An AI social broker supply and demand relationship accurate matching and intelligent referral method and system

Abstract

The application relates to an AI social broker supply-demand relationship accurate matching and intelligent referral method, which comprises the following steps: constructing a multi-dimensional user portrait containing explicit demand labels and implicit preference vectors according to target user static data, dynamic behavior data and historical interaction data, taking the target user as a center node, taking each candidate user as a candidate node, and taking the dynamic supply-demand matching degree between the target user and each candidate user as a connection edge weight to construct a supply-demand relationship network graph; generating a referral strategy based on the supply-demand relationship network graph through a social broker decision engine; and automatically sending a referral sentence in an authorized identity and establishing a referral connection after obtaining confirmation. The application realizes accurate identification of both supply and demand sides by constructing a multi-dimensional user portrait and calculating a dynamic supply-demand matching degree, sensing a matching degree change trend and judging a referral opportunity through a social broker decision engine, and automatically completing a referral connection in an authorized identity, thereby effectively improving the accuracy and efficiency of social matching.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for precise matching and intelligent referral of supply and demand relationships by an AI social broker. Background Technology

[0002] In recent years, AI-based online business social platforms have gradually emerged. These platforms aim to provide business users with a one-stop service for business opportunity discovery, networking, and knowledge acquisition. Existing similar platforms mostly adopt a function-centric technical architecture. Their core logic is that the platform acts as a passive information display and retrieval tool, providing users with discrete functions such as searching, filtering, and viewing. Users must actively operate various functional modules to obtain the information they need. Existing social matching solutions also largely rely on passively generated recommendation lists or maximized matching conversion values ​​based on single static data calculations, lacking awareness and responsiveness to the dynamic evolution of supply and demand relationships.

[0003] Chinese patent publication CN118820602A discloses a project planning and intelligent business intelligence push system based on AI and knowledge graphs. It includes: a data acquisition module for real-time industry data capture; a business intelligence analysis module, comprising a language processing unit, a knowledge graph construction unit, and a decision support unit; the language processing unit extracts key information from the data acquisition module and transmits it to the knowledge graph construction unit and decision support unit; the knowledge graph construction unit constructs a knowledge base and forms a knowledge graph based on the key information, which is then transmitted to the decision support unit; the decision support unit performs statistical analysis based on the results from the language processing unit and the knowledge graph construction unit to assist in making business decisions; and an intelligent push module, comprising a personalization unit and a push unit, for understanding user needs and pushing relevant information. This patented technology achieves intelligent push and decision support for business intelligence information by real-time industry data capture, knowledge graph construction, and calculation of the comprehensive value of text. However, its technical approach focuses on filtering high-value information content from massive amounts of text and pushing it to users. Essentially, it achieves information matching between industry information text and users, rather than social matchmaking between business entities with complementary supply and demand relationships. Specifically, the solution fails to identify users' core needs in finding funders, project owners, or potential partners, and it cannot quantify the complementarity between supply and demand sides or proactively initiate matchmaking.

[0004] It is evident that existing technologies lack precise matching and intelligent referral solutions for supply and demand relationships between business entities, failing to meet users' core needs for accurate discovery, proactive referral, and timely matchmaking of potential partners in business social scenarios. Summary of the Invention

[0005] To address these issues, this invention provides a method and system for precise matching and intelligent referral of supply and demand relationships using an AI social broker, thereby overcoming the problems in existing technologies where supply and demand matching relies on static information, cannot perceive dynamic changes in matching degree, lacks judgment on referral timing, and relies on manual operation for referral execution, resulting in low matching efficiency.

[0006] To achieve the above objectives, the present invention provides a method for precise matching and intelligent referral of supply and demand relationships using an AI social broker, the method comprising: Based on the acquired static data, dynamic behavior data, and historical interaction data of the target users, a multi-dimensional user profile is constructed, which includes explicit demand tags and implicit preference vectors. Using the target user as the central node and each candidate user as a candidate node, the dynamic supply and demand matching degree between the target user and each candidate user is comprehensively calculated based on the semantic similarity between the explicit demand tag and the supply tag of each candidate user, the affinity coefficient between the implicit preference vector and the interaction history of each candidate user, and the influence index of each candidate user in the supply and demand relationship network graph. The dynamic supply and demand matching degree is then used as the connection edge weight to construct a dynamically updated supply and demand relationship network graph. Based on the aforementioned supply and demand network diagram, a referral strategy is generated through a social broker decision engine. In response to a recommendation trigger, the system sends a referral message containing a description of the supply-demand match to the other party, using the authorized identity of either the target user or the candidate user, and establishes a referral connection after obtaining confirmation from both parties.

[0007] Furthermore, the method also includes: Obtain interaction result data after the referral connection is established. The interaction result data includes at least whether the referral request was accepted and the interaction behavior data after the referral connection was established. The referral success rate and subsequent interaction satisfaction are determined based on the interaction result data, and the referral success rate and subsequent interaction satisfaction are used as reward signals to input into the reinforcement learning framework to update the referral strategy of the social broker decision engine.

[0008] Furthermore, the engine strategy of the social broker decision engine is updated using a reinforcement learning framework, including: The current status is determined based on the matching degree of each candidate user at the current moment, the number of times they have been referred in the past, and the success rate of the historical referral. The current state is input into the policy network, which outputs a decision on whether each candidate user should initiate a referral, and then executes the output action. Calculate the reward value based on the feedback signal after the action is performed; The current state, the action, the reward value, and the state obtained at the next moment after performing the action are combined to form a state transition sample. The state transition samples at multiple consecutive moments are then combined according to the referral decision-making process to form trajectory data corresponding to the current strategy. When the trajectory data reaches the preset batch condition, the parameters of the policy network are updated based on the trajectory data using a near-end policy optimization algorithm; Repeat the above process until the preset convergence condition is met.

[0009] Furthermore, the generation of referral strategies based on the supply and demand network graph through the social broker decision engine includes: For any candidate user node in the supply and demand relationship network diagram, obtain the current dynamic supply and demand matching degree between the candidate user and the target user; When the current dynamic supply and demand matching degree is higher than the first threshold, the candidate user is marked as a referral candidate, and the matching degree time series data of the candidate user within a preset historical time window is obtained to calculate the matching degree change rate. A referral strategy is generated based on the rate of change of the matching degree and a preset rate of change threshold.

