Intelligent content resource matching method and system for innovation and entrepreneurship guidance needs

Abstract

The present invention provides an intelligent content resource matching method and system for innovation and entrepreneurship coaching needs, relating to the field of natural language processing technology. First, a target user's demand interaction data set in an innovation and entrepreneurship coaching scenario is obtained, including multiple historical coaching session records and corresponding user feedback annotation information. Subsequently, through coaching demand feature extraction processing, the target user's demand semantic association features and resource adaptation feature set are obtained. Next, a multi-level strategy matching process is performed based on a preset strategy adaptation network to generate a resource adaptation strategy set for the target user. The resource adaptation strategy set is dynamically adjusted based on the user feedback annotation information to obtain an optimized resource adaptation strategy set. Finally, the optimized resource adaptation strategy set is sent to a resource scheduling engine to trigger a coaching resource matching operation, thereby realizing intelligent matching of innovation and entrepreneurship coaching resources and improving matching efficiency and accuracy.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing technology, and in particular to an intelligent content resource matching method and system for innovation and entrepreneurship coaching needs. Background Art

[0002] In today's rapidly evolving innovation and entrepreneurship landscape, target users (such as entrepreneurs and innovative teams) face diverse coaching needs. These demands span a wide range of areas, from project conception and market analysis to business model design and financing strategies. To meet these needs, a vast array of innovation and entrepreneurship coaching resources has emerged, including but not limited to expert consultations, online courses, and case studies. However, efficiently and accurately matching these resources with the actual needs of target users has become a pressing challenge.

[0003] Traditional resource matching methods for innovation and entrepreneurship mentoring often rely on manual judgment or simple keyword searches. This approach is not only inefficient but also struggles to accurately capture users' deeper needs and preferences. Furthermore, due to the complexity and dynamic nature of the innovation and entrepreneurship field, user needs may constantly adjust as projects progress and the market environment changes, requiring resource matching methods to be highly flexible and adaptable. Therefore, it is particularly important to develop a method that can intelligently analyze user needs and dynamically adjust resource matching strategies. Summary of the Invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides an intelligent content resource matching method for innovation and entrepreneurship guidance needs, the method comprising:

[0005] Obtaining a target user's demand interaction data set in an innovation and entrepreneurship coaching scenario, wherein the demand interaction data set includes multiple historical coaching session records and corresponding user feedback annotation information;

[0006] Performing tutoring demand feature extraction processing on the demand interaction data set to obtain the target user's demand semantic association features and resource adaptation feature set;

[0007] Performing multi-level policy matching processing on the demand semantic association features and the resource adaptation feature set based on a preset policy adaptation network to generate a resource adaptation policy set for the target user;

[0008] Performing dynamic policy adjustment processing on the resource adaptation policy set according to the user feedback annotation information to obtain an optimized resource adaptation policy set;

[0009] The optimized resource adaptation policy set is sent to the resource scheduling engine to trigger a coaching resource matching operation.

[0010] On the other hand, an embodiment of the present invention also provides an intelligent content resource matching system for innovation and entrepreneurship coaching needs, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0011] Based on the above aspects, the embodiments of the present invention achieve a deep understanding and precise matching of target user needs in the innovation and entrepreneurship coaching scenario by comprehensively using demand interaction data acquisition, coaching demand feature extraction, multi-level strategy matching processing, and dynamic strategy adjustment processing. It can not only efficiently extract user demand semantic association features and resource adaptation features from a large number of historical coaching session records, but also perform multi-level strategy matching based on a preset strategy adaptation network to generate a personalized resource adaptation strategy set, and can also dynamically adjust the resource adaptation strategy set according to user feedback annotation information to ensure the accuracy and timeliness of the matching results, significantly improve the matching efficiency and accuracy of innovation and entrepreneurship coaching resources, and provide users with more personalized and professional coaching services. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a schematic diagram of the execution flow of the intelligent content resource matching method for innovation and entrepreneurship guidance needs provided by an embodiment of the present invention.

[0013] Figure 2 Schematic diagram of exemplary hardware and software components of an intelligent content resource matching system for innovation and entrepreneurship coaching needs provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of an intelligent content resource matching method for innovation and entrepreneurship coaching needs provided by an embodiment of the present invention. The intelligent content resource matching method for innovation and entrepreneurship coaching needs is introduced in detail below.

