Digital rights automated negotiation system and method based on embodied intelligent agent

Abstract

The application discloses a digital right automated negotiation system and method based on embodied intelligent agent, and relates to the technical field of digital right management and intelligent negotiation.The multi-modal perception module of the embodied intelligent agent collects visual expressions, voice emotions, public information and economic shopping data of negotiation participants, performs cross-modal feature fusion and context understanding, generates a structured consumer portrait and a real-time emotion state vector;based on the context understanding result, combined with a reinforcement learning algorithm and a historical memory library, an adaptive negotiation strategy is dynamically generated to drive the agent to perform negotiation actions such as offering, bargaining and condition changing;after the negotiation is reached, an intelligent contract is automatically generated to call a digital right cancellation system and a payment system to complete right delivery and fund settlement, while continuously monitoring the execution state and automatically compensating for abnormalities and self-healing.The deep collaboration of multi-modal perception and decision-making, the closed-loop integration of negotiation and execution significantly improve the intelligent level, success rate and user experience of digital right negotiation, and ensure the efficient, accurate and safe operation of transactions.

Description

Technical Field

[0001] This invention belongs to the field of digital rights management and intelligent negotiation technology, and relates to an automated digital rights negotiation system and method based on embodied intelligent agents. Background Technology

[0002] With the rapid development of the digital economy, the trading and circulation of digital rights (such as memberships, coupons, access to digital collectibles, and service subscriptions) are becoming increasingly frequent. In the complex multilateral platform ecosystem, issuers, demanders, and service providers of rights need to frequently negotiate to reach agreements on the transfer, exchange, or joint use of rights. Currently, such negotiation processes mainly rely on the following methods:

[0003] 1. Manual negotiation: Business personnel communicate via email and instant messaging tools, which is inefficient, difficult to cope with high-frequency and real-time equity transaction needs, and the negotiation results are greatly affected by subjective factors and lack consistency.

[0004] 2. Automated systems based on rule engines: The system has preset negotiation logic (such as price range and validity period matching), which cannot dynamically adjust the strategy according to the real-time market environment, the behavior pattern of the counterparty and its own status. It has poor adaptability in complex and ever-changing negotiation scenarios and is prone to negotiation failure or suboptimal results.

[0005] 3. Decision-making systems based on offline machine learning models: Although they can learn patterns from historical data, the models become fixed after training and cannot dynamically adjust strategies based on real-time feedback from the opponent during the negotiation process. They also have difficulty handling sequential decision-making problems.

[0006] 4. Separation of negotiation and execution processes: After the negotiation is reached, the results still need to be entered into the rights and interests verification and payment system manually or semi-manually. The process is fragmented, which increases operational risks and delays, and lacks closed-loop feedback and self-healing ability on the execution status of the negotiation results.

[0007] In summary, existing technologies have problems such as limited perception dimensions, rigid strategy generation, separation of negotiation and execution, and lack of continuous learning ability when dealing with high-frequency, complex, and multi-party digital rights negotiations, making it difficult to guarantee the efficiency, accuracy, and stability of digital rights transactions. Summary of the Invention

[0008] In view of the problems existing in the prior art, the present invention provides a digital rights automated negotiation system and method based on embodied intelligent agents to solve the above-mentioned technical problems.

[0009] To achieve the above and other objectives, the technical solution adopted by the present invention is as follows:

[0010] This invention provides a method for automated negotiation of digital rights based on embodied intelligent agents, the method comprising:

[0011] Step S1: Through the multimodal perception module of the embodied intelligent agent, multimodal interaction is carried out with the digital rights interaction environment to collect visual expression data, voice emotion data, public information data and economic shopping data of the negotiation participants. Cross-modal feature fusion and contextual understanding are performed on the collected multi-source data to generate a structured consumer profile and real-time emotional state vector, and consumer contextual understanding data is obtained.

[0012] Step S2: Based on the consumer contextual understanding data and combined with the historical successful negotiation experience stored in the historical memory bank of the embodied intelligent agent, an adaptive negotiation strategy is generated using a reinforcement learning algorithm. The adaptive negotiation strategy includes an initial offer strategy, a concession strategy, a condition combination strategy, and an exit strategy, thereby obtaining the optimal negotiation strategy data.

[0013] Step S3: Based on the preferred negotiation strategy data, drive the embodied intelligent agent to perform automated negotiation actions, including quoting, negotiating, changing conditions and generating draft agreements, and perceive the multimodal feedback signals of the counterparty in real time. Use the feedback signals as immediate rewards for reinforcement learning, dynamically update the negotiation strategy, and form an adaptive negotiation model.

[0014] Step S4: Based on the negotiation results output by the adaptive negotiation model, automatically generate a smart contract, call the digital rights verification system and payment system through API to complete the rights transfer and fund settlement, continuously monitor the execution status, automatically compensate for anomalies during the execution process and perform self-healing, and output the final negotiation execution report.

[0015] Preferably, step S1 includes:

[0016] Step S11: Through the visual perception module of the embodied intelligent agent, collect facial expression image sequences and body posture data of the negotiation participants, and use a deep learning model to identify emotion categories, emotion intensity and emotion fluctuations to generate visual emotion feature data.

[0017] Step S12: Through the voice perception module of the embodied intelligent agent, collect the voice signals of the negotiation participants, extract acoustic features and perform semantic sentiment analysis to identify the emotional tendencies and implicit intentions in the voice and generate voice sentiment feature data.

[0018] Step S13: With authorization, obtain public information of the negotiation participants through the data interface, including social network profiles, historical consumption records, and behavioral footprint data, and integrate them to generate public consumer profile data;

[0019] Step S14: Based on publicly available consumer profile data, analyze the consumption capacity level, shopping intention intensity, and predicted bargaining space of the negotiation participants to generate economic shopping situation analysis data;

[0020] Step S15: Synchronize and align the visual emotion feature data, voice emotion feature data, publicly available consumer profile data, and economic shopping situation analysis data in time. Use a cross-modal attention fusion network to fuse the features and generate a structured consumer profile and real-time emotional state vector to obtain consumer contextual understanding data.

