Method and system for carrying out obligation tracking and risk management and control on long-period dialogue based on speech contract object
By using a speech contract extraction module and state transition engine based on the Transformer architecture, unstructured dialogues are instantiated as speech contract objects (VCOs), which solves the problems of fragmented dialogue information and lagging risk management in existing technologies, and realizes automated obligation tracking and proactive risk warning.
Patent Information
- Application Number
- CN202510969184.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-15
- Publication Date
- 2025-10-17
AI Technical Summary
Existing technologies are unable to automatically identify, track, and manage commitment information in long-term unstructured conversations, resulting in information fragmentation, passive and lagging risk management, and the inability to achieve forward-looking early warning.
A verbal contract extraction module based on the Transformer architecture is adopted to instantiate unstructured dialogues into verbal contract objects (VCOs). Automated obligation tracking and risk management are achieved through a state transition engine and business action triggers, combined with data security and privacy protection measures.
It enables automated management and risk warning of commitment information in long-term dialogues, improving the certainty and reliability of business communication and reducing information loss and risk lag.
Smart Images

Figure CN120806646A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present invention relates to the field of computer information processing, and in particular to an artificial intelligence application that combines natural language processing, business process management and risk control. More specifically, the present invention relates to a method and system for systematically identifying, tracking, managing and executing implicit commitments and obligations arising from long-cycle, unstructured human conversations, and contains specific technical considerations for data security and personal information protection. BACKGROUND
[0002] In many modern business fields such as finance, insurance, legal consulting, large customer sales, project management, communication with customers or partners is the core link of business promotion. These communications often last for a long time, cross channels, and are mainly in the form of unstructured natural language conversations (including text chats, emails, and transcribed texts of phone recordings). Among these massive conversation data, there are often various commitments, agreements and to-do items made by one party to another. These "oral commitments" have de facto binding force in business activities and are directly related to customer satisfaction, smooth transaction and corporate reputation.
[0003] However, there is a huge, universally overlooked technical gap in the existing technology when dealing with this type of information, which leads to a technical pain point that is widespread in existing business systems and practices and needs to be addressed urgently:
[0004] First, the fragmentation and management gap of information. In the existing technology, systems that can handle conversations or tasks are mainly divided into several categories, but they are each their own and cannot form effective linkage. Customer Relationship Management (CRM) systems or task management software can allow business personnel to manually create customer follow-up tasks or to-do items, but the generation of all tasks depends on human initiative to enter, and cannot automatically discover and capture commitment information from massive, real-time conversations. On the other hand, there are also some conversation analysis tools or call quality inspection software on the market that can extract keywords or conduct sentiment analysis on transcribed text, but the purpose of these tools is post-analysis, and their technical core is statistics and tagging, completely lacking the ability to identify, track and manage the semantics of "obligations" which have legal and business logic attributes. This creates a huge gap between "conversation flow" and "task flow", and oral commitments are largely lost in this gap.
[0005] Second, the lack of automated semantic understanding and tracking mechanism. Although existing natural language processing (NLP) technologies have developed in general tasks such as intent recognition and entity extraction, they are usually domain general and are not specifically designed and applied to the highly complex technical problem of "automatically identifying commitment speech with future performance obligations". Therefore, the existing technology cannot automatically convert unstructured commitments into a machine-traceable and manageable state object, and the entire process still remains manual.
[0006] Third, the passivity and lag of risk management. Due to the above defects, the management of oral commitment risks under the existing technology is completely passive. Enterprises can only discover the existence of problems when customers complain or business is interrupted, and cannot perform any form of forward-looking risk management.
