocking edges in the third gate control process, and has a high call quality score and is not transferred to a human operator, then a high internal evaluation value can be formed and mapped to a high interface display value.

[0159] If an object passes the first gating process, but the second gating process only forms an anchor point for unconfirmed intention and the third gating process has unresolved conflict blocking edges, then the internal evaluation value can be in the low to medium range, and the value displayed on the interface remains in the unadmitted state.

[0160] If an object continues to be blocked during the three-gating process and enters manual review, the internal evaluation value will be in the low range, and the value displayed on the interface will remain in the "not allowed" state.

[0161] Therefore, the scoring field becomes a comprehensive field reflecting the combined effect of the three gating processes, rather than a display field independent of the core technical solution.

[0162] In one specific embodiment of the present invention, the electronic device pre-stores multiple preset filtering schemes. Each preset filtering scheme consists of filtering fields, value conditions, and logical relationships, and is used to extract the target record set from a structured record set, a push task queue, or a manual review queue.

[0163] Example 1: Filtering Scheme for Category A Objects

[0164] Preferred filter fields and conditions include:

[0165] Intent tag field = Grade A;

[0166] First gate processing result = Pass;

[0167] The result of the second gating process is the formation of a clear and stable anchor point of intent.

[0168] Third gating result = No unresolved conflict blocking edges;

[0169] Call quality score ≥ 80;

[0170] Transfer to human operator status field = No.

[0171] After the electronic device responds to the above filtering scheme call, it outputs the records that meet the conditions as a set of A-level target records and writes them into the push task queue.

[0172] Example 2: Supplementary Verification and Screening Scheme for Level B Objects

[0173] Preferred filter fields and conditions include:

[0174] Intent tag field = Grade B;

[0175] First gate processing result = Pass;

[0176] The result of the second gating process = the formation of an anchor point for unconfirmed intentions;

[0177] The third gating result indicates that there are unresolved conflict-blocked edges.

[0178] Call quality score ≥ 60;

[0179] Transfer to human operator status field = No.

[0180] After responding to the above screening scheme call, the electronic device outputs the records that meet the conditions as a set of supplementary verification records of level B, and distributes them to the supplementary verification task queue or the secondary follow-up queue.

[0181] Example 3: Manual review and screening scheme for C-level objects

[0182] Preferred filter fields and conditions include:

[0183] Intent tag field = Grade C;

[0184] The first gated processing result is either unstable or pending verification.

[0185] The second gating result = no intentional stable anchor point was formed;

[0186] The third gating result = there is a blocking result;

[0187] Call quality score field <60 or transfer to human operator status field = Yes.

[0188] After responding to the above screening scheme call, the electronic device outputs the records that meet the conditions as a set of C-level manual review records and distributes them to the manual review queue or the non-push processing set.

[0189] It should be noted that the above-mentioned A-level object push screening scheme, B-level object supplementary verification screening scheme, and C-level object manual review screening scheme are only specific implementation examples of the present invention, used to illustrate the automatic screening and distribution methods of objects with different willingness levels in the present invention, and do not constitute a limitation on the scope of protection of the present invention.

[0190] Example 1

[0191] like Figure 2 As shown, taking the object with number 178****0568 in the figure as an example, this illustrates the implementation method of obtaining the A-level willingness level according to the technical solution of the present invention.

[0192] In this embodiment, the dispute resolution digital asset is the object with the number 178****0568 in the figure. Its current interface fields are: gender is male, call duration is 00:40, rating is 200, call quality is high, the hanging party is AI, the tag is intentional, and the intention tag is A-level (has clear intention).