[0010] Further, the step of generating a referral strategy based on the matching degree change rate and a preset change rate threshold includes: When the rate of change of the matching degree is higher than the second threshold, the current moment is determined to be the optimal time for referral, and the referral action is triggered immediately. When the matching degree change rate is lower than the third threshold, the candidate user is stored in the monitoring queue, and when the matching degree change rate is higher than the second threshold, the candidate user is woken up from the monitoring queue and a referral action is triggered.

[0011] Furthermore, when there are multiple candidate users to be referred, the referral priority is determined based on the social distance between each candidate user and the target user and the corresponding rate of change in matching degree. For multiple candidate users to be recommended whose matching degree change rate is higher than the second threshold, priority is given to selecting candidate users with smaller social distance for recommendation; For candidate users with the same social distance, priority is given to those with a higher rate of change in match rate. In the supply and demand relationship network graph, nodes represent users participating in supply and demand matching, connecting edges represent supply and demand relationships between two users, the weight of connecting edges represents the dynamic supply and demand matching degree between two users, and the social distance is determined based on the shortest path length between the target user and the candidate user in the supply and demand relationship network graph.

[0012] Furthermore, the process of determining the semantic similarity includes: The explicit demand tags and the supply tags of each candidate user are mapped to corresponding nodes in the pre-built industry knowledge graph, and the paths between the corresponding nodes and the semantic relationship types on the paths are obtained. The supply and demand relationship between the explicit demand tag and the supply tag is determined based on the semantic relationship type, wherein the semantic relationship type includes at least complementary relationship, upstream and downstream relationship and similar competitive relationship; The corresponding semantic matching result is determined based on the supply and demand relationship, and the semantic similarity is determined based on the semantic matching result.

[0013] Furthermore, the update method for the dynamic supply and demand matching degree includes: Within a preset monitoring period, changes in the node data in the supply and demand relationship network diagram are detected; When the static attribute data or dynamic behavior data of any node changes, the node that has changed is identified as the changed node, and the set of adjacent nodes corresponding to the changed node is obtained. For each neighboring node in the set of neighboring nodes, only the dynamic supply and demand matching degree between the changed node and each neighboring node is recalculated, and the weight of the corresponding associated edge is updated using the recalculated dynamic supply and demand matching degree. For candidate nodes that have not been associated with the changed node, candidate relationship filtering is performed based on the updated user data of the changed node. When a candidate node that has not been associated meets the preset matching conditions, the dynamic supply and demand matching degree between the changed node and the corresponding candidate node is calculated, and a corresponding association edge is established based on the calculation result. For other nodes and associated edges that are not associated with the changed node, their corresponding dynamic supply and demand matching degree and edge weight remain unchanged to form a locally updated supply and demand relationship network graph.

[0014] Furthermore, the present invention also provides an AI social brokerage system for precise matching and intelligent referral of supply and demand relationships, the system comprising: The profile building module is used to construct a multi-dimensional user profile, including explicit demand tags and implicit preference vectors, based on the acquired static data, dynamic behavior data, and historical interaction data of the target user. The network graph construction module is used to comprehensively calculate the dynamic supply and demand matching degree between the target user and each candidate user, with the target user as the central node and each candidate user as the candidate node. Based on the semantic similarity between the explicit demand label and the supply label of each candidate user, the affinity coefficient between the implicit preference vector and the interaction history of each candidate user, and the influence index of each candidate user in the supply and demand relationship network graph, the module constructs a dynamically updated supply and demand relationship network graph using the dynamic supply and demand matching degree as the weight of the connection edge. The strategy generation module is used to generate referral strategies based on the supply and demand relationship network diagram through the social broker decision engine; The referral connection module is used to respond to a recommendation trigger by sending a referral message containing a description of the supply and demand match to the other party in the authorized identity of either the target user or the candidate user, and to establish a referral connection after obtaining confirmation from both parties.

[0015] Furthermore, the system also includes: An update module is used to obtain interaction result data after the referral connection is established. The interaction result data includes at least whether the referral request is accepted and the interaction behavior data after the referral connection is established. The referral success rate and subsequent interaction satisfaction are determined based on the interaction result data. The referral success rate and subsequent interaction satisfaction are used as reward signals and input into the reinforcement learning framework to update the referral strategy of the social broker decision engine.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention improves the accuracy of identifying complementary relationships between supply and demand by constructing a multi-dimensional user profile that includes explicit demand tags and implicit preference vectors, and by calculating dynamic supply and demand matching degree by integrating semantic similarity, affinity coefficient, and supply and demand relationship network graph influence index. It also achieves dynamic judgment of referral timing by generating referral strategies through a social broker decision engine based on the supply and demand relationship network graph. Furthermore, it achieves automated matching from matching decision to referral execution by automatically sending referral messages with authorized identity and establishing referral connections after confirmation. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the AI ​​social broker's method for precise matching and intelligent referral of supply and demand in this embodiment. Figure 2 A flowchart illustrating the process of generating the referral strategy for this embodiment; Figure 3 This is a flowchart illustrating the process of updating the referral strategy of the social broker decision engine in this embodiment. Figure 4 This is a schematic diagram of the supply and demand relationship precise matching and intelligent referral structure of the AI ​​social broker in this embodiment. Detailed Implementation

[0018] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0019] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0020] Please see Figure 1 The diagram shown is a flowchart illustrating the AI ​​social broker's method for precise matching and intelligent referral of supply and demand relationships in this embodiment. This embodiment specifically involves natural language processing, user profiling, and intelligent decision-making technologies, particularly a social broker intelligent referral system based on dynamic evolution of supply and demand relationships, applied in a supply and demand relationship network graph platform. This embodiment provides an AI social broker method for precise matching and intelligent referral of supply and demand relationships, comprising: Step S100: Based on the acquired static data, dynamic behavior data and historical interaction data of the target user, construct a multi-dimensional user profile that includes explicit demand tags and implicit preference vectors.