[0021] Preferably, step S2 includes:

[0022] Step S21: Based on consumer contextual understanding data, retrieve historical successful negotiation cases similar to the current consumer context from the long-term memory of the embodied intelligent agent, extract the initial pricing strategy, concession strategy and condition combination strategy used in the similar cases, and obtain historical successful strategy reference data.

[0023] Step S22: Using historical successful strategy reference data as the initial strategy candidate set, combined with the preset equity value model, the strategy parameters are quickly adjusted through meta-learning methods to adapt to the current personalized characteristics of consumers, and an initial adaptive negotiation strategy set is generated.

[0024] Step S23: Perform feasibility assessment and expected utility calculation for each strategy in the initial adaptive negotiation strategy set. The expected utility is calculated based on a comprehensive value function, which includes negotiation success rate, equity realization value, consumer satisfaction and friction cost factor, to obtain strategy expected utility data.

[0025] Step S24: Based on the strategy feasibility assessment data and the strategy expected utility data, the initial adaptive negotiation strategy set is screened and sorted, and the strategy with the highest comprehensive value is selected as the preferred negotiation strategy, and the preferred negotiation strategy data is output.

[0026] Preferably, step S3 includes:

[0027] Step S31: Based on the optimized negotiation strategy data, generate specific negotiation action instructions, send the negotiation actions to the negotiation environment through the interaction interface of the embodied intelligent agent, and set up a multimodal listener to capture the opponent's visual feedback, voice feedback and text feedback in real time, so as to obtain negotiation action execution data and real-time feedback data.

[0028] Step S32: Use the changes in visual expression, voice emotion, and semantic emotion in the real-time feedback data as immediate reward signals for reinforcement learning, update the current consumer context understanding data, and form a dynamic state space vector.

[0029] Step S33: Use the dynamic state space vector as the input of the reinforcement learning model and the next negotiation action as the output to construct a reinforcement learning model based on a deep Q-network or policy gradient algorithm, and continuously update the model parameters during the negotiation interaction process.

[0030] Step S34: Store the complete sequence of each negotiation interaction (including state, action, reward, and next state) into the short-term memory buffer for online learning and strategy fine-tuning, and periodically transfer successful cases to the long-term memory bank;

[0031] Step S35: Repeat steps S31 to S34 until a negotiation is reached or fails, and output the adaptive negotiation model for the current negotiation task.

[0032] Preferably, step S4 includes:

[0033] Step S41: Based on the negotiation results output by the adaptive negotiation model, automatically generate smart contract code that complies with legal regulations or platform rules to obtain smart contract data;

[0034] Step S42: Based on smart contract data, the digital rights verification system and payment system are called in parallel via API to complete rights delivery and fund settlement, and to obtain rights delivery and fund settlement execution data;

[0035] Step S43: Continuously monitor the execution status of equity delivery and fund settlement. If a success callback is received within the preset time, the execution is considered successful. If the timeout occurs or a failure callback is received, the execution is considered abnormal, and the execution status monitoring data is obtained.

[0036] Step S44: Based on the execution status monitoring data, perform automated compensation for execution anomalies, write the anomaly type, associated consumer characteristics, and occurrence time into the feedback queue, and trigger the compensation strategy. The compensation strategy includes payment rollback, rights freeze, retry mechanism, and backup plan switching to obtain automated compensation execution data.

[0037] Step S45: The execution exception information is used as feedback and input back to the perception module in step S1 and the strategy generation module in step S2 to optimize the identification and prevention of the abnormal links in subsequent negotiations, so as to realize the closed-loop collaborative optimization of negotiation and execution.

[0038] Step S46: Verify the consistency of the results after the automated compensation is executed, confirm that the equity status and the fund status are finally consistent, generate the final negotiation execution report, and output the report.

[0039] An automated digital rights negotiation system based on embodied intelligent agents. This system includes:

[0040] The multimodal perception module is used to collect visual expression data, voice emotion data, public information data, and economic shopping data of negotiation participants through visual perception unit, voice perception unit, and data interface, and to perform cross-modal feature fusion and contextual understanding to generate structured consumer profiles and real-time emotional state vectors.

[0041] The contextual cognition module, connected to the multimodal perception module, is used to construct a dynamic state space vector based on the structured consumer profile and real-time emotional state vector, combined with historical negotiation success experiences in the historical memory bank.

[0042] The strategy generation module, connected to the context cognition module, is used to generate an adaptive negotiation strategy based on a dynamic state space vector using a reinforcement learning algorithm, and output the preferred negotiation strategy data.

[0043] An interactive execution module, connected to the strategy generation module, is used to execute automated negotiation actions based on the preferred negotiation strategy data, and to capture the opponent's multimodal feedback signals in real time, feeding the feedback signals back to the strategy generation module for online learning.

[0044] The contract execution and self-healing module is connected to the interactive execution module. It is used to automatically generate a smart contract after the negotiation is reached, call the digital rights verification system and payment system to complete the rights transfer and fund settlement, and continuously monitor the execution status, and automatically compensate for anomalies and self-heal the status.

[0045] The memory module, connected to the context cognition module, strategy generation module, and contract execution and self-healing module, is used to store historical successful negotiation cases, strategy parameters, and execution anomaly modes, supporting context retrieval and experience reuse.

[0046] Preferably, the multimodal sensing module includes:

[0047] The visual perception unit is used to collect facial expression image sequences and body posture data, and generate visual emotion feature data through residual networks and micro-expression recognition models.

[0048] The speech perception unit is used to collect speech signals, extract acoustic emotion features through a recurrent neural network, and perform semantic emotion analysis through a natural language processing model to generate speech emotion feature data.

[0049] The public information collection unit is used to obtain social network profiles, historical consumption records, and behavioral footprint data under authorized conditions, and generate public consumer profile data.

[0050] The economic analysis unit, based on publicly available consumer profile data, uses regression models and clustering algorithms to analyze the predicted values ​​of consumption capacity, shopping intention intensity, and bargaining space, generating economic shopping situation analysis data.

[0051] The feature fusion unit employs a cross-modal attention fusion network to weight and fuse visual emotion features, voice emotion features, publicly available consumer profile features, and economic shopping situation features to generate a structured consumer profile and a real-time emotional state vector.