[0193] In this embodiment, the preset heterogeneous rebate determination rule is set as follows: when the contact information obtained by the current round of voice recognition is completely consistent with the historical valid contact information record, and is consistent with the last eight digits of the contact information in the existing binding record, the contact information node is determined to have passed the verification; the preset confirmation round determination rule is set as follows: the call duration is not less than 30 seconds, the call quality score is not less than 80 points, the hang-up party's tag is not an abnormal interruption, and the current round hits the "confirmation of continued communication" or "confirmation of plan sending" dialogue mode; the preset conflict blocking edge difference condition is set as follows: when the values ​​of the same item category are inconsistent in time, amount, role or channel and are not confirmed and eliminated in subsequent rounds, a conflict blocking edge is established; the preset stability condition is set as follows: the reachable unified asset index remains unchanged after two consecutive recalculations, the intention stability anchor point remains unchanged after two consecutive recalculations, and the number of unresolved conflict blocking edges remains 0 after two consecutive recalculations.

[0194] Step S1: The electronic device obtains the current round of voice interaction data of the object and its associated asset records, establishes the current round node in the dispute state diagram, and writes the contact information node, mediation willingness node and activation sensitive matter node extracted in the current round; for example, in the current round, it is recognized that the current object clearly indicates "can continue communication and accept the subsequent plan to be sent", and at the same time, the valid historical records of the previous two rounds are read in the associated asset records.

[0195] Step S2: Perform the first gating process. The electronic device performs heterogeneous verification between the contact information identified in the current round and the historical valid contact information records, existing binding records, and SMS receipt results. In this embodiment, the current round contact information is consistent with the historical valid contact information records and also consistent with the existing binding records. Therefore, a unified asset merging edge is written into the disputed state graph to generate a candidate set of reachable unified asset indexes for the object.

[0196] For example, if the contact information identified in the current round is "178****0568", the corresponding field in the historical valid contact information record is "178****0568", and the last eight digits of the corresponding contact information in the existing binding record are "0568", and the SMS receipt result shows that the confirmation SMS for the current round has been delivered, then the first gating process writes the "verification passed" status to the contact information node of the current round and writes a unified asset merging edge in the disputed state graph, so that the node of this round establishes a unique mapping relationship with the same asset node.

[0197] Step S3: Perform the second gating process. Based on the aforementioned candidate set of reachable unified asset indexes, the electronic device filters at least two non-adjacent target round nodes from the current round and historical rounds, with a confirmation round or clarification round between them. In this embodiment, the previous round was a willingness confirmation round, and the current round is a subsequent confirmation round. Both rounds of mediation willingness are categorized into the "continue communication" category and are both positive polarity. Therefore, willingness anchor edges are written between the mediation willingness nodes to generate a candidate set of willingness stable anchor points.

[0198] For example, in the three most recent valid historical records of the current object, the electronic device selects the first round and the current round as non-adjacent target round nodes. The first round has a call duration of 00:52, a call quality score of 86, and the end of the call is marked as completed by the human side after the disconnection. The current round has a call duration of 00:40, a call quality score of 92, and the end of the call is marked as completed by the AI ​​after the solution is confirmed. The mediation intentions in the two rounds are mapped to "continue communication / affirmative polarity" and "continue communication / affirmative polarity" respectively. Therefore, intention anchor edges are written between the corresponding mediation intention nodes, and the candidate set of stable intention anchors for the current object is marked as positively stable.

[0199] Step S4: Perform the third gating process. Based on the candidate set of stable anchor points of intent, the electronic device extracts the associated active sensitive matter nodes and performs a difference comparison on the dispute type, demand direction, amount range, and preferred communication channel. In this embodiment, the values ​​of each matter are consistent before and after, and no unresolved conflict blocking edge is formed. Therefore, the generated blocking result is empty or in a pass state.

[0200] For example, the comparison results of the differences in dispute type, demand direction, amount range, and preferred communication channel are as follows: the dispute type "after-sales negotiation" is consistent with the previous round, the demand direction "continue processing" is consistent with the previous round, the amount range is all below 5,000 yuan, and the preferred communication channel is "can receive subsequent pushes". Therefore, no unresolved conflict blocking edge was established, and the blocking result is empty.