[0021] Before step S100 in this embodiment, static attribute data, dynamic behavior data, and historical interaction data of the target user are first acquired. Static data includes one or more combinations of the user's industry, job title, company type, and business scale. Dynamic behavior data includes one or more combinations of the user's browsing behavior, search behavior, business opportunity attention, and interaction frequency. Historical interaction data includes one or more combinations of the user's past dialogue records, cooperation project records, and social evaluations. Step S100 constructs a multi-dimensional user profile containing explicit demand tags and implicit preference vectors based on the target user's static data, dynamic behavior data, and historical interaction data. This overcomes the shortcomings of existing technologies that rely solely on single tags or static user attributes for matching. Explicit demand tags are keywords such as industry, role, and resource requirements actively set by the user, extracted through user registration information, personal homepage annotations, or cooperation demand information actively posted by the user. Implicit preference vectors are user interest representation vectors extracted from the user's historical behavior data using a preset preference inference model.

[0022] Step S200: Taking the target user as the central node and each candidate user as a candidate node, based on the semantic similarity between the explicit demand label and the supply label of each candidate user, the affinity coefficient between the implicit preference vector and the interaction history of each candidate user, and the influence index of each candidate user in the supply and demand relationship network graph, the dynamic supply and demand matching degree between the target user and each candidate user is comprehensively calculated, and the dynamic supply and demand matching degree is used as the weight of the connection edge to construct a dynamically updated supply and demand relationship network graph.

[0023] Step S200 in this embodiment has higher accuracy in identifying complementary relationships between supply and demand sides, and can effectively screen out candidate users with potential cooperation value. The weight coefficients of semantic similarity, interaction affinity coefficient, and influence index in the supply-demand relationship network graph are preset static hyperparameters. Their initial values ​​are set by those skilled in the art according to the actual application scenario, and can be manually adjusted by the administrator during offline iteration based on the statistical analysis results of historical referral data.

[0024] Step S300: Based on the supply and demand relationship network diagram, generate referral strategies through the social broker decision engine.

[0025] In step S300 of this embodiment, the social broker decision engine is a rule-based decision engine, which is used to output a referral decision on whether and when to initiate a referral for each candidate user based on the comparison result of the dynamic supply and demand matching degree of the candidate user with the preset threshold and the time series change rate of the dynamic supply and demand matching degree.

[0026] In this embodiment, the dynamic supply and demand matching degree is used to characterize the current supply and demand fit between the target user and the candidate user, and the matching degree change rate is used to characterize the changing trend of the supply and demand fit. The two respectively play the roles of candidate user admission judgment and referral timing judgment. Therefore, the decision engine does not only statically sort the candidate users according to the current matching degree, but further determines the referral timing according to the matching degree change trend after the candidate user meets the matching degree admission condition.

[0027] Step S400: In response to the recommendation trigger, the user, authorized by either the target user or the candidate user, sends a referral message containing a description of the supply and demand match to the other user, and establishes a referral connection after obtaining confirmation from both parties.

[0028] In step S400 of this embodiment, in response to the referral trigger signal, the authorized agent execution module, acting as the authorized identity of either the target user or the candidate user, matches the corresponding referral template from a preset referral template library based on the semantic relationship type between the target user's explicit demand tags and the candidate user's supply tags in the industry knowledge graph. This generates a personalized referral message containing an explanation of the supply-demand match and sends a referral request to the other party. After confirmation from both parties, a referral connection is established. This embodiment achieves a complete closed loop from referral decision to referral execution, eliminating the need for users to manually edit messages or actively initiate contact. This significantly shortens the time from supply-demand matching to establishing an effective connection, reducing the icebreaker cost for users during initial contact.

[0029] This embodiment, through the above steps S100 to S400, achieves precise supply and demand matching based on multi-dimensional user profiles, intelligent referral timing decision-making based on dynamic evolution of matching degree, automated referral execution based on authorized agents, and subsequent continuous optimization of strategies based on feedback learning. This effectively improves the accuracy of discovering potential partners, referral acceptance rate, and matchmaking efficiency in business social scenarios.

[0030] The specific contents of steps S100 to S400 are described below.

[0031] In one embodiment of the present invention, the implicit preference vector is based on user historical behavior data through a preset... The preference inference model is obtained by collecting user behavior sequences within a preset historical time window, including browsing, clicking, reading, and consultation events. These behavioral sequences are encoded into feature vector sequences and input into a deep interest network, i.e., the preference inference model. The historical behaviors are weighted and summed using an attention mechanism to output the user's latent preference vector. Specifically, the latent preference vector is constructed using steps S10 to S30.

[0032] Step S10: Collect the target user's behavior sequence within a preset historical time window. The behavior sequence includes... Events such as the user's browsing of business opportunity content, reading time, whether they clicked the "interested" button, and whether they initiated an inquiry.

[0033] Step S20: Encode each event in the behavior sequence into a feature vector. The feature vector includes features such as event type encoding, content category encoding, and timestamp.

[0034] Step S30: Input the feature vector sequence into the Deep Interest Network (DIN), and use the DIN model to perform attention aggregation on the historical behavioral features to obtain the interest representation vector of the target user. Use the interest representation vector as the latent preference vector for subsequent affinity coefficient calculation.

[0035] In this embodiment, the preference inference model uses a deep interest network, takes the user behavior sequence in historical behavior data as input, captures the user's interest intensity on different content through an attention mechanism, and outputs the user's implicit preference vector.

[0036] In step S200 of this embodiment, the target user is taken as the central node and each candidate user is taken as the candidate node. The dynamic supply and demand matching degree between the target user and each candidate user is calculated respectively. The dynamic supply and demand matching degree is used as the edge weight connecting the target user and the corresponding candidate user to construct a supply and demand relationship network graph centered on the target user. The nodes in the supply and demand relationship network graph are used to represent the target user and candidate user participating in the referral matching. The edges are used to represent the supply and demand relationship between two nodes. The weight value of the edge is the dynamic supply and demand matching degree.

[0037] In one embodiment of the present invention, the process of determining semantic similarity includes: mapping explicit demand tags and supply tags of each candidate user to corresponding nodes in a pre-constructed industry knowledge graph, obtaining the paths between corresponding nodes and the semantic relationship types on the paths; determining the supply and demand relationship between explicit demand tags and supply tags according to the semantic relationship types, wherein the semantic relationship types include at least complementary relationships, upstream and downstream relationships, and similar competitive relationships; determining the corresponding semantic matching results according to the supply and demand relationship, and determining semantic similarity based on the semantic matching results.