[0052] Preferably, the strategy generation module includes:

[0053] The context retrieval unit is used to retrieve similar historical success cases from the memory module based on the current consumer context understanding data, and generate an initial strategy candidate set;

[0054] The meta-learning adaptation unit is used to quickly adapt the initial policy candidate set to the current consumer features through meta-learning methods, and generate an initial adaptive negotiation policy set.

[0055] The comprehensive evaluation unit is used to assess the feasibility of the strategy and calculate its expected utility. The expected utility is calculated based on a comprehensive value function, which includes negotiation success rate, value of rights realization, consumer satisfaction, and friction cost factor.

[0056] The reinforcement learning optimization unit is used to update the dynamic state space vector based on real-time feedback signals during the negotiation and interaction process, and to continuously optimize the policy parameters through a deep Q-network or policy gradient algorithm.

[0057] Preferably, the contract execution and self-healing module includes:

[0058] The smart contract generation unit is used to automatically generate smart contract code based on the negotiation results.

[0059] The execution scheduling unit is used to call the APIs of the digital rights verification system and the payment system in parallel to complete the rights transfer and fund settlement.

[0060] The anomaly monitoring unit is used to continuously monitor the execution status and identify abnormal events such as timeouts and failures.

[0061] The compensation execution unit is used to trigger compensation strategies based on the type of anomaly, including payment rollback, rights and interests freeze, retry mechanism and backup plan switching;

[0062] The feedback closed-loop unit is used to transmit execution exception information back to the multimodal perception module and the policy generation module to optimize the exception prevention capability in subsequent negotiations.

[0063] Preferably, the system further includes an interpretable output module, which is used to output the basis for generating the negotiation strategy in a synchronous manner, including the basis for emotion recognition, the basis for judging purchasing power, the basis for predicting bargaining space, and the reference of historical similar cases, and to record decision nodes during the negotiation process for audit traceability.

[0064] As described above, the automated digital rights negotiation system and method based on embodied intelligent agents provided by the present invention have at least the following beneficial effects:

[0065] The invention provides an automated digital rights negotiation method based on embodied intelligent agents. This method collects data on visual expressions, vocal emotions, public information, and economic shopping situations through a multimodal perception module, and performs cross-modal feature fusion and contextual understanding, effectively solving the problem of single perception dimensions in existing technologies. On one hand, this method not only understands what consumers "said," but also "how they said it, their emotions, and the context," constructing a three-dimensional consumer profile and a real-time emotional state vector. This provides a data foundation for precise negotiation and significantly improves the targeting and effectiveness of negotiation strategies. On the other hand, through deep collaboration between reinforcement learning and historical memory databases, the agent can dynamically adjust negotiation strategies based on the consumer's real-time emotional state, economic situation, and shopping intentions, achieving truly personalized interaction and a unique negotiation experience for each individual. Simultaneously, by deeply integrating the negotiation process with rights verification and payment systems, and introducing automated compensation and state self-healing mechanisms, a fully automated closed loop from negotiation to agreement to execution is achieved. This greatly reduces operational delays and the risk of human error, ensuring the efficient, accurate, and secure operation of digital rights transactions, guaranteeing the system's economic viability and security, and significantly enhancing business continuity and reliability in complex environments. Attached Figure Description

[0066] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 This is a schematic diagram showing the connections between the steps of the method of the present invention. Detailed Implementation

[0068] The following description, in conjunction with the implementation of this invention, is merely an example and illustration of the concept of this invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the inventive concept or exceed the scope defined in these claims, all of which should fall within the protection scope of this invention.

[0069] Example 1

[0070] Please see Figure 1 As shown, the method for automated negotiation of digital rights based on embodied intelligent agents includes the following steps:

[0071] Step S1: Through the multimodal perception module of the embodied intelligent agent, multimodal interaction is carried out with the digital rights interaction environment to collect visual expression data, voice emotion data, public information data and economic shopping situation data of the negotiation participants. Cross-modal feature fusion and contextual understanding are performed on the collected multi-source data to generate a structured consumer profile and real-time emotional state vector, thus obtaining consumer contextual understanding data.

[0072] For example, step S1 includes:

[0073] Step S11: Through the visual perception module of the embodied intelligent agent, collect facial expression image sequences and body posture data of the negotiation participants, and use a deep learning model to identify emotion categories, emotion intensity and emotion fluctuations to generate visual emotion feature data.

[0074] Step S12: Through the voice perception module of the embodied intelligent agent, collect the voice signals of the negotiation participants, extract acoustic features and perform semantic sentiment analysis to identify the emotional tendencies and implicit intentions in the voice and generate voice sentiment feature data.

[0075] Step S13: With authorization, obtain public information of the negotiation participants through the data interface, including social network profiles, historical consumption records, and behavioral footprint data, and integrate them to generate public consumer profile data;

[0076] Step S14: Based on publicly available consumer profile data, analyze the consumption capacity level, shopping intention intensity, and predicted bargaining space of the negotiation participants to generate economic shopping situation analysis data;

[0077] Step S15: Synchronize and align the visual emotion feature data, voice emotion feature data, publicly available consumer profile data, and economic shopping situation analysis data in time. Use a cross-modal attention fusion network to fuse the features and generate a structured consumer profile and real-time emotional state vector to obtain consumer contextual understanding data.

[0078] It should be added that step S15 includes:

[0079] Step S151: Time-stamp alignment and sampling frequency unification are performed on visual emotion feature data, voice emotion feature data, consumer public profile data, and economic shopping situation analysis data to establish a multimodal data correspondence table and obtain multimodal synchronized data;

[0080] Step S152: Based on multimodal synchronous data, a cross-modal attention mechanism is used to perform weighted fusion of features of each modality, calculate the correlation weight between features of different modalities, and when speech semantic features and visual emotion features show strong correlation within the time window, the joint features are given higher fusion weights to obtain cross-modal correlation weight data.