[0201] In step S5, after the electronic device writes the third gating processing result back to the disputed state graph, it does not trigger the substantial correction of the first two gating processes. During continuous recalculation, the reachable unified asset index remains unchanged, the intention stability anchor point remains unchanged, and the number of unresolved conflict blocking edges remains 0. Therefore, the stability condition is met, and a push admission identifier is generated and written to the push task queue. At the same time, the A-level intention tag, high score, and no transfer to human status of the object are retained in the structured record.

[0202] In this embodiment, Figure 2The displayed score of 200 represents a specific implementation result of the scoring field in this invention. The current score is generated by the first gating processing result, the second gating processing result, the third gating processing result, the call quality score field, and the intention tag field according to preset weights. For example, passing the first gating processing is recorded as 60 points, forming a stable anchor point in the second gating processing is recorded as 60 points, having no unresolved conflict blocking edges in the third gating processing is recorded as 40 points, a high call quality score is added as 20 points, and no transfer to human agent status is added as 20 points, thus forming a high-priority record. At the same time, the current object is matched by a preset screening scheme, for example, the screening conditions are "intention tag = A level, call quality score ≥ 80, transfer to human agent status = no, third gating processing result = passed". The electronic device automatically distributes the current object to the push task queue accordingly. This demonstrates that the score of this invention does not come from the high or low of a single field, but rather from the mutual inheritance relationship between the first gating processing locking the correct object boundary, the second gating processing forming a stable intention, and the third gating processing finally completing the clearance judgment and reverse verifying the results of the first two.

[0203] Therefore, in this embodiment, the object with the number 178****0568 in the figure is stably identified as a Level A willing customer after undergoing three gating processes, making it suitable as a digital asset for dispute resolution that can be directly pushed into the subsequent push link.

[0204] Example 2

[0205] by Figure 2 Taking the object with the number 139****9385 as an example, this paper illustrates the implementation method of obtaining the B-level willingness level according to the technical solution of the present invention.

[0206] In this embodiment, the preset confirmation round judgment rule still adopts the combination of "call duration not less than 30 seconds, call quality score reaching a preset threshold and hitting the confirmation or clarification dialogue mode"; however, since the current object has a low call quality score in the current round, the electronic device does not directly use the current round as a strong confirmation round in the second gating process, but instead uses it together with a round of historical intention confirmation round and a round of historical matter clarification round to participate in intention merging and matter comparison, thereby avoiding the excessive amplification of stable intention judgment by a single round of low-quality voice.

[0207] In this embodiment, the dispute resolution digital asset is the object with the number 139****9385 in the figure. Its current interface fields are: gender is male, call duration is 01:01, rating is 0, call quality is low, the hang-up party is AI, the tag is ordinary hang-up, and the intention tag is B (may have intention).

[0208] Step S1: The electronic device obtains the current round of voice interaction data of the object and its associated asset records, establishes the current round node in the dispute status diagram, and writes the contact information node, mediation willingness node and activation sensitive matter node extracted in the current round; for example, in the current round, it is identified that the object does not directly refuse to continue communication, but expresses "I can understand first, and I need to confirm whether to continue processing", and at the same time, the system extracts the historical rounds in the previous two rounds where there was one confirmation of willingness and one clarification of matter from the associated asset records.

[0209] Step S2: Perform the first gating process. The electronic device performs heterogeneous verification by comparing the current round of contact information with historical valid contact information records and existing binding records; in this embodiment, the current round of contact information is consistent with the historical valid contact information, thus generating a candidate set of reachable unified asset indexes.

[0210] For example, the contact information identified in the current round is "139****9385", and the corresponding field in the historical valid contact information record is also "139****9385". The last eight digits of the existing binding record match it, so the first gate control process passes and a unique reachable unified asset index is generated. However, since the call quality score in the current round is only 58 points, which is lower than the preferred threshold for A-level objects, although the current round can participate in object merging, it is given a lower confidence weight in subsequent intention merging.