[0038] The affinity coefficient is determined based on the degree of matching between the target user's implicit preference vector and the interaction history of each candidate user. It is also calculated by combining the interaction frequency, interaction duration, and emotional polarity within a preset historical time window. Specifically, the interaction history of candidate users is encoded into interaction features according to the interaction object, interaction content category, and interaction behavior type. The similarity between the implicit preference vector and the interaction features is calculated, and the affinity coefficient is determined by combining the interaction frequency, interaction duration, and emotional polarity within a preset historical time window.

[0039] The influence index is used to characterize a candidate user's ability to reach supply and demand relationships within the supply and demand network graph. In this embodiment, the influence index is determined based on the candidate user's node degree, network centrality, and the range of users with whom they have established supply and demand relationships in the supply and demand network graph. Specifically, the richer the supply and demand relationships between a candidate user and other users, and the wider the range of supply and demand relationships they can connect to, the higher their influence index.

[0040] In step S200 of this embodiment, the dynamic supply-demand matching degree is calculated using semantic similarity, affinity coefficient, influence index, and corresponding weighted comprehensive calculations. The weights of semantic similarity, affinity coefficient, and influence index are all preset weight coefficients, whose values ​​are set during initialization to balance the relative importance of semantic similarity, affinity coefficient, and influence index in the calculation of dynamic supply-demand matching degree. Those skilled in the art can preset different initial values ​​according to actual application scenarios, and the weight coefficients can be manually adjusted by the administrator during offline iteration based on the statistical analysis results of historical referral data.

[0041] In this embodiment, explicit demand tags are used to calculate semantic similarity, enabling precise matching of demand and supply at the industry knowledge graph level; implicit preference vectors are used to calculate affinity coefficients, enabling precise capture of users' potential preferences; the combination of the two allows the matching degree calculation to take into account both the user's active demand expression and implicit preferences; this embodiment has higher accuracy in identifying complementary relationships between supply and demand, and can effectively screen out candidate users with potential cooperation value.

[0042] In one embodiment of the present invention, the dynamic supply and demand matching degree is updated using a local incremental method. Changes in the node data in the supply and demand relationship network diagram are detected according to a preset monitoring period, wherein the node data includes at least the static attribute data and dynamic behavior data of the user corresponding to the node.

[0043] When a change in the data of any node is detected, that node is marked as a changed node, and the set of adjacent nodes corresponding to the changed node is determined based on the currently existing connecting edges in the supply and demand relationship network graph. Subsequently, the input data for calculating the dynamic supply and demand matching degree is re-acquired only for the changed node and each adjacent node in the set of adjacent nodes, and the dynamic supply and demand matching degree between the changed node and each adjacent node is recalculated.

[0044] For candidate nodes that have not yet established a supply and demand relationship with the changed node, candidate relationships are filtered based on the updated static attribute data or dynamic behavior data of the changed node. When a candidate node without an established relationship meets a preset matching condition, the dynamic supply and demand matching degree is recalculated for the changed node and the corresponding candidate node. When the calculated dynamic supply and demand matching degree meets the preset edge establishment condition, an association edge between the changed node and the candidate node is established in the supply and demand relationship network graph, and the calculated dynamic supply and demand matching degree is used as the initial edge weight of the association edge.

[0045] For the recalculated dynamic supply and demand matching degree, update the edge weights of the corresponding connecting edges between the changed node and its adjacent nodes in the supply and demand relationship network graph; for other nodes and connecting edges that have not undergone data changes and are not affected by the changed node, do not recalculate their dynamic supply and demand matching degree, nor modify their original edge weights.

[0046] For example, when the dynamic behavior data of the target user changes, instead of recalculating the matching for all users in the entire supply and demand network graph, the candidate users with supply and demand relationships with the target user are first identified, and the dynamic supply and demand matching degree is recalculated only for the relationship edges between the target user and these candidate users. If there are 1000 candidate users for the target user, and this change only involves 20 candidate users with established supply and demand relationships, then this update only processes the corresponding 20 relationship edges, and the remaining relationship edges retain the matching degree and edge weight of the previous monitoring period.

[0047] By using the aforementioned local incremental update method, the entire supply and demand relationship network graph is avoided from being repeatedly calculated when user behavior changes frequently. This allows the supply and demand relationship network graph to be updated in a timely manner for local relationships that have actually changed, while reducing redundant calculations for user relationships that have not changed.

[0048] This embodiment adopts an incremental update method, recalculating the matching degree only for the edges involved in the nodes whose behavior has changed and updating the edge weights synchronously. While ensuring that the supply and demand relationship network graph reflects the latest status of users in real time, it significantly reduces the consumption of computing resources and improves the system's response efficiency in high-frequency behavior change scenarios.

[0049] In one embodiment of the present invention, such as Figure 2 As shown, the referral strategy is generated using the following steps S310 to S330.

[0050] Step S310: For any candidate user node in the supply and demand relationship network diagram, obtain the current dynamic supply and demand matching degree between the candidate user and the target user.

[0051] Step S320: When the current dynamic supply and demand matching degree is higher than the first threshold, the candidate user is marked as a referral candidate, and the matching degree time series data of the candidate user within the preset historical time window is obtained. The matching degree change rate is calculated. Specifically, the matching degree change rate is determined based on the difference in dynamic supply and demand matching degree between adjacent monitoring times within the preset historical time window and the corresponding time interval.

[0052] Step S330: Generate a referral strategy based on the matching degree change rate and the preset change rate threshold.

[0053] In this embodiment, when the current dynamic supply and demand matching degree is lower than or equal to the first threshold, it is determined that the candidate user does not currently meet the referral conditions, and no referral process is performed on him / her. The matching degree change of the candidate user continues to be monitored. In this embodiment, through steps S310 to S330, the candidate user admission is first determined by the dynamic supply and demand matching degree, and then the trend of matching degree change is determined by the matching degree change rate. Based on different trends, corresponding referral strategies are generated, thereby forming a progressive referral decision-making mechanism of candidate admission, trend judgment and strategy generation. This embodiment can distinguish different evolution states of matching degree such as rising, stable and falling, and take differentiated strategies such as immediate referral, continuous monitoring or monitoring and waiting accordingly, so as to achieve dynamic optimization of referral timing while ensuring the accuracy of referral.