[0081] Step S153: Based on the cross-modal association weight data, perform weighted summation and nonlinear transformation on the features of each modality to generate a fused high-dimensional context feature vector, thus obtaining context feature fusion data;

[0082] Step S154: Based on the contextual feature fusion data, the structured consumer profile and real-time emotional state vector are output through classifiers and regression models to obtain consumer contextual understanding data.

[0083] In this embodiment of the invention, firstly, the visual perception module of the embodied intelligent agent collects a sequence of facial expression images of the negotiation participants after they enter the digital rights negotiation interaction interface. A micro-expression recognition model based on residual networks is used to analyze the facial muscle movement unit features, detecting a micro-expression pattern in the negotiation participants' faces characterized by brow retraction, lip closure, and frequent gaze shifts. A classifier determines the emotion category to be a composite state of "hesitation" and "dissatisfaction," with an emotion intensity index of 0.72 (with a set threshold of 0-1) and an emotion fluctuation coefficient of 0.28, generating visual emotion feature data. Simultaneously, the voice perception module collects the voice input stream of the negotiation participants, extracting acoustic features such as pitch standard deviation, speech rate change rate, average volume, and pause frequency. A recurrent neural network is used to identify the voice emotion tendency as "anxiety," and a Transformer-based natural language processing model is used for semantic sentiment analysis, identifying implicit intention labels of "tentative bargaining" and "potential decision-making anxiety," generating voice emotion feature data.

[0084] With authorization, publicly available information from negotiation participants is obtained through data interfaces: social network profile data includes characteristics such as frequency of consumer behavior sharing and brand interaction preferences; historical consumption record data shows that the total consumption amount in the past 12 months is in the top 20% quantile of the platform's user group; behavioral footprint data records that in the past 30 days, the number of similar benefits viewed 5 times, the number of high-priced items in the shopping cart 3, and the average page dwell time 120 seconds. The above data is integrated to generate publicly available consumer profile data. Based on this publicly available consumer profile data, the economic analysis unit predicts the consumption capacity level as "medium-high" using a gradient boosting regression model, analyzes the shopping intention intensity index as 0.9 using a K-means clustering algorithm (with a threshold of 0-1), and calculates the bargaining elasticity coefficient as 0.15 to 0.35 (based on the original price) using a bargaining space prediction model, generating economic shopping situation analysis data.

[0085] The feature fusion unit aligns visual emotion features, vocal emotion features, consumer public profile features, and economic shopping situation features with timestamps and unifies sampling frequencies to establish a multimodal data correspondence table, obtaining multimodal synchronous data. A cross-modal attention mechanism is used to weight and fuse the features of each modality. When the "bargaining intention" label in the vocal semantic features and the "hesitation" emotion in the visual emotion features show a strong correlation (correlation coefficient ≥ 0.8) within the same time window, the model assigns a joint weight of 0.85 to the "price sensitivity" feature. This is combined with the shopping intention intensity index (0.9) and the consumption capacity level index (0.8) to calculate the cross-modal association weight data. Based on the association weights, the features of each modality are weighted, summed, and nonlinearly transformed to generate a fused 128-dimensional contextual feature vector, obtaining the contextual feature fusion data. Finally, a structured consumer profile is output through a multilayer perceptron classifier and a linear regression model: spending power level = "medium to high level" (confidence 0.8), price sensitivity coefficient = 0.65, preference label vector = [quality priority, brand sensitivity]; real-time emotional state vector: emotion type encoding = [anxiety, tentative], emotion intensity value = 0.7, emotion volatility value = 0.3, thus obtaining consumer contextual understanding data.

[0086] Step S2: Based on the consumer contextual understanding data and combined with the historical successful negotiation experiences stored in the historical memory bank of the embodied intelligent agent, an adaptive negotiation strategy is generated using a reinforcement learning algorithm. The adaptive negotiation strategy includes an initial offer strategy, a concession strategy, a condition combination strategy, and an exit strategy, thereby obtaining the optimal negotiation strategy data.

[0087] For example, step S2 includes:

[0088] Step S21: Based on consumer contextual understanding data, retrieve historical successful negotiation cases similar to the current consumer context from the long-term memory of the embodied intelligent agent, extract the initial pricing strategy, concession strategy and condition combination strategy used in the similar cases, and obtain historical successful strategy reference data.

[0089] Step S22: Using historical successful strategy reference data as the initial strategy candidate set, combined with the preset equity value model, the strategy parameters are quickly adjusted through meta-learning methods to adapt to the current personalized characteristics of consumers, and an initial adaptive negotiation strategy set is generated.

[0090] Step S23: Perform feasibility assessment and expected utility calculation for each strategy in the initial adaptive negotiation strategy set. The expected utility is calculated based on the comprehensive value function to obtain the strategy expected utility data.

[0091] Step S24: Based on the strategy feasibility assessment data and the strategy expected utility data, the initial adaptive negotiation strategy set is screened and sorted, and the strategy with the highest comprehensive value is selected as the preferred negotiation strategy, and the preferred negotiation strategy data is output.

[0092] It should be added that step S23 includes:

[0093] Step S231: Based on the initial adaptive negotiation strategy set, verify the feasibility of each strategy item by item, check whether the quotation required by the strategy is within the preset authorization range, and whether the exchange conditions proposed in the strategy are owned and delivered by our side, and obtain strategy feasibility assessment data.

[0094] Step S232: Calculate the expected utility of the strategies that have passed the feasibility assessment, and construct a comprehensive value function V = α·P_success + β·V_equity + γ·S_satisfaction - δ·C_friction, where P_success is the predicted negotiation success rate, V_equity is the value of realized rights, S_satisfaction is the predicted consumer satisfaction, and C_friction is the friction cost, to obtain the expected utility data of the strategy;

[0095] Step S233: Based on the expected utility data of the strategies, sort and compare the expected utility values ​​of each strategy, identify the strategy with the highest utility value and its corresponding strategy parameter combination, and obtain the optimal strategy identification data;

[0096] Step S234: Based on the optimal strategy identification data, output the strategy with the highest expected utility value as the preferred negotiation strategy, and record the expected utility value and feasibility verification results of the strategy to obtain the output data of the preferred negotiation strategy.