[0211] Step S3: Perform the second gating process. Based on the candidate set of the reachable unified asset index, the electronic device filters valid rounds from the current round and historical rounds, and performs category merging and polarity merging on the mediation intentions. In this embodiment, multiple rounds of intentions are not merged into explicit rejection, but into categories related to "continue communication pending confirmation" or "postponement of processing", and the polarity is expressed as pending confirmation polarity. Therefore, a candidate set of stable anchor points of intentions is formed, but this anchor point is a non-explicit positive anchor point.

[0212] For example, the electronic device selects rounds 1, 3, and the current round to form a merge analysis set. Round 1 has a call duration of 00:47, a call quality score of 81, and hits the "continue communication for confirmation" phrase; Round 3 has a call duration of 00:36, a call quality score of 76, and hits the "contact later for clarification" phrase; the current round has a call duration of 01:01, a call quality score of 58, and the call ended using AI's normal termination method. The three rounds of mediation intention mapping results are "continue communication / pending confirmation polarity," "delayed processing / pending confirmation polarity," and "continue communication pending confirmation / pending confirmation polarity," thus grouping them at the category level into "continued communication but pending confirmation" and at the polarity level into "pending confirmation polarity," thereby forming only a candidate set of non-clearly positive intention stable anchor points.

[0213] Step S4: Perform the third gating process. The electronic device further extracts the active sensitive matter nodes associated with the candidate set of stable anchor points of the intention, and performs a difference comparison on the dispute type, claim direction, amount range, and processing time. In this embodiment, the dispute type and claim direction are consistent, but there is an unresolved conflict blocking edge at the processing time. That is, the current round indicates "contact later", while the previous round indicated "can be processed as soon as possible". Therefore, the third gating process generates a blocking result and writes it back to the dispute state graph.

[0214] For example, in the category of processing time items, the first round record is "can continue to be processed within 24 hours", while the current round record is "need to be contacted later". The two are inconsistent in terms of time values, exceeding the preset difference condition. Therefore, the electronic device establishes a conflict blocking edge between the corresponding activated sensitive item nodes and generates a blocking result accordingly. After the current blocking result is written back, the unified asset merging edge formed by the first gating process is not deleted, but the validity of the intention anchoring edge related to the processing time is frozen, so that the second gating process still remains in the pending confirmation state when recalculated.

[0215] In step S5, the electronic device triggers a recalculation of the first two gating processes based on the blocking result. After recalculation, the reachable unified asset index remains stable, but the intention stability anchor point remains in a pending confirmation state, and the number of conflict-free blocking edges has not decreased to zero. Therefore, no push access identifier is generated; instead, a supplementary verification task is generated, and clarification verification questions and answers regarding "processing time" and "willingness to continue communication" are output. Since the target did not show explicit refusal, and the contact information is validly merged, and the intention still retains the possibility of subsequent conversion, the target is retained as a Level B intention customer in the structured record.

[0216] In this embodiment, Figure 2 A score of 0 displayed on the interface does not mean the current object has no processing value, but rather that it has not yet met the scoring criteria for direct entry into the push task queue in the current structured record. For example, the scoring fields can be set as follows: 40 points for passing the first gated processing, 20 points for forming a pending confirmation intention anchor point in the second gated processing, 40 points deducted for the existence of unresolved conflict blocking edges in the third gated processing, no points added for low call quality scores, and no points added for normal hang-ups, thus making the overall score lower than the direct push threshold. At the same time, the preset filtering scheme can be set as "Intention tag = B level, first gated processing result = passed, second gated processing result = pending confirmation, third gated processing result = blocked result exists, transfer to human agent status = no". Based on this, the electronic device will not directly import the current object into the push task queue, but will distribute it to the supplementary verification queue or the secondary follow-up queue. This shows that B-level objects are not simply stratified between A and C based on experience, but are determined by a specific combination of states in the three gated processing: "object has been correctly merged - intention still needs confirmation - matter still has partial conflicts".

[0217] Therefore, in this embodiment, the object with the number 139****9385 in the figure is identified as a B-level customer when the first gate control process passes, the second gate control process forms a stable anchor point for the intention to be confirmed, and the third gate control process discovers some uncleared matters. The processing result is not directly pushed, but enters the supplementary verification or secondary follow-up link.