[0054] In one embodiment of the present invention, step S330 uses the following method to generate a referral strategy.

[0055] When the rate of change of the matching degree is higher than the second threshold, the current moment is determined to be the optimal time for referral, and the referral action is triggered immediately. When the rate of change of the matching degree is lower than or equal to the second threshold, or higher than or equal to the third threshold, the referral will not be triggered for the time being. Instead, the system will continue to monitor the change of the matching degree of the candidate user and wait for the next round of judgment. When the matching degree change rate is lower than the third threshold, the candidate user is stored in the monitoring queue. During subsequent monitoring, when the matching degree change rate of the candidate user is higher than the second threshold, the candidate user is woken up from the monitoring queue and a referral action is triggered.

[0056] In this embodiment, candidate users in the monitoring queue do not participate in each round of complete referral decisions; instead, only the candidate user identifier, the most recent matching degree, the most recent matching degree change rate, and the corresponding change rate trigger status are retained. When a partial update occurs in the supply and demand relationship network graph, the matching relationship corresponding to the candidate user in the monitoring queue is checked first. When the updated matching degree change rate is higher than the second threshold again, the corresponding candidate user is removed from the monitoring queue and restored to the referral candidate set, thereby triggering the referral action.

[0057] In this embodiment, the specific values ​​of the first threshold, the second threshold, and the third threshold are preset by those skilled in the art based on actual application scenarios. The first threshold is used for admission judgment of dynamic supply-demand matching degree; the second threshold is used for judgment triggered by an upward trend in the matching degree change rate; and the third threshold is used for judgment of a downward or low trend in the matching degree change rate. The first threshold is used to filter candidate users whose current dynamic supply-demand matching degree meets the basic referral conditions, while the second and third thresholds are used to further judge the matching degree change trend of the candidate users. Together, these three constitute a complete referral timing decision-making system.

[0058] This embodiment differentiates candidate users by setting a matching score admission threshold, an upward trend trigger threshold, and a downward / low trend determination threshold, combining the absolute value of the matching score and the trend of its change rate. When the matching score change rate is higher than the second threshold, it is determined that the current trend is suitable for referral and a referral is triggered immediately. When the matching score change rate is between the second and third thresholds, a referral is not triggered temporarily, and monitoring continues. When the matching score change rate is lower than the third threshold, the candidate user is stored in a monitoring queue, and when the matching score change rate subsequently rises above the second threshold again, the candidate user is retrieved from the monitoring queue and a referral is triggered. Through this three-level judgment mechanism, the accuracy of referrals is ensured while dynamically optimizing the timing of referrals, avoiding matchmaking failures caused by referrals that are too early or too late.

[0059] In one embodiment of the present invention, when there are multiple candidate users to be referred, the referral priority is determined based on the social distance between each candidate user and the target user and the corresponding rate of change in matching degree. For multiple candidate users to be recommended whose matching degree change rate is higher than the second threshold, priority is given to selecting candidate users with smaller social distance for recommendation; For candidate users with the same social distance, priority is given to those with a higher rate of change in match rate. In the supply and demand network graph, nodes represent users participating in supply and demand matching, connecting edges represent supply and demand relationships between two users, the weight of the connecting edge represents the dynamic supply and demand matching degree between the two users, and the social distance is determined based on the shortest path length between the target user and the candidate user in the supply and demand network graph. Nodes in the supply and demand network graph represent users participating in supply and demand matching, and connecting edges represent established supply and demand relationships between users.

[0060] In this embodiment, when multiple referral candidates exist, the social broker decision engine sorts them based on the social distance between each candidate user and the target user in a pre-constructed supply-demand relationship network graph. The social distance is determined based on the shortest path length between user nodes in the supply-demand relationship network graph: if there is a direct connection between two user nodes, the social distance is 1; if a connection requires an intermediate node, the social distance is 2, and so on. A shorter shortest path length indicates a shorter connection path between the two users in the supply-demand relationship network graph, which can serve as an auxiliary criterion for referral priority. The decision engine prioritizes referencing candidate users whose social distance is less than a fourth threshold and whose dynamic supply-demand matching degree is higher than the first threshold. That is, only when candidate users meet the basic matching degree requirements is social distance further used as a priority ranking criterion, thereby reducing the cost of breaking the ice while ensuring matching quality.

[0061] This embodiment introduces social distance as a priority ranking criterion for referrals, prioritizing the referral of candidate users who are closer to the target user in the supply and demand relationship network graph. It leverages the trust transfer effect brought about by social proximity to reduce the cost of breaking the ice, increase the referral acceptance rate and the success rate of subsequent cooperation, and at the same time avoids over-recommending unfamiliar users with too far social distance, which could lead to referral rejection.

[0062] In step S400 of this embodiment, the industry knowledge graph is a directed graph structure pre-built and stored in a graph database, containing multiple nodes and directed edges connecting the nodes. Nodes include at least the following types: industry category nodes, sub-domain nodes, business type nodes, skill tag nodes, and resource type nodes; directed edges include at least the following types: hierarchical relationship edges, complementary relationship edges, upstream and downstream relationship edges, and similar competitive relationship edges. Industry category nodes include primary industry classifications such as artificial intelligence, new energy, and biomedicine; sub-sector nodes include secondary or tertiary sub-sectors such as machine learning, lithium battery materials, and gene sequencing; business type nodes include business processes such as technology research and development, manufacturing, and channel sales; skill tag nodes include specific ability tags such as Python development, supply chain management, and cross-border e-commerce operations; resource type nodes include resource categories such as capital, facilities, and patented technologies; hierarchical relationship edges indicate the hierarchical relationship between nodes, such as artificial intelligence and machine learning; complementary relationship edges indicate that nodes have complementary businesses, such as project owners and funders; upstream and downstream relationship edges indicate that nodes have upstream and downstream relationships in the industrial chain, such as chip design and chip manufacturing; and similar competitive relationship edges indicate that nodes have similar competitive businesses.