[0007] In summary, the point of difference of the present application from the prior art is that it does not simply piece together several existing technologies, but rather, it creatively proposes a completely new core concept - "Verbal Contract Object (VCO)", and around the "automated instantiation" and "full life cycle closed-loop management" of this object, it builds a complete, end-to-end solution. It first bridges the gap between "dialogue understanding" and "business process risk management", and improves risk management from post-event, passive remediation to in-event, automated tracking and pre-event, forward-looking early warning, which is an ability and technical concept that the existing technology does not have. SUMMARY
[0008] The purpose of the present application is to overcome the fundamental defects of the background art, such as the loss of oral commitment information, the difficulty of tracking and management, the lag of risk, and the low efficiency of traceability, and to provide a completely new, automated method and system for obligation tracking and risk control of long-period dialogues based on a "Verbal Contract Object (VCO)" created by the present application.
[0009] To achieve this core purpose, the present application provides a method, the core technical solution of which is as follows:
[0010] First, the system collects the unstructured dialogue content from one or more sources through a unified dialogue interface in a safe and compliant manner.
[0011] Next, the system calls a dedicated speech contract extraction module. The "speech contract extraction module" referred to here specifically refers to a deep learning model based on natural language understanding. In a preferred embodiment of the present application, the model can be a sequence labeling and relation extraction joint model based on the Transformer architecture (such as BERT) and fine-tuned for commitment speech recognition tasks. Its training corpus needs to contain a large number of artificially or semi-automatically annotated samples from real business conversations, including commitment intent, obligation subject, right subject, specific content of commitment, and performance deadline, etc. entities and relationships. The workflow of this module is: perform word segmentation and vectorization on the input dialogue sentence, then capture the context semantics through the self-attention mechanism of the model, and finally output the entity label of each word element and the relationship classification between entities, thereby realizing the automatic and structured extraction of commitment information.
[0012] After identifying the commitment speech segment, the module immediately performs an "instantiation" operation to convert it into a standardized speech contract object (VCO).
[0013] All successfully instantiated VCOs are securely stored in a dedicated, persistent database, which the present application refers to as the "dynamic obligation ledger". In a preferred embodiment of the present application, the "dynamic obligation ledger" can be implemented using a relational database or a document database. To ensure data security, all data stored in the ledger should be stored using field-level encryption, and the database should have complete audit log functionality enabled.
[0014] To realize dynamic tracking of the commitment fulfillment process, the present application designs a VCO state transition engine. The "state transition engine" referred to here specifically refers to a background service process based on configurable rules. The engine maintains a state machine model, the core of which is a set of explicit state transition rules. Each rule defines the trigger events and conditions that must be met to migrate from one state to another.
[0015] When the state transition engine captures solid evidence of performance or finds that the performance deadline has passed, it will automatically update the state of the VCO according to these logical rules.
[0016] Finally, the system provides a flexibly configurable business action trigger. This module subscribes to VCO state change events and automatically executes the corresponding business actions when it finds that the VCO meets the pre-set trigger conditions.
[0017] The present application fully considers the compliance requirements of the "Personal Information Protection Law of the People's Republic of China" and the "Data Security Law of the People's Republic of China" at every stage of the design. The system will use the following specific technical measures to ensure data security and privacy protection:
[0018] 1. Minimum necessary principle: the speech contract extraction module only extracts the necessary information directly related to the performance of commitment when processing the conversation, and does not extract and save irrelevant personal chats, sensitive information, etc.
[0019] 2. De-identification and encrypted storage: the obligation subject and right subject identification stored in the VCO should preferentially use the internal system generated de-identified ID that cannot be directly linked to the individual. All fields related to the original text of the conversation and the customer's identity information must be strongly encrypted in the "dynamic obligation ledger".
[0020] 3. Permission and access control: the system backend provides fine-grained permission management functions. By default, only the direct obligation subject of the VCO, its immediate superior, and authorized auditors can view the complete content of a specific VCO.
[0021] 4. User notification and consent: before the enterprise puts this system into use, it should explicitly inform users in its terms of service or privacy policy that their business conversation content may be used by the system to automatically extract and manage service commitments, and obtain the user's explicit consent.
[0022] 5. Intervenability of management dashboard: the management dashboard provided by the system allows authorized users (including customers as right subjects) to view, complain, and even request to delete incorrect VCOs under certain rules, ensuring the user's right to know and correct.
