An AI agent-based service discovery transaction and credit evaluation method and system

Abstract

This invention discloses a service discovery, transaction, and credit evaluation method and system based on AI agents, belonging to the fields of artificial intelligence, e-commerce, and reservation services. The method proposes that a merchant's dedicated AI agent autonomously connects to the business's inventory management system to achieve intelligent operation; the AI ​​agent uses structured capability cards to achieve multimodal semantic discovery and proactive demand matching across categories (goods / services / reservations); AI agents from both buyers and sellers conduct multiple rounds of intelligent negotiation on complex terms such as price, inventory, and refunds / changes, and implement a tiered human confirmation control mechanism by identifying transaction risks and amounts; after performing multi-mode performance supervision for various transactions, a dual-separation credit evaluation based on tamper-proof interactive proof is generated, and trust transfer and recommendation are executed through social relationship graphs, achieving an efficient and secure decentralized automated transaction ecosystem through a credit scoring closed-loop mechanism.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, e-commerce, and online services. Specifically, it relates to a method and system for an AI agent to achieve cross-category publishing and semantic discovery of physical goods, reservation services, and professional services through structured capability declarations, enabling intelligent operation by relying on the merchant's AI agent to autonomously access the business system, completing terms negotiation through a multi-round negotiation mechanism, requiring human confirmation at key stages based on a tiered strategy, and establishing a credit system after the transaction is completed through multi-dimensional credit evaluation and social trust transfer. Background Technology

[0002] In social interactions and business activities involving both humans and AI agents, AI agents face the following technical challenges when acting on behalf of users to execute transactions such as purchasing goods and booking services:

[0003] First, traditional e-commerce and service platforms suffer from data silos and deep human involvement. In existing service and e-commerce platforms, product and booking information is locked within various apps, requiring users to manually search, compare, communicate, place orders, and track fulfillment across multiple platforms—a process entirely reliant on cumbersome human intervention.

[0004] Secondly, there is a lack of genuine business interaction and autonomous operation mechanisms between AI agents. Current technologies mostly remain at the level of simple API calls or one-way web scraping, leaving merchants without dedicated "autonomous AI agents" to represent them in business activities. An ideal merchant AI agent should be able to directly connect to the merchant's internal business systems such as ERP (Enterprise Resource Planning), proactively seeking customers, promoting products, and conducting multiple rounds of negotiations autonomously, just like a human salesperson. It should also be able to provide purchasing or replenishment suggestions to human merchants based on inventory dynamics.

[0005] Third, existing inter-agent communication lacks the ability to conduct multi-round negotiations and negotiate complex terms. Most existing inter-agent communication solutions are single-request-response scenarios, which cannot cope with the repeated bargaining and reservation management of complex terms such as price, delivery time, inventory lock-in, returns and exchanges, and liability for breach of contract in real business.

[0006] Fourth, transaction security lacks a tiered human verification mechanism. Existing systems are either completely automated, leading to financial risks, or require human intervention at every step, lacking a "tiered human verification lock" mechanism based on transaction amount, risk level, and historical trust relationships.

[0007] Fifth, the means of supervising contract performance are limited. There is a lack of a unified AI-assisted supervision and performance promotion mechanism for different types of transactions (such as physical goods requiring logistics tracking, reservation services requiring electronic voucher issuance and verification, and professional services requiring quality inspection of delivered goods).

[0008] Sixth, the evaluation system is disconnected from AI social networks. Existing systems cannot effectively prevent fraudulent orders and lack the function of using "friends' purchase records and reviews" in AI-assisted social relationship chains to transfer trust.

[0009] Seventh, the differences between this invention and existing technologies. Compared with traditional service-oriented architecture (SOA), the discovery mechanism of this invention supports cross-modal matching for complex natural language requirements; compared with web crawlers and price comparison software, the merchant AI agent of this invention has the ability to autonomously operate and actively sell by connecting to the inventory management system; compared with existing e-commerce, this invention supports a unified transaction framework for three types of assets: physical goods, reservations, and services, and prevents fraudulent transactions through interactive proofs, and expands credit from a single evaluation to a networked trust endorsement by utilizing social trust relationship chains.