[0063] The method for constructing the industry knowledge graph is as follows: Industry classification standard documents and supply chain data are acquired; entities are extracted using named entity recognition technology; relationship extraction technology is used to determine the relationship types between entities; and entities are stored as nodes and relationships as directed edges in a graph database. The semantic relationship type query method is as follows: demand tags and supply tags are mapped to corresponding nodes in the industry knowledge graph; the shortest path between two nodes and the edge types along the path are queried in the graph; and the semantic relationship type between the demand tag and the supply tag is determined based on the edge type. Semantic relationship types include complementary relationships, upstream / downstream relationships, and competitive relationships.

[0064] When there are multiple candidate paths between demand tag nodes and supply tag nodes, candidate supply and demand relationships are determined based on the semantic relationship types contained in the paths, and the paths that can directly represent supply and demand relationships are prioritized for semantic matching. Among them, complementary relationships are used to represent the relationship between demanders and suppliers that can form complementary resources or capabilities, upstream and downstream relationships are used to represent the supply and demand relationship between different links in the industrial chain, and similar competitive relationships are used to represent users with similar business attributes but may have competitive relationships.

[0065] The referral template library is a collection of pre-stored referral templates, each associated with at least one semantic relationship type tag. Once the semantic relationship type between the demand tag and the supply tag is determined, a template matching the semantic relationship type is selected from the template library, and the identity information of both parties and the supply-demand matching points are filled into the replaceable fields in the template to generate a personalized referral.

[0066] In one embodiment of the present invention, after obtaining confirmation from both parties and establishing a referral connection in step S400... It also includes step S500: obtaining interaction result data after the referral connection is established, wherein the interaction result data includes at least whether the referral request is accepted and the interaction behavior data after the referral connection is established; determining the referral success rate and subsequent interaction satisfaction based on the interaction result data, and inputting the referral success rate and subsequent interaction satisfaction as reward signals into the reinforcement learning framework to update the referral strategy of the social broker decision engine.

[0067] In step S500 of this embodiment, a reinforcement learning framework is used to update the referral strategy of the social broker decision engine. Specifically, a strategy network is constructed, which uses the matching degree sequence and referral feedback in historical referral data as training samples, and the referral success rate and subsequent interaction satisfaction as reward signals. The parameters of the strategy network are updated through a proximal strategy optimization algorithm. The updated strategy network is used to optimize the referral strategy of the social broker decision engine, including suggestions for adjusting preset thresholds and / or suggestions for optimizing the timing of monitoring queue wake-up. Through the closed-loop feedback mechanism constructed in step S500 above, this embodiment continuously optimizes the referral timing decision in long-term operation, achieving an adaptive learning effect that becomes more accurate with use.

[0068] Specifically, the referral success rate is the ratio of the number of successfully established referral connections to the total number of referral requests initiated within a preset statistical time window. Interaction satisfaction characterizes the actual interaction quality between the two parties after the referral connection is established, and is determined based on the interaction frequency and user-initiated feedback ratings. Interaction frequency reflects whether effective interaction continues between the two parties, while user-initiated feedback ratings reflect their subjective evaluation of the referral result. When the interaction frequency reaches a preset condition and user-initiated feedback reaches a preset satisfaction level, the corresponding referral is considered to have high interaction satisfaction.

[0069] The reward signal is determined based on the success of the referral and the satisfaction of the interaction: a positive reward is given for a referral that successfully establishes a connection and whose interaction satisfaction meets the preset satisfaction criteria; a zero reward or a low positive reward is given for a referral that successfully establishes a connection but whose interaction satisfaction does not meet the preset satisfaction criteria; and a negative reward is given for a referral that fails to establish a connection.

[0070] In one embodiment of the present invention, such as Figure 3 As shown, the engine strategy of the social broker decision engine is updated using the following steps S510 to S560, including: Step S510: Determine the current status based on the dynamic supply and demand matching degree of each candidate user at the current moment, the number of times they have been referred in history, and the success rate of their referrals in history.

[0071] Step S520: Input the current state into the policy network, output the action decision of whether each candidate user initiates a recommendation, and execute the output action.

[0072] In step S520 of this embodiment, the policy network outputs the action probability distribution corresponding to each candidate user, and determines whether to initiate a referral action for each candidate user based on the sampling of the probability distribution. If a referral action is initiated, the referral action is executed and a referral request is sent to the selected user.

[0073] Step S530: Calculate the reward value based on the feedback signal after the action is executed.

[0074] In step S530 of this embodiment, feedback signals corresponding to each action are collected within a preset response time after the referral request is issued, and the total reward value is calculated. The feedback signals include: whether the referral request is accepted or rejected, and the interaction frequency and user evaluation tags within a preset observation time window after the referral connection is established. These are used to determine whether a single referral is successful to calculate the referral success rate, and to calculate interaction satisfaction, which is used to calculate the reward value.

[0075] Step S540: The current state, action, reward value, and next moment state are arranged into state transition samples according to the time sequence of action execution, and multiple state transition samples are organized into trajectory data corresponding to the current policy according to the continuous referral decision process.

[0076] Step S550: When the trajectory data corresponding to the current strategy reaches the preset batch condition, calculate the advantage function and strategy objective function required for strategy update based on the trajectory data, and update the parameters of the strategy network using the near-end strategy optimization algorithm.

[0077] In step S550 of this embodiment, when the trajectory data corresponding to the current strategy reaches the preset batch size, the advantage function and strategy objective function corresponding to each state transition sample are calculated based on the trajectory data, and the parameters of the strategy network are updated in batches using the near-end strategy optimization algorithm, so that the updated strategy network improves the expected cumulative reward.

[0078] Step S560: Repeat the above process until the preset convergence condition is met.

[0079] In step S560 of this embodiment, steps S510 to S550 are repeated until the referral success rate of the policy network on the validation set converges or reaches the preset iteration limit, thus completing the training and updating of the policy model.

[0080] This embodiment achieves continuous and progressive optimization of the referral strategy through the closed-loop process constructed in steps S510 to S560. The policy network adopts a multi-layer fully connected neural network structure, which includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The first hidden layer contains 128 nodes, the second hidden layer contains 64 nodes, the activation function of each hidden layer is the ReLU function, and the output layer uses the Softmax activation function.