[0023] The beneficial effects of the present application are achieved through the above-mentioned systematic technical innovation, which solves the fundamental problems that the prior art cannot solve, and brings unprecedented certainty and reliability to modern business communication. BRIEF DESCRIPTION OF DRAWINGS
[0024] ATTACHMENT Figure 1 For an embodiment of the present application, the core process of the speech contract object-based method interacts with the system module schematic diagram. In the figure, the reference signs represent as follows:
[0025] 100 - unstructured conversation input;
[0026] 110 - speech contract extraction module;
[0027] 120 - dynamic obligation ledger;
[0028] 130 - monitoring and triggering hub;
[0029] 131 - state transition engine;
[0030] 132 - business action trigger;
[0031] 140 - Business system output.
[0032] Attachment Figure 2 This is a flowchart describing the complete lifecycle of a business interaction scenario, from the creation of a verbal contract to its tracking and ultimately triggering of an alert action, in one embodiment of the present invention. In the figure, the reference numerals represent the following:
[0033] 201-Instantiation step of speech contract;
[0034] 202-Risk warning condition satisfaction event;
[0035] 203-The system automatically executes the warning action;
[0036] 204-External evidence of promise performance;
[0037] 205 - Automatic update step of the status of the speech contract object;
[0038] 206 - Triggering step of subsequent business process.
[0039] Attachment Figure 3 This is a typical state transition path relationship diagram of a speech contract object during its life cycle in one embodiment of the present invention. DETAILED DESCRIPTION
[0040] To make the objectives, technical solutions, and advantages of the present invention more clear and complete, the technical solutions of the present invention will be further and fully described below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to illustrate the present invention and do not constitute any limitation to the present invention.
[0041] The present invention proposes a method for tracking obligations and controlling risks in long-term conversations based on verbal contract objects. Figure 1 As shown, its core process and system architecture include the following organically integrated parts.
[0042] First, the first step of this method is to collect the conversation content. Figure 1 As shown in the figure 100, the system receives unstructured dialogue input from business scenarios through a unified dialogue interface.
[0043] Next, the system sends the collected conversation content to the speech contract extraction module 110 for processing in real time or in batches. Figure 1 As shown, after identifying the commitment, the module 110 will immediately perform a key "instantiation" operation, that is, automatically converting the commitment and its necessary context information into a standardized data object, namely, a "verbal contract object (VCO)".
[0044] In the technical solutions of the present application, a core creative idea is to instantiate the unstructured commitment speech segment into a standardized data object, i.e., a "speech contract object (VCO)". This step is the key to converting natural language information into a machine-manageable unit. In order to ensure full disclosure of the invention, the preferred data structure of the VCO is described in detail here, which is sufficient to support the subsequent processing flows of the present application. A VCO instance logically includes at least the following attribute fields:
[0045] 1. VCO unique identifier: a string (String) type field used to uniquely identify each VCO instance in the system. To ensure its global uniqueness, its value can be generated by the universal unique identifier (UUID) generation algorithm, or by combining high-precision timestamps, machine identifiers, and self-incrementing serial numbers, etc. For example: "VCO-20250715-001".
[0046] 2. Obligation subject: a composite object (Object) type field used to record the detailed information of the commitment maker. The object internally includes at least two subfields: a type subfield used to distinguish the subject type (such as "internal employee"); and an ID subfield recording the unique identification of the subject (such as employee number or system user ID).
[0047] 3. Right subject: a composite object (Object) type field used to record the detailed information of the commitment receiver. Its internal structure is similar to that of the obligation subject, such as type "external customer" and ID the unique identification of the customer in the system.
[0048] 4. Original speech segment: a long text (Text / String) type field used to record the original dialogue text recognized by the speech contract extraction module as the direct basis for creating the VCO without modification. This field is an important basis for future auditing and tracing.