[0010] The service discovery, transaction, and credit evaluation method described in this invention is applicable to various social interaction scenarios involving both humans and AI agents, and can be extended into a standard protocol for third-party merchants, ERP service providers, OTA platforms, and other users. Summary of the Invention

[0012] The technical problems to be solved by this invention are: how to enable merchant AI agents to deeply integrate with inventory management systems to achieve autonomous operation and intelligent sales; how to establish three standardized capability declarations and cross-category discovery mechanisms covering physical goods, reservation services, and professional services; how to complete the negotiation of complex transaction terms through multiple rounds of negotiation; how to design a hierarchical human confirmation mechanism that balances efficiency and security and implement multi-mode verification in contract performance; and how to establish a tamper-proof multi-dimensional credit evaluation system based on interactive proofs and alleviate the cold start of AI agents through trust transfer via social graphs.

[0013] Technical solution

[0014] Step S1, Structured Capability Declaration and Multi-Category Publication: The AI ​​agent packages the physical goods, reservation services, or professional consultations it represents into structured capability cards and publishes them to the HASN (Human and AI Social Network) service index. Capability cards include: category identifier, specifications / SKU set, pricing strategy, available inventory / time slots, logistics / refund / change conditions, etc. Capability cards are stored in a vectorized manner to support cross-modal semantic search.

[0015] Step S1.5, Merchant AI Agent Autonomous Access and Intelligent Operation: Merchants create a dedicated AI agent representing their business entity. This agent connects to the merchant's internal business systems, such as ERP and room management, through standard interfaces. The merchant AI agent acts like a human store manager, autonomously reading real-time inventory and prices, converting products into the aforementioned capability cards, and updating them to the network. Simultaneously, the merchant AI agent proactively identifies potential customers in the network, develops dynamic promotional strategies, answers customer inquiries, and autonomously assesses and notifies its human owner (merchant) when the inventory of a certain type of product falls below a threshold or shows a clear trend of high demand, prompting the agent to make purchasing or replenishment decisions.

[0016] Step S2, Multimodal Service Discovery and Proactive Demand Mining: This includes passive discovery and proactive mining. Passive discovery—Users express complex, cross-category needs verbally (e.g., "Arrange a business trip to Shanghai next Wednesday"). The user's AI agent performs semantic decomposition and simultaneously matches different candidate merchant agents (such as airline tickets and hotels) across the network. Proactive discovery—The buyer's AI agent continuously perceives the user's daily conversations, schedules, and consumption preferences, searching the network for matching products or services. When a high-quality match is found, it is recommended to the user in the form of a message summary. Search ranking comprehensively considers semantic matching degree, merchant AI agent credit score, and social trust transfer distance.

[0017] Step S3, Multi-round Intelligent Negotiation and Terms Consultation: The buyer's AI agent can initiate negotiations simultaneously with multiple merchant AI agents. Negotiation supports multiple rounds of interaction and intent game theory. For physical goods, the scope of negotiation includes price discounts, gifts, and delivery time; for booking services, it includes room availability locking (lock-in validity period) and cancellation / change penalty rates; for professional services, it includes the number of modifications and delivery / acceptance standards. When negotiations reach a stalemate beyond a preset number of rounds, an escalation mechanism is triggered, allowing the respective human owners to intervene and make a decision.

[0018] Step S4, Tiered Human Confirmation and Contract Generation: A tiered control strategy is introduced. If the transaction amount and risk assessment are below the user-set password-free threshold and the merchant's trust level is high, the buyer's AI agent can make the decision independently and only generate a notification report (automatic mode). If the amount is higher or it is the first transaction, the buyer's AI agent will submit the optimal negotiation result and multiple comparative drafts to the human owner, forcing the human to "confirm signing and payment" (single or double confirmation mode). After confirmation by both parties, a structured smart service contract is generated and stored in encrypted form.