[0081] like Figure 4 As shown, the present invention also provides an AI social brokerage system for precise matching and intelligent referral of supply and demand relationships, the system comprising: The profile building module 100 is used to build a multi-dimensional user profile containing explicit demand tags and implicit preference vectors based on the acquired static data, dynamic behavior data and historical interaction data of the target user.

[0082] In this embodiment, the profile construction module 100 performs the content of the above step S100. Static data includes one or more combinations of the user's industry, position, type of enterprise, and business scale; dynamic behavioral data includes one or more combinations of the user's browsing behavior, search behavior, business opportunity attention, and interaction frequency; historical interaction data includes one or more combinations of the user's past dialogue records, cooperation project records, and social evaluations.

[0083] The network graph construction module 200 uses the target user as the central node and each candidate user as a candidate node. Based on the semantic similarity between the explicit demand label and the supply label of each candidate user, the affinity coefficient between the implicit preference vector and the interaction history of each candidate user, and the influence index of each candidate user in the supply and demand relationship network graph, it comprehensively calculates the dynamic supply and demand matching degree between the target user and each candidate user, and uses the dynamic supply and demand matching degree as the weight of the connection edge to construct a dynamically updated supply and demand relationship network graph.

[0084] In this embodiment, the network graph construction module 200 executes the content of step S200 above. The dynamic supply and demand matching degree is calculated by comprehensively weighting the semantic similarity between the explicit demand label and the supply label of each candidate user, the affinity coefficient between the implicit preference vector and the interaction history of each candidate user, and the influence index of each candidate user in the supply and demand relationship network graph. The semantic similarity is calculated based on the semantic distance of the industry knowledge graph. The affinity coefficient is calculated based on the interaction frequency, interaction duration, and emotional polarity. The influence index is determined based on the node degree, network centrality, and the range of users who have established supply and demand relationships with the candidate user in the supply and demand relationship network graph. The weight coefficients corresponding to the semantic similarity, affinity coefficient, and influence index are preset hyperparameters.

[0085] The strategy generation module 300 is used to generate referral strategies based on the supply and demand relationship network graph through the social broker decision engine.

[0086] In this embodiment, the strategy generation module 300 executes the content of the above step S300. Through the rule-based social broker decision engine, it generates a referral strategy for each candidate user, including whether and when to initiate a referral, based on the comparison result of the dynamic supply and demand matching degree of the candidate user with the preset threshold and the time series change rate of the dynamic supply and demand matching degree.

[0087] The referral connection module 400 is used to respond to a recommendation trigger by sending a referral message containing a description of the supply and demand match to the other party in the authorized identity of either the target user or the candidate user, and to establish a referral connection after obtaining confirmation from both parties.

[0088] In this embodiment, the referral connection module 400 executes the content of step S400 above. The authorized identity is the proxy permission pre-granted by the target user or candidate user. The referral text is generated by matching the semantic relationship type between the target user's explicit demand tag and the candidate user's supply tag in the industry knowledge graph from the preset referral text template library. A personalized referral text containing a description of the supply and demand matching point is generated, and a referral request is sent to the other party in the authorized direction. After obtaining confirmation from both parties, a referral connection is established.

[0089] In this embodiment, the profile construction module 100 constructs standardized multi-dimensional digital profiles for target users and candidate users, enabling the system to perceive explicit needs and implicit preferences. Based on this, the network graph construction module 200 quantifies the supply and demand relationship between users into a dynamic matching degree and constructs an evolvable relationship graph. The strategy generation module 300, based on the matching degree and its changing trends in the graph, outputs the optimal referral timing and target selection through a preset rule engine. The referral connection module 400 automatically generates referral messages and establishes connections with the user's authorized identity. These four modules sequentially complete the entire chain from perception, quantification, decision-making to execution, jointly realizing the full-process automation of social brokers in supply and demand matching and intelligent referral.

[0090] In one embodiment of the present invention, the system further includes: an update module for obtaining referral links. The interaction result data after the establishment includes at least whether the referral request was accepted and the interaction behavior data after the referral connection was established. The referral success rate and subsequent interaction satisfaction are determined based on the interaction result data. The referral success rate and subsequent interaction satisfaction are used as reward signals and input into the reinforcement learning framework to update the referral strategy of the social broker decision engine.

[0091] In this embodiment, the update module performs the method described in step S500 during the update operation. Specifically, it acquires the interaction result data after the referral connection is established, determines the referral success rate and subsequent interaction satisfaction based on the interaction result data, and uses this as a reward signal to input into the reinforcement learning framework to batch update the policy network, thereby optimizing the referral strategy of the social broker decision engine. State transition samples are formed by combining the state, action, reward value, and the state at the next moment, and trajectory data is generated according to the continuous referral decision process. When the trajectory data reaches a preset batch condition, a near-end policy optimization algorithm is used to batch update the policy network, repeating the above process until a preset convergence condition is met. The referral success rate is the ratio of the number of referrals successfully established to the total number of referral requests initiated within a preset statistical time window. Interaction satisfaction is determined based on the interaction frequency and user evaluation tags after the referral connection is established. The reward signal is jointly determined by whether the referral is successful and whether the interaction satisfaction reaches a preset satisfaction condition.

[0092] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for precise matching and intelligent referral of supply and demand relationships using an AI social broker, characterized in that, The method includes: Based on the acquired static data, dynamic behavior data, and historical interaction data of the target users, a multi-dimensional user profile is constructed, which includes explicit demand tags and implicit preference vectors. Using the target user as the central node and each candidate user as a candidate node, the dynamic supply and demand matching degree between the target user and each candidate user is comprehensively calculated based on the semantic similarity between the explicit demand tag and the supply tag of each candidate user, the affinity coefficient between the implicit preference vector and the interaction history of each candidate user, and the influence index of each candidate user in the supply and demand relationship network graph. The dynamic supply and demand matching degree is then used as the connection edge weight to construct a dynamically updated supply and demand relationship network graph. Based on the aforementioned supply and demand network diagram, a referral strategy is generated through a social broker decision engine. In response to a recommendation trigger, the system sends a referral message containing a description of the supply-demand match to the other party, using the authorized identity of either the target user or the candidate user, and establishes a referral connection after obtaining confirmation from both parties.