[0049] 5. Structured commitment content: a composite object (Object) type field that is the result of deep semantic analysis and information structuring of the original speech segment, allowing the machine to programmatically understand the specific content of the commitment. In an embodiment of the present application, the object can include but is not limited to the following subfields: core action (such as "add new function", "send document"), object or feature name of the action (such as "custom report filtering"), and other action-related parameters such as execution environment "user acceptance test environment", etc.
[0050] 6. Creation timestamp: a timestamp (Timestamp / DateTime) type field accurate to seconds or milliseconds, recording the exact time when the VCO was instantiated by the system.
[0051] 7. Expected Fulfillment Time Stamp: A timestamp type field that records the time point when the commitment should be fulfilled at the latest, as parsed from the dialog content. For example, for the expression "end of this month", the system will parse it to the last second of the last day of the month, combined with the creation timestamp.
[0052] 8. Business Priority: A field that indicates the importance of the commitment, which can be an enumerated type (such as "high", "medium", "low") or a numerical type. Its value can be automatically assessed by the system based on keywords in the dialog (such as the customer emphasizing "very important", "must") or manually adjusted by authorized business personnel in subsequent management.
[0053] 9. Fulfillment Status: An enumerated type field that indicates the current stage of the VCO in the life cycle, which is the core of the VCO state transition engine management. Its initial value is fixed as "to be fulfilled". Referring to the attached Figure 3 , its acceptable state values include but are not limited to: to be fulfilled, fulfilled, overdue, breached, canceled, etc.
[0054] 10. Status Change Log: An array type field, each array member of which is a log object that records the history of state changes. Each log object contains at least: timestamp (the precise time when the change occurs), new state (the new state after the change), and change reason or evidence description (for example, "associated project management task ID: PROJ-123 status is updated to 'completed')".
[0055] After the VCO is instantiated, the system writes it to a dedicated persistent storage unit. As shown in the attached Figure 1 , this storage unit is called dynamic obligation ledger 120.
[0056] The subsequent process of the invention is completed by the monitoring and triggering hub 130. This hub logically contains two sub-modules: state transition engine 131 and business action trigger 132. The VCO state transition engine 131 continuously monitors the life cycle status of each VCO. As shown in the attached Figure 1 , it reads the state information from the dynamic obligation ledger 120 and simultaneously receives subsequent dialog content and external data as input. The business action trigger 132 subscribes to the state change events of the VCO. When a VCO's state undergoes a critical transition or meets a pre-set risk condition, the trigger is activated and automatically performs the pre-set business action.
[0057] These business actions are ultimately reflected on the business system output 140. As shown in the attached Figure 1As shown in the bottom, these outputs can be in various forms, such as sending an automated reminder message to the obligor of the VCO, or sending a risk warning to the supervisor.
[0058] In summary, referring to the overall flow of the attached Figure 1 , the present application generates a VCO through the processing of the dialogue input 100 by the speech contract extraction module 110, stores it in the dynamic obligation ledger 120, and continuously tracks and judges the risk by the monitoring and triggering hub 130, finally produces intelligent business system output 140, thereby forming a complete, automated dialogue obligation management and risk control closed loop.
[0059] Embodiment one: applied to the customer communication management scenario of an enterprise software customization service company
[0060] This embodiment describes how the method and system of the present application are applied to a technical company that provides software customization development services for large enterprises.
[0061] Referring to the attached Figure 2 , a typical processing flow is as follows:
[0062] 1. Speech contract generation: In a chat for a project progress discussion, the sales representative promises the customer, "I will add the custom report function to the test environment for you before the end of this month (July 31st)".
[0063] 2. VCO instantiation (Fig. 201 in the attached Figure 2 ): The speech contract extraction module receives and analyzes the promise, and immediately automatically instantiates a VCO and stores it in the dynamic obligation ledger. The event corresponds to time point T1 on the time axis (July 15th).
[0064] 3. Risk warning (Figs. 202, 203 in the attached Figure 2 ): On July 29, 2025 (time point T2), the system detects that the VCO is about to expire but the status is still "to be performed", so it meets the warning rule and automatically executes the warning action, sending an urgent reminder to the relevant personnel.