[0019] Step S5, Multi-mode Execution Supervision and Performance Acceptance: A supervision engine is automatically matched based on the transaction category. For physical goods, the buyer's AI agent tracks the shipment via a logistics interface, and enters a post-sales observation period after receipt. For pre-booked services, the buyer's AI agent receives and verifies the electronic voucher record issued by the merchant's AI agent and provides a time-sensitive reminder before departure. For professional services, the buyer's AI agent automatically urges progress according to contract deadlines and performs basic compliance and duplication checks upon receipt of the delivered goods. All performance can ultimately be configured with a human final acceptance and confirmation step.

[0020] Step S6, Dual Separation of Evaluation and Interaction Proof: After the transaction is completed, the buyer's AI agent generates multi-dimensional evaluation records. The evaluation system adopts a dual separation mechanism of merchant AI agent service attitude response score and specific product / service quality score. The hash digests of key events during the transaction (such as inquiries, price changes, shipping fingerprints, and electronic voucher signing) are recorded on the blockchain or stored tamper-proofly as "interaction proofs." Only evaluations with genuine interaction proofs are included in the valid accounting pool. When calculating reputation, a time decay coefficient (the more recent the evaluation, the higher the weight) and the evaluator's historical reputation are weighted.

[0021] Step S7, Social Graph Trust Transfer and Credit Loop: Solving the Cold Start Problem for New Merchants. When a new merchant's AI agent lacks direct review records, the buyer's AI agent will penetrate the user's social relationship graph (such as "friends," "friends of friends," or "industry partners") to find genuine purchase and review records for the merchant's AI agent or similar products within the relationship chain. The system calculates a derived reference trust score based on the path hop decay coefficient of the social graph. Based on this, when a user browses a product, the AI ​​agent can directly prompt "Your friend X has purchased and given a positive review." Ultimately, the reputation score and the derived trust score work together in the search ranking of step S2, achieving a complete ecosystem loop.

[0022] Beneficial effects

[0023] First, it establishes the autonomous operating status of merchants' AI agents, supports direct integration with the inventory management system for proactive sales and inventory decisions, and completely revitalizes the intelligent ecosystem of the service supply side.

[0024] Secondly, a three-dimensional transaction standard has been established that is compatible with physical e-commerce, life reservations, and professional consultations, solving the problem that traditional technical architectures cannot negotiate cross-category combinations.

[0025] Third, it enables multi-round negotiation of terms that is closer to real business logic, and supports advanced mechanisms such as reserved inventory lock-in and dynamic pricing, which are not limited to simple price matching.

[0026] Fourth, the tiered human verification mechanism achieves the best balance between agent efficiency and fund security, avoids the legal risks of fully automated transactions, and reduces user disturbance.

[0027] Fifth, the multi-dimensional reputation system based on interactive proof eradicates the chronic problem of order-brushing on existing e-commerce platforms; the innovative social trust transfer scheme effectively solves the cold start problem for newly registered merchant agents and creates a highly sticky trust network. Attached Figure Description

[0029] Figure 1 A closed-loop diagram of the entire process of service transactions and credit evaluation. Figure 2 Merchant AI agent integration with inventory management and self-management architecture diagram. Figure 3 To support multi-round negotiation and tiered confirmation distribution maps across product categories. Figure 4 This is a flowchart for calculating the comprehensive credit score. Figure 5 This is a schematic diagram of trust transfer based on social graph chains. Figure 6 This is an overview diagram of the transaction and evaluation process of this invention. Detailed Implementation

[0031] Example 1: Merchant Agent Autonomous Access and Intelligent Management (Physical Product Listing and Proactive Replenishment Alerts)

[0032] The owner of a digital brand specialty store configured a merchant AI agent (Merchant Agent C) and granted it access to the store's ERP inventory management system API. Merchant Agent C can autonomously read the inventory quantity (500 units), cost price, and official suggested retail price of "latest noise-canceling headphones" in the warehouse without manual input from humans, and encapsulate it as a product-type capability card to register it in the HASN network.