[0218] Example 3

[0219] by Figure 2 Taking the object with the number 181****8933 as an example, this paper illustrates the implementation method of obtaining the C-level willingness level according to the technical solution of the present invention.

[0220] In this embodiment, the electronic device's determination of a customer's willingness level (C-level) is not simply based on the current round's "lack of willingness," but rather on the fact that the current object continuously fails to meet stable admission conditions throughout the interconnected chain of the first, second, and third gating processes. In other words, the C-level result is not a direct output of a single label rule, but rather a convergence result of the closed-loop gating mechanism's comprehensive effect on object boundaries, willingness stability, and the event's clearance status.

[0221] In this embodiment, the dispute resolution digital asset is the object with the number 181****8933 in the figure. Its current interface fields are: gender is male, call duration is 00:43, rating is 0, call quality is low, the hang-up party is AI, and the intention tag is C level (no intention).

[0222] Step S1: The electronic device acquires the current round of voice interaction data of the object and its associated asset records, establishes the current round node in the dispute state diagram, and writes the contact information node, mediation willingness node, and activation of sensitive matters node extracted in the current round; for example, the current round identification results show that the object does not show any willingness to continue communication, and the associated asset records can be extracted to show that there was a record of a quick end to the call and a record of an identity not clearly confirmed in the previous round.

[0223] Step S2: Perform the first gating process. The electronic device performs heterogeneous verification by comparing the current round of contact information with historical valid contact information records, key input results, and existing binding records. In this embodiment, although the current round of contact information partially matches the historical records, due to low call quality and insufficient valid confirmation, the contact information node only forms a low-confidence verification result, and the generated candidate set of reachable unified asset index is unstable.

[0224] For example, if the contact information identified in the current round is "181****8933", there are two other contact information records with different last four digits, "181****8933" and another record with different last four digits. The existing binding records failed to provide stable binding results, and the call quality score in the current round is only 54 points. Therefore, although the first gating process can form a candidate contact information match, it cannot generate a high-confidence reachable unified asset index and only writes the corresponding contact information node into the low-confidence verification state.

[0225] Step S3: Perform the second gating process. The electronic device filters valid rounds based on the unstable reachable unified asset index candidate set and attempts to perform category merging and polarity merging on the mediation intentions. In this embodiment, one round in the current round and the previous rounds showed a refusal to communicate, and the other round showed an unconfirmed identity. Neither a positive stable anchor point nor a conversion anchor point can be formed. Therefore, the second gating process does not form a valid candidate set of stable anchor points for intentions.

[0226] For example, an electronic device extracts two rounds of records from historical rounds. One round has a call duration of 00:18, with the call ending quickly on the user's side; the other round has a call duration of 00:43, a call quality score of 60, and the script only matches the "identity not confirmed" mode. The current round is identified as "refusal to continue communication." In this case, the electronic device cannot simultaneously meet the preferred condition of "at least two non-adjacent valid rounds with a confirmation or clarification round in between." Even if it barely participates in the merging, the resulting intention categories are mapped to "refusal to communicate," "pending confirmation," and "refusal to communicate," respectively. The polarity exhibits a coexistence of "negative polarity" and "pending confirmation polarity." Therefore, the second gating process cannot form an effective stable anchor point candidate set.

[0227] Step S4: Perform the third gating process. Due to the lack or instability of the candidate set of stable anchor points, the electronic device can only perform partial checks on the active sensitive item nodes associated with the current round. In this embodiment, there are unresolved conflicts in the dispute type and processing time, and there is a lack of sufficient clarification rounds for item clearance. Therefore, the third gating process generates a blocking result and writes the blocking result back to the dispute state graph.

[0228] For example, a comparison of the activated sensitive items in the current round and historical rounds reveals that there is a historical conflict between the dispute type "after-sales dispute" and "normal follow-up visit", and a conflict between the processing time "no need to continue contact" and "contact later". Furthermore, due to the lack of sufficient confirmation rounds or clarification rounds, the above conflicts were not eliminated in subsequent rounds. Therefore, the third gating process establishes conflict blocking edges on the two item categories respectively and generates blocking results.