[0015] Step S110: obtaining a target user's demand interaction data set in an innovation and entrepreneurship coaching scenario, where the demand interaction data set includes a plurality of historical coaching session records and corresponding user feedback annotation information.

[0016] In the actual scenario of Internet product innovation and entrepreneurship guidance, in order to accurately match intelligent content resources, it is first necessary to obtain a set of target user demand interaction data.

[0017] In today's digital internet startup landscape, various innovation and entrepreneurship coaching platforms have emerged. These platforms offer a wide variety of services, such as online courses, expert consultations, and project exchange forums. As target users utilize these services, they generate a large amount of interaction data with the platforms and other participants.

[0018] Historical coaching session records detail the conversations between target users, coaching experts, and other entrepreneurs. For example, on a coaching platform focused on internet social product startups, users might discuss product design features with experts, such as how to increase user interactivity and optimize social recommendation algorithms. They might also discuss marketing strategies, such as how to use social media for promotion and how to integrate offline promotional activities. These conversations are recorded in text format by the platform, forming historical coaching session records.

[0019] User feedback annotations represent users' evaluations and feedback on the tutoring services and resources they received. After completing a tutoring session or using a resource, the platform will guide them to complete a feedback questionnaire. This feedback may include satisfaction with the course content, an evaluation of the practicality of the expert advice, and the compatibility of the resource with their project. The platform will categorize and annotate this feedback, for example, by assigning different levels of satisfaction and providing a detailed description of the resource's practicality, to facilitate subsequent analysis.

[0020] There are multiple ways to obtain demand interaction data sets. By writing database query statements, you can directly extract the target user's historical coaching session records and corresponding user feedback annotation information from the platform's database. You can also obtain data by connecting with other related systems, such as connecting with industry information platforms to obtain the latest market trends and competitor information, further enriching the demand interaction data set. During the data acquisition process, it is necessary to attach great importance to data security and privacy protection. Collection and use must be carried out with user permission and authorization; for data involving user personal privacy, such as ID number and contact information, encryption algorithms are used to encrypt and process it to prevent data leakage. At the same time, a strict data access permission management system should be established, and only authorized personnel can access and process this data.

[0021] Step S120: performing coaching demand feature extraction processing on the demand interaction data set to obtain the target user's demand semantic association features and resource adaptation feature set.

[0022] After obtaining the demand interaction data set, the next step is to extract and process the coaching demand features.

[0023] Step S121: semantically decompose the demand text sequence in the historical tutoring session record to obtain multiple demand semantic units and corresponding semantic weight annotations, and perform context association analysis on each demand semantic unit to generate semantic association features and demand temporal features between the demand semantic units.

[0024] The demand text sequences in historical tutoring session records are the textual embodiment of user needs and usually contain multiple different semantic information. Semantic decomposition is to split the demand text sequences into multiple demand semantic units with independent semantics.

[0025] Step S1211: performing a co-occurrence relationship analysis on adjacent demand semantic units in the demand text sequence in the historical tutoring session record to determine the contextual co-occurrence frequency and semantic similarity parameters between the adjacent demand semantic units.

[0026] For example, in a coaching scenario for starting an internet e-commerce product startup, there's a requirement text sequence: "We want to develop an e-commerce app with personalized recommendations to enhance the user shopping experience. We also need to establish an efficient logistics and distribution system to ensure timely delivery of products. Furthermore, we need to develop an effective marketing and promotion strategy to attract more users to our app."

[0027] Therefore, this demand text sequence can be decomposed into multiple demand semantic units, such as "developing an e-commerce APP with personalized recommendation function", "improving user shopping experience", "establishing an efficient logistics distribution system", "ensuring that goods are delivered to users in a timely manner", "formulating effective marketing and promotion strategies", "attracting more users to use the APP", etc.

[0028] For adjacent requirement semantic units, a co-occurrence analysis is required. Assume that the set of requirement text sequences is all requirement text sequences. Traverse all requirement text sequences and count the number of sequences containing two adjacent requirement semantic units. This is denoted as the number of sequences containing a specific adjacent requirement semantic unit. The total number of sequences is the number of all requirement text sequences. The contextual co-occurrence frequency is then calculated as the number of sequences containing a specific adjacent requirement semantic unit divided by the number of all requirement text sequences.