[0097] In this embodiment of the invention, based on consumer contextual understanding data, historical successful negotiation cases similar to the current consumer context are retrieved from the long-term memory of the embodied intelligent agent. The cosine similarity between the current consumer context vector and the historical context vector in the memory is calculated, and a similarity threshold of ≥0.85 is set. The three historical successful cases with the highest similarity are retrieved. The consumer context characteristics in these cases are all of the type "high purchase intention, moderate to high price sensitivity, and anxiety tendency". The effective strategy combination used in the successful cases is "rapid price reduction strategy + additional benefit gift strategy". The strategy parameters of these strategies (price reduction range of 8%-12%, additional benefit type is trial benefit) are extracted as the initial strategy candidate set to obtain historical successful strategy reference data.

[0098] Using historical successful strategy reference data as the initial strategy candidate set, and combining it with a preset benefit value model (the marginal cost of the benefit approaches zero and the expected value coefficient is 0.3), the strategy parameters are quickly adjusted through a model-independent meta-learning algorithm to adapt to the current personalized characteristics of consumers, generating an initial adaptive negotiation strategy set, including three candidate strategies: Strategy A "10% price reduction + free trial benefit", Strategy B "15% price reduction + no additional benefits", and Strategy C "maintain original price + free multiple trial benefits".

[0099] Each strategy in the initial adaptive negotiation strategy set undergoes a feasibility assessment. This involves checking whether the required offer is within the preset authorization range (the lower limit of authorization is 80% of the original price) and whether the exchange conditions proposed in the strategy are owned by our equity pool and are deliverable. Verification shows that the offers of strategies A, B, and C are all within the authorization range, and the exchange conditions are all deliverable, yielding strategy feasibility assessment data. The expected utility of the strategies that pass the feasibility assessment is calculated, constructing a comprehensive value function V = α·P_success + β·V_equity + γ·S_satisfaction - δ·C_friction, with weighting coefficients set to α=0.5, β=0.2, γ=0.2, and δ=0.1. Using Monte Carlo simulation and combining it with opponent behavior pattern data, the expected utility value is calculated to be 0.86 for strategy A, 0.72 for strategy B, and 0.68 for strategy C, yielding the strategy expected utility data. The expected utility values ​​of each strategy are sorted in descending order. The strategy with the highest utility value is identified as Strategy A, "10% price reduction + free trial benefits". This strategy is output as the preferred negotiation strategy, and the expected utility value of this strategy (0.86) and the feasibility verification result are recorded to obtain the output data of the preferred negotiation strategy.

[0100] Step S3: Based on the preferred negotiation strategy data, drive the embodied intelligent agent to perform automated negotiation actions, including quoting, negotiating, changing conditions and generating draft agreements, and perceive the multimodal feedback signals of the counterparty in real time. Use the feedback signals as immediate rewards for reinforcement learning, dynamically update the negotiation strategy, and form an adaptive negotiation model.

[0101] For example, step S3 includes:

[0102] Step S31: Based on the optimized negotiation strategy data, generate specific negotiation action instructions, send the negotiation actions to the negotiation environment through the interaction interface of the embodied intelligent agent, and set up a multimodal listener to capture the opponent's visual feedback, voice feedback and text feedback in real time, so as to obtain negotiation action execution data and real-time feedback data.

[0103] Step S32: Use the changes in visual expression, voice emotion, and semantic emotion in the real-time feedback data as immediate reward signals for reinforcement learning, update the current consumer context understanding data, and form a dynamic state space vector.

[0104] Step S33: Use the dynamic state space vector as the input of the reinforcement learning model and the next negotiation action as the output to construct a reinforcement learning model based on a deep Q-network or policy gradient algorithm, and continuously update the model parameters during the negotiation interaction process.

[0105] Step S34: Store the complete sequence of each negotiation interaction (including state, action, reward, and next state) into the short-term memory buffer for online learning and strategy fine-tuning, and periodically transfer successful cases to the long-term memory bank;

[0106] Step S35: Repeat steps S31 to S34 until a negotiation is reached or fails, and output the adaptive negotiation model for the current negotiation task.

[0107] It should be added that step S33 includes:

[0108] Step S331: Based on the dynamic state space vector, construct a deep Q-network model, take the state vector as the input layer, extract state features through a multi-layer fully connected network, and output the Q value of each candidate action to obtain the initial reinforcement learning model.

[0109] Step S332: During the negotiation interaction, an ε-greedy strategy is adopted to select actions. New actions are explored randomly with probability ε, and the action with the highest current Q value is selected with probability 1-ε. The selected action is sent to the interaction execution module for execution to obtain action selection data.

[0110] Step S333: Receive the instant reward signal returned by the interactive execution module, calculate the target Q value, update the parameters of the deep Q network by minimizing the mean squared error loss function, so that the Q value predicted by the model gradually approaches the real cumulative reward, and obtain the model parameter update data;

[0111] Step S334: Periodically sample interactive sequence samples from the short-term memory buffer for experience replay training, break sample correlation, improve model training stability, and output the updated reinforcement learning model.

[0112] In this embodiment of the invention, based on the preferred negotiation strategy data "10% price reduction + free trial benefits," specific negotiation action instructions are generated. These instructions are then sent to the negotiation participants via the voice synthesis interface of the embodied intelligent agent, using a preset gentle tone template: "Based on your current needs and historical behavior patterns, we can offer a 10% price discount on the original price, plus a free 7-day trial benefit as an added value service. This offer is valid until the end of this round of negotiation. Please confirm your acceptance." Simultaneously, a multimodal listener is set up to capture the opponent's visual, voice, and text feedback in real time, obtaining negotiation action execution data and real-time feedback data.

[0113] The agent captures real-time feedback from negotiation participants: the visual perception module detects a change in facial expression from "brows furrowed" to "brows relaxed, corners of mouth raised," and a change in head posture from "tilted back" to "tilted forward"; the voice perception module detects a decrease in speech rate from 180 words per minute to 120 words per minute, a decrease in volume from 72 decibels to 65 decibels, and a change in voice emotion recognition output from "anxious" to "calm"; the text semantic analysis result is "acceptable." The changes in visual expression (positive shift), voice emotion (anxiety → calm), and semantic emotion transition (hesitation → acceptance) in the real-time feedback data are used as immediate reward signals for reinforcement learning, with a reward value of +10. This updates the current consumer contextual understanding data, forming a dynamic state space vector S_t = [emotional intensity value = 0.3, emotional tendency encoding = positive, purchasing power level encoding = 3, current shopping intention value = 0.95, bargaining elasticity coefficient = 0.2, historical interaction rounds = 2, opponent's behavioral characteristics encoding = quick acceptance].