[0229] In step S5, after the electronic device performs recalculation based on the blocking result, it still cannot simultaneously meet the stable conditions of "the reachable unified asset index is valid, the intention stable anchor point is established, and the number of unresolved conflict blocking edges is not greater than the threshold". Therefore, it does not generate a push admission identifier, but instead writes the object into the manual review task or non-push processing set, and retains its low score, low call quality, no clear positive intention and C-level intention label in the structured record.

[0230] In this embodiment, after the blocking result is written back, the electronic device triggers the first and second gating processes for recalculation. However, in two consecutive recalculations, the reachable unified asset index still cannot remain stable, the intention stability anchor point is never formed, and the number of unresolved conflict blocking edges continues to exceed the preset threshold. Therefore, after reaching the recalculation limit, the electronic device writes the current object into the manual review task queue or the non-push processing set, and retains its low call quality score, C-level intention tag, and transfer to human agent status field in the structured record. This embodiment can be hit by the preset filtering scheme of "call quality score < 60, intention tag = C-level, transfer to human agent status = yes or waiting for human agent, third gating result = blocking", thus automatically entering the manual review queue instead of the push task queue. This further demonstrates that the formation of C-level objects in this invention does not originate from a single round of negative expression, but is caused by the instability of the merging of preceding objects, the failure of subsequent intention merging, and the continuous failure to clear key conflicts in the three gating processes. This hierarchical result, achieved by the three gating processes and their mutual inheritance relationship, is different from the conventional method of direct diversion based on single-round tags in the prior art.

[0231] Therefore, in this embodiment, the object with the number 181****8933 in the figure is identified as a C-level customer due to unstable contact information verification, failure to merge intentions, and failure to clear key matters. The result is that it will not enter the subsequent push link, but will be transferred to manual review or terminated follow-up link.

[0232] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0233] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A full-scenario AI-assisted method, characterized in that, include: Step S1: Obtain the voice interaction data and associated asset records of the dispute resolution object, and construct a dispute status diagram including round nodes, contact information nodes, mediation willingness nodes, and activation of sensitive matters nodes; Step S2: Perform the first gating process on the disputed state graph to generate a candidate set of reachable unified asset indexes; Step S3: Perform the second gating process to generate a candidate set of intentional stable anchor points based on the candidate set of reachable unified asset index; Step S4: Perform the third gating process, establish conflict blocking edges based on the candidate set of intentional stable anchor points, and generate blocking results; Step S5: Write the blocking result back to the dispute state graph, constrain the first gating process and the second gating process to recalculate until the number of reachable unified asset indexes, intentional stable anchor points and unresolved conflict blocking edges in the dispute state graph meets the preset conditions. Then, generate a push access identifier and write it into the push task queue. Otherwise, generate a supplementary verification task or a manual review task and output the verification Q&A content.

2. The AI-assisted method based on all scenarios according to claim 1, characterized in that, In step S1, the contact information node of the dispute status diagram records the contact information value, source round identifier, value confidence status and verification status; the mediation willingness node records the willingness category, willingness polarity, source round identifier and merging status; and the activation sensitive matter node records the matter category, matter value, source round identifier and conflict status. The edges in the disputed state diagram include round-related edges, unified asset merging edges, intention-anchored edges, and conflict-blocking edges.

3. The AI-assisted method based on all scenarios according to claim 1, characterized in that, In step S2, the first gating process includes: The contact information value corresponding to the current round is taken as the first contact information value. The second contact information value is obtained from the SMS receipt result, key input result, historical valid contact information record or existing binding record. A consistency comparison is performed between the first contact information value and the second contact information value. When the consistency comparison passes, write the unified asset merging edge into the disputed state graph and generate a candidate set of reachable unified asset indexes. When the consistency comparison fails, the corresponding contact information node is marked as pending verification in the dispute status diagram, and a contact information supplementary verification task or contact information confirmation task is generated.