2. The method according to claim 1, characterized in that, The method further includes: obtaining interaction result data after the referral connection is established, wherein the interaction result data includes at least whether the referral request is accepted and interaction behavior data after the referral connection is established; The referral success rate and subsequent interaction satisfaction are determined based on the interaction result data, and the referral success rate and subsequent interaction satisfaction are used as reward signals to input into the reinforcement learning framework to update the referral strategy of the social broker decision engine.

3. The method according to claim 2, characterized in that, The engine strategy of the social broker decision engine is updated using a reinforcement learning framework, including: The current status is determined based on the matching degree of each candidate user at the current moment, the number of times they have been referred in the past, and the success rate of the historical referral. The current state is input into the policy network, which outputs a decision on whether each candidate user should initiate a referral, and then executes the output action. Calculate the reward value based on the feedback signal after the action is performed; The current state, the action, the reward value, and the state obtained at the next moment after performing the action are combined to form a state transition sample. The state transition samples at multiple consecutive moments are then combined according to the referral decision-making process to form trajectory data corresponding to the current strategy. When the trajectory data reaches the preset batch condition, the parameters of the policy network are updated based on the trajectory data using a near-end policy optimization algorithm. Repeat the above process until the preset convergence condition is met.

4. The method according to claim 1, characterized in that, The process of generating referral strategies based on the supply and demand network diagram and through a social broker decision engine includes: For any candidate user node in the supply and demand relationship network diagram, obtain the current dynamic supply and demand matching degree between the candidate user and the target user; When the current dynamic supply and demand matching degree is higher than the first threshold, the candidate user is marked as a referral candidate, and the matching degree time series data of the candidate user within a preset historical time window is obtained to calculate the matching degree change rate. A referral strategy is generated based on the rate of change of the matching degree and a preset rate of change threshold.

5. The method according to claim 4, characterized in that, The step of generating a referral strategy based on the matching degree change rate and a preset change rate threshold includes: When the rate of change of the matching degree is higher than the second threshold, the current moment is determined to be the optimal time for referral, and the referral action is triggered immediately. When the matching degree change rate is lower than the third threshold, the candidate user is stored in the monitoring queue, and when the matching degree change rate is higher than the second threshold, the candidate user is woken up from the monitoring queue and a referral action is triggered.

6. The method according to claim 5, characterized in that, When there are multiple candidate users to be introduced, the introduction priority is determined based on the social distance between each candidate user and the target user and the corresponding rate of change in matching degree. For multiple candidate users whose matching degree change rate is higher than the second threshold, the candidate users to be recommended are determined in order of increasing social distance. For candidate users with the same social distance, the recommended candidate users are determined in descending order of the rate of change of matching degree; In the supply and demand relationship network graph, nodes represent users participating in supply and demand matching, connecting edges represent supply and demand relationships between two users, the weight of connecting edges represents the dynamic supply and demand matching degree between two users, and the social distance is determined based on the shortest path length between the target user and the candidate user in the supply and demand relationship network graph.

7. The method according to any one of claims 1 to 6, characterized in that, The process of determining the semantic similarity includes: The explicit demand tags and the supply tags of each candidate user are mapped to corresponding nodes in the pre-built industry knowledge graph, and the paths between the corresponding nodes and the semantic relationship types on the paths are obtained. The supply and demand relationship between the explicit demand tag and the supply tag is determined based on the semantic relationship type, wherein the semantic relationship type includes at least complementary relationship, upstream and downstream relationship and similar competitive relationship; The corresponding semantic matching result is determined based on the supply and demand relationship, and the semantic similarity is determined based on the semantic matching result.

8. The method according to any one of claims 1 to 6, characterized in that, The methods for updating the dynamic supply and demand matching degree include: Within a preset monitoring period, changes in the node data in the supply and demand relationship network diagram are detected; When the static attribute data or dynamic behavior data of any node changes, the node that has changed is identified as the changed node, and the set of adjacent nodes corresponding to the changed node is obtained. For each neighboring node in the set of neighboring nodes, only the dynamic supply and demand matching degree between the changed node and each neighboring node is recalculated, and the weight of the corresponding associated edge is updated using the recalculated dynamic supply and demand matching degree. For candidate nodes that have not been associated with the changed node, candidate relationship filtering is performed based on the updated user data of the changed node. When a candidate node that has not been associated meets the preset matching conditions, the dynamic supply and demand matching degree between the changed node and the corresponding candidate node is calculated, and a corresponding association edge is established based on the calculation result. For other nodes and associated edges that are not associated with the changed node, their corresponding dynamic supply and demand matching degree and edge weight remain unchanged to form a locally updated supply and demand relationship network graph.

9. A precise matching and intelligent referral system for supply and demand relationships using AI social brokerage, characterized in that, The system includes: The profile building module is used to construct a multi-dimensional user profile, including explicit demand tags and implicit preference vectors, based on the acquired static data, dynamic behavior data, and historical interaction data of the target user. The network graph construction module uses the target user as the central node and each candidate user as a candidate node. Based on the semantic similarity between the explicit demand label and the supply label of each candidate user, the affinity coefficient between the implicit preference vector and the interaction history of each candidate user, and the influence index of each candidate user in the supply and demand relationship network graph, it comprehensively calculates the dynamic supply and demand matching degree between the target user and each candidate user, and uses the dynamic supply and demand matching degree as the weight of the connection edge to construct a dynamically updated supply and demand relationship network graph. The strategy generation module is used to generate referral strategies based on the supply and demand relationship network diagram through the social broker decision engine; The referral connection module is used to respond to a recommendation trigger by sending a referral message containing a description of the supply and demand match to the other party in the authorized identity of either the target user or the candidate user, and to establish a referral connection after obtaining confirmation from both parties.

10. The system according to claim 9, characterized in that, The system also includes: An update module is used to obtain interaction result data after the referral connection is established. The interaction result data includes at least whether the referral request is accepted and the interaction behavior data after the referral connection is established. The referral success rate and subsequent interaction satisfaction are determined based on the interaction result data. The referral success rate and subsequent interaction satisfaction are used as reward signals and input into the reinforcement learning framework to update the referral strategy of the social broker decision engine.

Images