[0114] Based on dynamic state space vectors, a deep Q-network model is constructed. A 14-dimensional state vector is used as the input layer, and state features are extracted through a three-layer fully connected network (with hidden layer dimensions of 64, 32, and 16 respectively). The output layer is the Q-value of five candidate actions (candidate actions include: confirming the deal, making a small concession, maintaining the original offer, proposing a conditional exchange, and ending the negotiation), resulting in an initial reinforcement learning model. During the negotiation interaction, an ε-greedy strategy is used to select actions. Initially, ε=0.1, new actions are randomly explored with a probability of 0.1, and the action with the highest current Q-value ("confirming the deal," Q-value = 8.6) is selected with a probability of 0.9. The selected action is sent to the interaction execution module for execution, obtaining action selection data. The immediate reward signal (+10) returned by the interaction execution module is received, the target Q-value is calculated, and the parameters of the deep Q-network are updated by minimizing the mean squared error loss function. The learning rate is set to 0.001, allowing the model's predicted Q-value to gradually approach the actual cumulative reward, resulting in updated model parameter data. Periodically sample 32 interaction sequences randomly from the short-term memory buffer for experience replay training to break sample correlation and improve model training stability. Repeat the above steps, and the negotiation is reached within 3 rounds of interaction, with a total time of 45 seconds. Output an adaptive negotiation model for this negotiation task.

[0115] Step S4: Based on the negotiation results output by the adaptive negotiation model, automatically generate a smart contract, call the digital rights verification system and payment system through API to complete the rights transfer and fund settlement, continuously monitor the execution status, automatically compensate for anomalies during the execution process and perform self-healing, and output the final negotiation execution report.

[0116] For example, step S4 includes:

[0117] Step S41: Based on the negotiation results output by the adaptive negotiation model, automatically generate smart contract code that complies with legal regulations or platform rules to obtain smart contract data;

[0118] Step S42: Based on smart contract data, the digital rights verification system and payment system are called in parallel via API to complete rights delivery and fund settlement, and to obtain rights delivery and fund settlement execution data;

[0119] Step S43: Continuously monitor the execution status of equity delivery and fund settlement. If a success callback is received within the preset time, the execution is considered successful. If the timeout occurs or a failure callback is received, the execution is considered abnormal, and the execution status monitoring data is obtained.

[0120] Step S44: Based on the execution status monitoring data, perform automatic compensation for execution anomalies, write the anomaly type, associated consumer characteristics, and occurrence time into the feedback queue, and trigger the compensation strategy to obtain automatic compensation execution data;

[0121] Step S45: The execution exception information is used as feedback and input back to the perception module in step S1 and the strategy generation module in step S2 to optimize the identification and prevention of the abnormal links in subsequent negotiations, so as to realize the closed-loop collaborative optimization of negotiation and execution.

[0122] Step S46: Verify the consistency of the results after the automated compensation is executed, confirm that the equity status and the fund status are finally consistent, generate the final negotiation execution report, and output the report.

[0123] It should be added that step S44 includes:

[0124] Step S441: Based on the execution status monitoring data, classify and identify the types of exceptions. If the payment system successfully returns a callback timeout, mark it as a payment timeout exception; if the rights and interests redemption system returns a failure code, mark it as a redemption failure exception; if the interface call returns a network error, mark it as a network exception, and obtain exception classification data.

[0125] Step S442: Based on the anomaly classification data, match the corresponding compensation strategy. For payment timeout anomalies, trigger the payment rollback compensation strategy and call the payment system's refund interface; for reconciliation failure anomalies, trigger the retry compensation strategy and set the number of retries and intervals; for network anomalies, trigger the backup channel switching strategy and obtain compensation strategy matching data.

[0126] Step S443: Execute the compensation strategy, monitor the execution status of the compensation operation in real time, record the success rate and time of the compensation operation, and obtain the compensation execution data;

[0127] Step S444: Write the anomaly type, compensation strategy, compensation result, and consumer characteristic association information into the feedback queue, so that the perception module and strategy generation module can subscribe to and consume it to obtain feedback queue data.

[0128] In this embodiment of the invention, based on the negotiation result output by the adaptive negotiation model (final transaction price of 52.5 yuan, transaction quantity of 100 units, benefit type of premium membership, and benefit validity period extended by 15 days from the original validity period), smart contract code conforming to platform rules and legal regulations is automatically generated. The smart contract includes core elements such as benefit delivery conditions, fund settlement conditions, and breach of contract liability clauses, thus obtaining smart contract data. Based on the smart contract data, the payment system deduction interface and the benefit transfer interface of the benefit verification system are called in parallel via API to complete the benefit delivery and fund settlement, obtaining benefit delivery and fund settlement execution data.

[0129] The execution status of rights settlement and fund settlement is continuously monitored. The monitoring component records that the payment system's deduction interface returns an HTTP status code 200 and success callback data within 1.2 seconds. However, if the rights reimbursement system does not return any response within a preset timeout threshold of 5 seconds, it is determined to be an execution anomaly, and execution status monitoring data is obtained. Based on the execution status monitoring data, the anomaly type is classified and identified. The lack of response from the rights reimbursement system is determined to be a reimbursement timeout anomaly, and anomaly classification data is obtained. According to the anomaly classification data, the corresponding compensation strategy is matched. For reimbursement timeout anomalies, a retry compensation strategy is triggered, with the number of retries set to 3, and the retry interval adopting an exponential backoff algorithm (initial interval of 3 seconds, followed by 6 seconds and 12 seconds). Compensation strategy matching data is obtained.