4. The AI-assisted method based on all scenarios according to claim 3, characterized in that, In the first gating process of step S2, the consistency comparison includes at least one of the following: full value consistency comparison of the number, consistency comparison of the preset number digits, consistency comparison of the identifier, and consistency comparison of the contact information with the existing binding relationship; the blocking result written back by the third gating process is used to delete, freeze, or reduce the validity of the unified asset merging edge associated with the blocking event.

5. The AI-assisted method based on all scenarios according to claim 1, characterized in that, The second gating process in step S3 includes: Based on the candidate set of reachable unified asset index, multiple round nodes associated with the same asset node through unified asset merging edges are extracted in the dispute state graph. At least two non-adjacent target round nodes that contain a confirmation round or a clarification round are selected from the multiple round nodes. The mediation intention values ​​associated with the target round nodes are extracted and intention category merging and intention polarity merging are performed. When the result of merging the willingness categories and the result of merging the willingness polarities are consistent, write willingness anchor edges between the mediation willingness nodes and generate a candidate set of willingness stable anchor points; When any merge result is inconsistent, generate a supplementary verification task and output clarification verification questions and answers.

6. The AI-assisted method based on all scenarios according to claim 5, characterized in that, In step S5, the blocking results written back by the third gating process are used to remove round nodes associated with the blocking items, freeze the corresponding intention anchor edges, or trigger the recalculation of intention category merging and intention polarity merging. Based on the call duration field, call quality score field, and hang-up party label field, the historical round nodes are filtered for valid rounds, and round nodes that meet the preset conditions are marked as confirmation rounds or clarification rounds.

7. The AI-assisted method based on all scenarios according to claim 1, characterized in that, In step S4, the third gating process includes: Based on the candidate set of stable anchor points of intent, the active sensitive item nodes associated with the candidate set of stable anchor points of intent are extracted in the dispute state graph, and the differences of active sensitive item nodes belonging to the same item category are compared. When the difference comparison results are inconsistent, write conflict blocking edges between the nodes that activate sensitive items, and generate blocking results based on the conflict blocking edges; Among them, activating sensitive matters includes at least one of the following: dispute type, demand direction, amount range, processing time, decision-making role, and preferred communication channel.

8. The stability conditions include at least two of the following: the reachable unified asset index remains unchanged in two consecutive recalculations, the intended stable anchor remains unchanged in two consecutive recalculations, and the number of unresolved conflict-blocked edges remains unchanged or is less than or equal to a preset threshold in two consecutive recalculations.

9. The AI-assisted method based on all scenarios according to claim 7, characterized in that, In the third gating process of step S4, when the stability condition is not met and the number of recalculations reaches the preset upper limit, a supplementary verification task or a manual review task is generated.

10. The AI-assisted method based on all scenarios according to claim 1, characterized in that, The method also includes generating and writing structured records based on the dispute status diagram. The structured records include fields such as gender, call duration, rating, call quality rating, hang-up party label, intention label, transfer to human agent status, contact information, mediation willingness, activation of sensitive matters, round identifier, and task type. The scoring field is generated based on at least two of the following: the first gating processing result, the second gating processing result, the third gating processing result, the blocking result, the number of supplementary verifications, the call quality scoring field, the intention tag field, and the transfer to human agent status field.

11. The AI-assisted method based on all scenarios according to claim 9, characterized in that, The method also includes pre-storing a preset filtering scheme, which is a filtering condition combination template consisting of one or more filtering fields, the value conditions corresponding to each filtering field, and the logical relationship between multiple value conditions. The filtering fields include at least one of the following: gender, call duration, rating, call quality rating, hang-up party label, intention label, transfer to human agent status, gating result, blocking result, and task type. In response to the call of the preset filtering scheme, condition matching is performed on the structured record set, the push task queue or the manual review queue, the target record set is output, and the target record set is displayed, distributed to the corresponding queue or exported as a record file.

Images