[0065] 4. Automatic state transfer (Figs. 204, 205 in the attached Figure 2 ): The development team completes the function on July 30, 2025 and updates the task status in the external project management tool. The VCO state transfer engine captures this change through the interface and automatically updates the VCO status to "performed" after verification. The event corresponds to time point T3 on the time axis.
[0066] 5. Closed loop of the process (Fig. 206 in the attached Figure 2 ): The change of the state can further trigger the subsequent notification process, thereby completing the closed loop management of the entire event.
[0067] Example 2: Application in insurance sales and claims service scenarios
[0068] The application scenario of this embodiment is the sales and claims service of the insurance industry. Wang Wu, an agent of an insurance company, communicates with customer Zhao Liu about automobile insurance renewal through the company's official online customer service platform (the conversation is recorded by the system).
[0069] The creation of a verbal contract: During a conversation, client Zhao Liu asked about the claims process for minor accidents. Agent Wang Wu responded, "Rest assured, Mr. Zhao, for small claims under 5,000 yuan, our company offers an industry-leading fast-payment service. Simply upload photos and information about the accident online through our mobile app, and I guarantee the compensation will be deposited into your account within 24 hours."
[0070] VCO instantiation: refer to the attached Figure 1 Following the process, the system's verbal contract extraction module (110) immediately extracts Wang Wu's promise and creates a new VCO. The uniqueness of this VCO is that its fulfillment deadline rule is a relative time triggered by a condition, rather than a fixed absolute timestamp. Its due_ts_rule attribute is set to "within 24 hours after the claim event is triggered", and its initial fulfillment status is set to a special "pending_trigger" state, indicating that although the VCO has been generated, its fulfillment timer has not yet started.
[0071] Event-driven state transfer and risk warning: This process can also refer to the attached Figure 2 The life cycle model shown can be understood.
[0072] At some point in the future, customer Zhao Liu gets into a traffic accident that meets the requirements and submits a claim online. This “report event” is a key external data input for this system, which constitutes the starting point for VCO execution. Figure 2 When the system captures this event, it immediately activates the corresponding VCO, updates its fulfillment status from "waiting for trigger" to "pending", and calculates the precise expiration timestamp (i.e., current time + 24 hours) based on the fulfillment deadline rules.
[0073] From this moment on, the system background begins to continuously monitor the VCO, which corresponds to the attached Figure 2 During the entire life cycle from T1 to T3, the state transfer engine periodically queries the payment status of the company's internal claims system through the API.
[0074] If the performance period (e.g. 24 hours) is about to end, for example, Figure 2 At a similar time point T2, if the engine finds that the compensation has not been completed, it will meet the risk warning condition (202) and automatically perform the warning action (203), such as sending an emergency notice to the agent Wang Wu and his supervisor, prompting them to follow up immediately.
[0075] When the claim is finally completed and the payment gateway returns a "success" status, the state change of this external system constitutes an external evidence event of the promise fulfillment (see Appendix). Figure 2 204). The VCO state transfer engine captures this information and immediately automatically updates the VCO state to "fulfilled" (see Attachment 204). Figure 2 205), and trigger the subsequent archiving or notification process to complete the closed loop.
[0076] Possible state transitions: The VCO in this embodiment also follows the following in its life cycle: Figure 3 The state transition logic is shown. For example, if a claim is not paid within 24 hours, its status will automatically change from "pending" to "overdue." If the claim is subsequently paid, it will change from "overdue" to "fulfilled." If the commitment no longer needs to be fulfilled for some reason (such as the customer canceling the claim), it may change to "canceled."
[0077] This embodiment further demonstrates that the method and system proposed in the present invention can flexibly handle commitments with relative time limits triggered by specific events, and demonstrates its versatility and powerful functionality in different business scenarios through consistency with the logic of the accompanying drawings.