[0130] The compensation strategy is executed immediately by calling the payment system's refund interface to roll back the deducted amount. The rights verification task is marked as "pending retry" and written to the retry queue. The execution status of the compensation operation is monitored in real time, and the success rate and time taken are recorded to obtain compensation execution data. During the first retry (after a 3-second delay), the system calls the rights verification interface again. The interface returns an HTTP status code 200 and success callback data, indicating that the rights transfer is complete and the retry is recorded as successful. The exception type (verification timeout), compensation strategy (retry strategy), compensation result (success), and consumer characteristic association information are written to the feedback queue for the perception module and strategy generation module to subscribe to and consume, thus obtaining feedback queue data.

[0131] The abnormal execution information is used as feedback and fed back to the perception module in step S1 and the strategy generation module in step S2. This ensures that in subsequent negotiation interactions with consumers exhibiting similar characteristics, the system will prioritize the type of rights with a higher success rate of redemption or increase the monitoring frequency threshold of the redemption process during the strategy generation phase, achieving closed-loop collaborative optimization of negotiation and execution. The system verifies the consistency of the results after automated compensation execution, checking the transaction status in the payment system as "refunded - successful retry of deduction" and the rights status in the rights redemption system as "transferred," confirming that the rights status and fund status are ultimately consistent (successful rights transfer and successful fund settlement). A final negotiation execution report is generated, including a summary of the negotiation process (3 rounds of negotiation, total time 45 seconds, final transaction price 52.5 yuan / unit), execution results (payment transaction number PAY123456, redemption transaction number VER789012), abnormal compensation records (abnormality type: redemption timeout, compensation strategy: 3 retries with 3-second intervals, compensation result: 1st retry successful), and the final status (transaction completed). The report is then output.

[0132] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0133] It should be understood that determining B based on A does not mean determining B solely based on A; it also means determining B based on A and / or other information.

[0134] 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.

[0135] 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 method for automated negotiation of digital rights based on embodied intelligent agents, characterized in that, The method includes: Step S1: Through the multimodal perception module of the embodied intelligent agent, multimodal interaction is carried out with the digital rights interaction environment to collect visual expression data, voice emotion data, public information data and economic shopping data of the negotiation participants. Cross-modal feature fusion and contextual understanding are performed on the collected multi-source data to generate a structured consumer profile and real-time emotional state vector, and consumer contextual understanding data is obtained. Step S2: Based on the consumer contextual understanding data and combined with the historical successful negotiation experience stored in the historical memory bank of the embodied intelligent agent, an adaptive negotiation strategy is generated using a reinforcement learning algorithm. The adaptive negotiation strategy includes an initial offer strategy, a concession strategy, a condition combination strategy, and an exit strategy, thereby obtaining the optimal negotiation strategy data. Step S3: Based on the preferred negotiation strategy data, drive the embodied intelligent agent to perform automated negotiation actions, including quoting, negotiating, changing conditions and generating draft agreements, and perceive the multimodal feedback signals of the counterparty in real time. Use the feedback signals as immediate rewards for reinforcement learning, dynamically update the negotiation strategy, and form an adaptive negotiation model. Step S4: Based on the negotiation results output by the adaptive negotiation model, automatically generate a smart contract, call the digital rights verification system and payment system through API to complete the rights transfer and fund settlement, continuously monitor the execution status, automatically compensate for anomalies during the execution process and perform self-healing, and output the final negotiation execution report.

2. The method according to claim 1, characterized in that, Step S1 includes: Step S11: Through the visual perception module of the embodied intelligent agent, collect facial expression image sequences and body posture data of the negotiation participants, and use a deep learning model to identify emotion categories, emotion intensity and emotion fluctuations to generate visual emotion feature data. Step S12: Through the voice perception module of the embodied intelligent agent, collect the voice signals of the negotiation participants, extract acoustic features and perform semantic sentiment analysis to identify the emotional tendencies and implicit intentions in the voice and generate voice sentiment feature data. Step S13: With authorization, obtain public information of the negotiation participants through the data interface, including social network profiles, historical consumption records, and behavioral footprint data, and integrate them to generate public consumer profile data; Step S14: Based on publicly available consumer profile data, analyze the consumption capacity level, shopping intention intensity, and predicted bargaining space of the negotiation participants to generate economic shopping situation analysis data; Step S15: Synchronize and align the visual emotion feature data, voice emotion feature data, publicly available consumer profile data, and economic shopping situation analysis data in time. Use a cross-modal attention fusion network to fuse the features and generate a structured consumer profile and real-time emotional state vector to obtain consumer contextual understanding data.

3. The method according to claim 2, characterized in that, Step S2 includes: Step S21: Based on consumer contextual understanding data, retrieve historical successful negotiation cases similar to the current consumer context from the long-term memory of the embodied intelligent agent, extract the initial pricing strategy, concession strategy and condition combination strategy used in the similar cases, and obtain historical successful strategy reference data. Step S22: Using historical successful strategy reference data as the initial strategy candidate set, combined with the preset equity value model, the strategy parameters are quickly adjusted through meta-learning methods to adapt to the current personalized characteristics of consumers, and an initial adaptive negotiation strategy set is generated. Step S23: Perform feasibility assessment and expected utility calculation for each strategy in the initial adaptive negotiation strategy set. The expected utility is calculated based on a comprehensive value function, which includes negotiation success rate, equity realization value, consumer satisfaction and friction cost factor, to obtain strategy expected utility data. Step S24: Based on the strategy feasibility assessment data and the strategy expected utility data, the initial adaptive negotiation strategy set is screened and sorted, and the strategy with the highest comprehensive value is selected as the preferred negotiation strategy, and the preferred negotiation strategy data is output.