[0033] In the lead-up to Singles' Day, Merchant Agent C, based on market demand analysis, independently developed a tiered price reduction strategy and updated its capability cards. As sales surged, when inventory dropped to 50 units, Merchant Agent C, based on historical shipping rates, determined that there was a risk of stockouts and proactively sent a pop-up notification to the human store manager's mobile device: "Dear Manager, we only have 50 units of the latest noise-canceling headphones left in stock. Our recent daily shipments have reached 20 units, making us highly susceptible to overselling and defaulting on orders. Please contact the supplier to replenish stock as soon as possible."

[0034] Example 2: Proactive Demand Discovery and Matching Across Categories by Buying and Selling Agents (Comprehensive Business Travel Booking Scenario)

[0035] User A says to their AI agent (buyer Agent A) via voice: "I'm going on a three-day business trip to the Guomao area of ​​Beijing next Wednesday." Buyer Agent A quickly performs semantic analysis and identifies two core needs: airfare and hotel.

[0036] Buyer Agent A proactively searched the HASN network, finding AI agents representing two different airlines and three merchant AI agents representing hotels near the China World Trade Center. Agent A initially sorted these merchant AI agents based on the buyer's historical travel preferences (early morning flights, four-star or higher hotels on Ctrip, and within a 15-minute walk of the meeting location) and the overall reputation scores of these agents. Among them, while a hotel merchant agent D was newly listed, Agent A discovered during its social graph scan that buyer A's colleague E had recently stayed there and given it a high rating for "excellent location and convenient transportation." Therefore, Agent A generated a summary and recommended this flight + hotel combination to user A.

[0037] Example 3: Multi-round complex negotiation and tiered human confirmation (refund and change terms and price negotiation)

[0038] Following Example 2, before placing the formal order, buyer Agent A and merchant Agent D, representing the hotel, conducted multiple rounds of negotiations.

[0039] Round 1: Agent A asks: "How much does it cost to book a standard room for three consecutive nights starting next Wednesday?" Merchant Agent D checks the room availability in the internal PMS and replies: "There are rooms available. The total price for three consecutive nights is 1500 yuan. It is non-cancellable."

[0040] Round Two: Agent A countersuggests: "The host's itinerary may change, so we need to include a free cancellation clause 24 hours before check-in. Can the price remain at 1500 yuan?" After calculating using its internal decision-making model, Agent D countersuggests: "Including the free cancellation clause would increase the price to 1650 yuan, or if full prepayment is made in advance, we can offer 1550 yuan."

[0041] In the third round, Agent A, based on the owner's usual requirements for cost-effectiveness, accepted the terms of "1550 yuan, full prepayment plus free cancellation", and thus an agreement was reached.

[0042] Subsequently, because the amount of 1550 yuan exceeded the "Agent automatic password-free payment threshold of less than 500 yuan" set by user A, the system triggered a tiered confirmation lock. Agent A submitted the agreed contract to user A, who clicked "Agree to payment and sign" on their mobile device to complete the final confirmation.

[0043] Example 4: Multi-mode performance monitoring and transaction closed loop (physical goods logistics tracking)

[0044] User A wants to buy a coffee machine. Buyer Agent A places an order and pays on behalf of User A through appliance retailer Agent F. Once the contract takes effect, the performance monitoring mechanism is automatically activated. Retailer Agent F notifies its warehouse to ship the goods and sends the tracking number and carrier's digital signature to Buyer Agent A. Buyer Agent A periodically checks the logistics interface for tracking information. After the package is delivered to the parcel locker, Agent A sends a message to User A to remind them to pick it up. Two days after the package is signed for, Agent A asks User A, "Are you using the coffee machine well? Do you need to confirm receipt and release the payment?" After User A confirms, the transaction payment is officially transferred to the retailer's account.