[0078] Refer to the attached Figure 3 This diagram summarizes the various possible state transitions of a VCO throughout its lifecycle. Upon creation, a VCO begins in the pending state. Depending on various events, it can transition to fulfilled, overdue, defaulted, or canceled. Each state transition can be captured by the system and trigger corresponding, highly customized business processes.
Claims
1. A method for tracking obligations and managing risks in long-term conversations, characterized in that: The following steps are involved: a) collecting one or more unstructured conversations between conversation participants through a conversation interface; b) performing semantic analysis on the conversation content by a speech contract extraction module to identify whether there are any speech segments expressing a binding commitment, agreement, or obligation of one party to the other party; c) If the speech fragment is recognized, the speech fragment and its contextual information are automatically instantiated into a standardized data object, called a Verbal Contract Object (VCO); the VCO contains at least the following attributes: an identifier of the obligor, an identifier of the obligee, a structured description of the commitment content, a creation timestamp, an expected fulfillment deadline, and an initial fulfillment status of "pending"; d) storing the VCO in a dedicated, persistent dynamic obligation ledger for unified management; e) Continuously monitor subsequent conversation content, associated external system data, or the passage of time through a VCO state transition engine; f) The state transfer engine, based on pre-set logical rules, automatically updates the performance status of the corresponding VCO in the dynamic obligation ledger, for example, from "pending performance" to "fulfilled" or "defaulted," upon detecting an event that proves that the commitment has been fulfilled, modified, or determined to be unfulfillable; g) The system sets one or more business action trigger conditions, which are associated with the VCO's performance status or its relationship with the performance deadline; when any VCO meets the trigger condition, the system automatically executes one or more preset business actions, which may include sending a notification to a designated person, creating a task in the business system, or adjusting the service policy for the user.
2. The method according to claim 1, characterized in that The attributes of the VCO in step c) further include: a business priority for evaluating the importance of the commitment, and a state change log that records the history of all state changes.
3. The method according to claim 1, characterized in that The verbal contract extraction module in step b) adopts a model based on natural language understanding, which is trained on a corpus containing a large number of commitment sentences to achieve the joint extraction of intentions, time entities and obligation relationships.
4. The method according to claim 1, wherein The logic rules of the state transfer engine described in step f) include: Content matching rules: Monitor subsequent conversations for statements confirming the fulfillment of a commitment. If so, update the status to "fulfilled." External data verification rules: Query the external business system through the application program interface. If the query results confirm that the commitment has been fulfilled, the status is updated to "Fulfilled"; Time monitoring rule: Monitors the current time. If the current time exceeds the expected fulfillment period and the status is still "To be fulfilled", the status will be updated to "Overdue".
5. The method according to claim 1, wherein The service action triggering conditions described in step g) include: Status change trigger: Triggered immediately when the VCO status changes to "fulfilled", "defaulted" or "overdue"; Risk warning trigger: When the current time distance to the expected fulfillment period is less than a preset time threshold and the VCO status is still "pending fulfillment", the "imminent overdue" warning action is triggered.
6. The method according to claim 1, characterized in that The preset business action described in step g) is a configurable action sequence. For example, when an "imminent overdue" warning is triggered, the first step is to send an automatic, friendly reminder message to the obligor. If the status remains unchanged within the specified time, the second step is to automatically create a high-priority follow-up ticket for the customer's responsible person in the customer relationship management system, and attach relevant VCO information as background material.
7. A conversation obligation tracking and risk management system for implementing the method according to any one of claims 1 to 6, characterized in that: The system physically includes at least one processor and memory, and logically includes: A conversation interface responsible for collecting conversations; a speech contract extraction module configured to perform steps b) and c); a dynamic obligation ledger constructed on said memory for performing step d); a VCO state transfer engine configured to perform steps e) and f) interacting with the dynamic obligation ledger; A business action triggering and execution module configured to execute step g).
8. The system according to claim 7, characterized in that The system also includes a management dashboard that provides a visual interface for authorized users to view a list of all VCOs, their current status, fulfillment history, and perform manual intervention on VCOs, such as modifying fulfillment deadlines or manually updating their status.