4. The method according to claim 3, characterized in that, Step S3 includes: Step S31: Based on the optimized negotiation strategy data, generate specific negotiation action instructions, send the negotiation actions to the negotiation environment through the interaction interface of the embodied intelligent agent, and set up a multimodal listener to capture the opponent's visual feedback, voice feedback and text feedback in real time, so as to obtain negotiation action execution data and real-time feedback data. Step S32: Use the changes in visual expression, voice emotion, and semantic emotion in the real-time feedback data as immediate reward signals for reinforcement learning, update the current consumer context understanding data, and form a dynamic state space vector. Step S33: Use the dynamic state space vector as the input of the reinforcement learning model and the next negotiation action as the output to construct a reinforcement learning model based on a deep Q-network or policy gradient algorithm, and continuously update the model parameters during the negotiation interaction process. Step S34: Store the complete sequence of each negotiation interaction (including state, action, reward, and next state) into the short-term memory buffer for online learning and strategy fine-tuning, and periodically transfer successful cases to the long-term memory bank; Step S35: Repeat steps S31 to S34 until negotiation is reached or fails, and output the adaptive negotiation model for the current negotiation task.

5. The method according to claim 4, characterized in that, Step S4 includes: Step S41: Based on the negotiation results output by the adaptive negotiation model, automatically generate smart contract code that complies with legal regulations or platform rules to obtain smart contract data; Step S42: Based on smart contract data, the digital rights verification system and payment system are called in parallel via API to complete rights delivery and fund settlement, and to obtain rights delivery and fund settlement execution data; Step S43: Continuously monitor the execution status of equity delivery and fund settlement. If a success callback is received within the preset time, the execution is considered successful. If the timeout occurs or a failure callback is received, the execution is considered abnormal, and the execution status monitoring data is obtained. Step S44: Based on the execution status monitoring data, perform automated compensation for execution anomalies, write the anomaly type, associated consumer characteristics, and occurrence time into the feedback queue, and trigger the compensation strategy. The compensation strategy includes payment rollback, rights freeze, retry mechanism, and backup plan switching to obtain automated compensation execution data. Step S45: The execution exception information is used as feedback and input back to the perception module in step S1 and the strategy generation module in step S2 to optimize the identification and prevention of the abnormal links in subsequent negotiations, so as to realize the closed-loop collaborative optimization of negotiation and execution. Step S46: Verify the consistency of the results after the automated compensation is executed, confirm that the equity status and the fund status are finally consistent, generate the final negotiation execution report, and output the report.

6. A digital rights automated negotiation system based on embodied intelligent agents, characterized in that: The system includes: The multimodal perception module is used to collect visual expression data, voice emotion data, public information data, and economic shopping data of negotiation participants through visual perception unit, voice perception unit, and data interface, and to perform cross-modal feature fusion and contextual understanding to generate structured consumer profiles and real-time emotional state vectors. The contextual cognition module, connected to the multimodal perception module, is used to construct a dynamic state space vector based on the structured consumer profile and real-time emotional state vector, combined with historical negotiation success experiences in the historical memory bank. The strategy generation module, connected to the context cognition module, is used to generate an adaptive negotiation strategy based on a dynamic state space vector using a reinforcement learning algorithm, and output the preferred negotiation strategy data. An interactive execution module, connected to the strategy generation module, is used to execute automated negotiation actions based on the preferred negotiation strategy data, and to capture the opponent's multimodal feedback signals in real time, feeding the feedback signals back to the strategy generation module for online learning. The contract execution and self-healing module is connected to the interactive execution module. It is used to automatically generate a smart contract after the negotiation is reached, call the digital rights verification system and payment system to complete the rights transfer and fund settlement, and continuously monitor the execution status, and automatically compensate for anomalies and self-heal the status. The memory module, connected to the context cognition module, strategy generation module, and contract execution and self-healing module, is used to store historical successful negotiation cases, strategy parameters, and execution anomaly modes, supporting context retrieval and experience reuse.

7. The system according to claim 6, characterized in that, The multimodal sensing module includes: The visual perception unit is used to collect facial expression image sequences and body posture data, and generate visual emotion feature data through residual networks and micro-expression recognition models. The speech perception unit is used to collect speech signals, extract acoustic emotion features through a recurrent neural network, and perform semantic emotion analysis through a natural language processing model to generate speech emotion feature data. The public information collection unit is used to obtain social network profiles, historical consumption records, and behavioral footprint data under authorized conditions, and generate public consumer profile data. The economic analysis unit, based on publicly available consumer profile data, uses regression models and clustering algorithms to analyze the predicted values ​​of consumption capacity, shopping intention intensity, and bargaining space, generating economic shopping situation analysis data. The feature fusion unit employs a cross-modal attention fusion network to weight and fuse visual emotion features, voice emotion features, publicly available consumer profile features, and economic shopping situation features to generate a structured consumer profile and a real-time emotional state vector.

8. The system according to claim 6, characterized in that, The strategy generation module includes: The context retrieval unit is used to retrieve similar historical success cases from the memory module based on the current consumer context understanding data, and generate an initial strategy candidate set; The meta-learning adaptation unit is used to quickly adapt the initial policy candidate set to the current consumer features through meta-learning methods, and generate an initial adaptive negotiation policy set. The comprehensive evaluation unit is used to assess the feasibility of the strategy and calculate its expected utility. The expected utility is calculated based on a comprehensive value function, which includes negotiation success rate, value of rights realization, consumer satisfaction, and friction cost factor. The reinforcement learning optimization unit is used to update the dynamic state space vector based on real-time feedback signals during the negotiation and interaction process, and to continuously optimize the policy parameters through a deep Q-network or policy gradient algorithm.

9. The system according to claim 6, characterized in that, The contract execution and self-healing module includes: The smart contract generation unit is used to automatically generate smart contract code based on the negotiation results. The execution scheduling unit is used to call the APIs of the digital rights verification system and the payment system in parallel to complete the rights transfer and fund settlement. The anomaly monitoring unit is used to continuously monitor the execution status and identify abnormal events such as timeouts and failures. The compensation execution unit is used to trigger compensation strategies based on the type of anomaly, including payment rollback, rights and interests freeze, retry mechanism and backup plan switching; The feedback closed-loop unit is used to transmit execution exception information back to the multimodal perception module and the policy generation module to optimize the exception prevention capability in subsequent negotiations.

10. The system according to claim 6, characterized in that, The system also includes an interpretable output module, which is used to output the basis for generating the negotiation strategy in a synchronous manner, including the basis for emotion recognition, the basis for judging purchasing power, the basis for predicting bargaining space, and the reference of similar historical cases, and to record decision nodes during the negotiation process for audit traceability.

Images