[0045] Example 5: Multidimensional Evaluation and Interactive Proof (Preventing False Prosperity)

[0046] After the coffee machine transaction was completed, Buyer Agent A generated a structured review for the seller. The review included: a merchant agent response speed score (4.9 / 5, based on the speed of responses in the logs) and a product quality score (4.5 / 5, derived from the buyer's subjective feedback). To validate the review, Buyer Agent A attached an encrypted "interaction proof"—containing a combination of "initial inquiry timestamp, negotiated price hash value, and actual logistics delivery tracking number." The HASN network verified the interaction proof, confirming it was a genuine transaction, before adding the review to Merchant Agent F's overall credit pool. The system effectively prevented automated order manipulation by assigning higher weightings to high-reputation reviewers (frequent shoppers with fair reviews) and introducing a decay function that degrades over time.

[0047] Example 6: Social Network Purchase Trends and Discovery Closed Loop (Trust Transfer Based on Social Graph)

[0048] A large number of users, such as A and E, have engaged in real transactions online, establishing a reliable interactive evaluation system. New user G recently wanted to buy a pair of noise-canceling headphones but was unsure which brand to choose. Buyer Agent G, by searching the HASN network and overlaying it with social graph insights, discovered that three AI agents representing G's colleagues in his department had recently ordered the same model of headphones from the aforementioned "Merchant Agent C," and all had ratings above 4.7. The system, through two levels of social distance attenuation (attenuation coefficient set to 0.8), assigned Merchant Agent C an extremely high "friends circle reference recommendation index (local high credibility)." Agent G directly suggested to the user: "Several of your colleagues recently purchased these headphones and had excellent feedback; should you place an order directly?" This mechanism leverages the real shopping chain of acquaintances and social circles to provide strong trust endorsement, accelerating the closed-loop decision-making of the transaction discovery engine.

Claims

1. A service discovery, transaction, and credit evaluation method based on AI agents, characterized in that, Includes the following steps: Step S1, Structured Capability Declaration and Multi-Category Release: The AI ​​agent packages the physical goods, booking services or professional consultations it represents into structured capability cards and releases them to the service index. The capability cards include category identifiers, specifications or SKUs, pricing strategies, available inventory or time slots, input and output parameters and delivery conditions. The capability cards are stored in a vectorized manner to support cross-modal semantic search. Step S1.5, Autonomous Operation and Marketing Integration of Merchant AI Agents in Social Networks: Human merchants entrust their physical goods, reserved stores, or professional skills to their created exclusive "Merchant AI Agents"; these Merchant AI Agents, as independent social nodes, are connected to the HASN hybrid social network and interface with the merchant's internal business systems such as inventory management or hotel management through standard interfaces; the Merchant AI Agents then gain environmental awareness and autonomous decision-making power, not only autonomously reading real-time inventory and price dynamic update capability cards, but also autonomously initiating replenishment warnings and approvals to human merchants based on inventory thresholds, and proactively discovering potential customers in the HASN social network through social interaction with buyer AI agents for anthropomorphic sales promotion and retention operations; Step S2: When a user expresses cross-category service or product demand in a hybrid social network using natural language, the buyer AI agent performs semantic decomposition of the demand, searches for the most similar merchant AI agent combination in the service index through semantic vector matching and passive discovery, and incorporates credit score and trust factor based on social topology association into the recommendation ranking of the search results. Step S3: The buyer's AI agent and the candidate merchant's AI agent initiate a negotiation session for multiple rounds of intelligent interaction, and engage in intentional game on the attributes of different product categories (including price discounts, delivery time, schedule lock-in, return and exchange, and breach of contract terms); the system maintains the negotiation status, and when the negotiation exceeds the preset number of rounds without reaching an agreement, the points of disagreement are escalated and submitted to the respective human users for decision. Step S4, Tiered Human Confirmation and Contract Generation: A tiered confirmation control strategy is implemented based on the transaction amount, risk assessment, and merchant trust level; for regular transactions that meet the conditions for small-amount password-free transactions, the buyer's AI agent autonomously facilitates the agreement and generates a notification report; for higher amounts or the first large transaction, the human user is forced to make the final confirmation of signing and payment; after confirmation by both parties, a structured smart service contract is generated. Step S5, Multi-mode execution supervision and performance acceptance: After the contract takes effect, the buyer's AI agent automatically matches the supervision engine according to the transaction category. For physical goods, it connects to the logistics interface to track the trajectory. For pre-booked services, it verifies the electronic vouchers issued by the merchant's AI agent. For professional services, it automatically urges progress according to nodes and reviews the delivered goods. The final result is given to the user for confirmation and acceptance. Step S6: After the transaction is completed, the buyer's AI agent generates a multi-dimensional evaluation record. This record performs a dual-separation scoring of the merchant's communication response attitude and the quality of specific goods or services. The system attaches the hash digest of the interaction events during the transaction as an interaction proof to the evaluation. Only valid evaluations verified by the interaction proof are included in the comprehensive credit score calculation. The calculation is weighted by the time decay coefficient and the evaluator's historical reputation. Step S7, Trust Transfer Based on Social Graph: When the direct review records of the evaluated merchant's AI agent are insufficient, the buyer's AI agent penetrates the user's social relationship graph, queries the real purchase and review records of contacts within the associated distance for the merchant's AI agent or the same product, applies a trust decay coefficient based on the number of path hops to obtain a derived reference trust score, and weights and fuses it with the direct review comprehensive credit score to obtain the final credibility score, which affects the search recommendation ranking closed loop in step S2.

2. The method according to claim 1, characterized in that, The buyer AI agent described in step S2 has a proactive service discovery mechanism: the buyer AI agent continuously senses the user's daily conversations, schedules and consumption preferences, actively searches for matching goods or services in the network, and recommends them to the user in the form of message summaries when a high-quality match is found.

3. The method according to claim 1, characterized in that, In step S3, the buyer's AI agent can simultaneously initiate cross-category combination negotiations with multiple candidate merchant AI agents, and generate a comprehensive recommendation plan based on the total price of the multi-category package, the comprehensive reputation score, and the coordination of delivery time among the merchants.

4. The method according to claim 1, characterized in that, Step S5 includes the following execution supervision for the booking service: the buyer's AI agent automatically receives the order confirmation information and generates a time constraint record, automatically sends a reminder to the user as the booking event approaches its deadline, and verifies the validity of the electronic voucher during the fulfillment stage.

5. The method according to claim 1, characterized in that, The interactive proof mechanism in step S6 stipulates that the interactive event sequence includes the negotiated price change hash value, the real logistics tracking number or the receipt fingerprint. The verifier determines the absolute authenticity of the transaction and service process by checking the digest hash.

6. The method according to claim 1, characterized in that, The trust transfer support based on social graph described in step S7 extends to purchase decision-making: when providing product search results, the buyer AI agent visualizes the purchase records and credit ratings of related contacts in the social graph, serving as a trust-assisted decision-making basis to facilitate the transaction.

7. A service discovery, transaction, and credit rating system based on AI agents, characterized in that, include: The multi-category capability registration module is used to manage the structured service claims published by AI agents, covering physical goods, reservation services, and professional consultations, and to maintain a service index library that supports semantic cross-modal search. The merchant system integration engine module enables merchants' dedicated AI agents to connect directly with internal business systems such as inventory management, sales and marketing, achieving autonomous inventory and price awareness and proactive marketing promotion. The semantic matching module is used to vectorize user needs to the capabilities of the merchant's AI agent and product information, and integrate credit scores and social connections for comprehensive recommendation and ranking. The intelligent negotiation engine module is used to support dynamic game and multi-round negotiation between AI agents regarding multi-dimensional terms such as price, delivery time, and cancellation / change. The hierarchical control and contract module is used to implement a hierarchical human authorization mechanism (automatic approval or mandatory confirmation) based on risk and amount assessment, and to generate structured smart contracts for performance. The multi-mode supervision module is used to adapt to the differentiated execution supervision mechanisms for different transaction categories, including logistics tracking, voucher verification and compliance checks of delivered goods; The credit evaluation and trust transfer module is used to collect two-dimensional evaluation data supported by tamper-proof interactive proof, and calculate derived trust scores based on the user's social communication graph according to the relationship distance to realize the evaluation closed loop.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images