[0029] The semantic similarity parameter is used to measure the degree of semantic similarity between two adjacent demand semantic units. It can be calculated using a word vector-based method. First, the words in each demand semantic unit are converted into corresponding word vectors. Suppose that demand semantic unit one contains a set of words, and demand semantic unit two contains a set of words. The word vectors of these words are averaged to obtain the vectors for demand semantic unit one and demand semantic unit two. Then, the cosine similarity of these two vectors is calculated as the semantic similarity parameter.

[0030] Step S1212: Calculate edge weight parameters between adjacent demand semantic units according to the context co-occurrence frequency and the semantic similarity parameter, wherein the edge weight parameter is positively correlated with the context co-occurrence frequency and the semantic similarity parameter.

[0031] After obtaining the contextual co-occurrence frequency and semantic similarity parameters between adjacent demand semantic units, the edge weight parameter is calculated. The edge weight parameter is used to represent the strength of the association between adjacent demand semantic units and is positively correlated with the contextual co-occurrence frequency and semantic similarity parameters.

[0032] For example, the edge weight parameter can be calculated by weighted summation, that is, the edge weight parameter is equal to the weight coefficient of the context co-occurrence frequency multiplied by the context co-occurrence frequency plus the weight coefficient of the semantic similarity parameter multiplied by the semantic similarity parameter, and the sum of the weight coefficient of the context co-occurrence frequency and the weight coefficient of the semantic similarity parameter is 1. The values ​​of the weight coefficient of the context co-occurrence frequency and the weight coefficient of the semantic similarity parameter can be adjusted according to actual conditions. For example, if more attention is paid to the impact of the context co-occurrence frequency on the association strength, the value of the weight coefficient of the context co-occurrence frequency can be appropriately increased; if more attention is paid to semantic similarity, the value of the weight coefficient of the semantic similarity parameter can be increased, where the context co-occurrence frequency and the semantic similarity parameters need to be standardized and converted.

[0033] Step S1213: Traverse all requirement semantic units in the requirement text sequence, take the requirement semantic units as nodes and construct edges according to the edge weight parameters to generate a temporal dependency graph, in which the nodes represent the requirement semantic units and the edges represent the contextual association strength between adjacent requirement semantic units.

[0034] After calculating the edge weight parameters between all adjacent requirement semantic units, we start to build the temporal dependency graph. We traverse all requirement semantic units in the requirement text sequence and treat each requirement semantic unit as a node.

[0035] Taking the demand text sequence of Internet e-commerce product entrepreneurship as an example, demand semantic units such as "developing an e-commerce APP with personalized recommendation function", "improving user shopping experience", "establishing an efficient logistics distribution system", "ensuring that goods are delivered to users in a timely manner", "formulating effective marketing and promotion strategies", and "attracting more users to use the APP" are used as nodes respectively.

[0036] Based on the calculated edge weight parameter, edges are constructed between nodes. For example, if the edge weight parameter between "Develop an e-commerce app with personalized recommendation functionality" and "Improve the user shopping experience" is a certain value, then an edge with that value is constructed between the two nodes. This forms a temporal dependency graph, where the node set is the set of demand semantic units, and the edge weights in the edge set are the edge weight parameters. This temporal dependency graph reflects the contextual association strength and temporal relationships between demand semantic units.

[0037] Step S1214: Determine the semantic weight labeling of the requirement semantic unit in the requirement text sequence according to the degree centrality parameter and edge weight parameter of each node in the temporal dependency graph.

[0038] Calculate the degree centrality parameter for each node in the time series dependency graph. The degree centrality parameter reflects the importance of a node in the graph and is the ratio of the number of direct connections between the current node and other nodes to the maximum number of possible connections in the graph. Let the number of direct connections of a node be the number of nodes connected to it, and the maximum number of possible connections in the graph be the maximum number of connections. Then, the degree centrality parameter of a node is the number of nodes connected to it divided by the maximum number of connections.

[0039] Perform weighted aggregation on all edge weight parameters corresponding to each node to obtain the aggregate weight value of each node. Here, the arithmetic mean of the edge weight parameters is used for weighted aggregation. Let the set of edge weight parameters corresponding to a node be multiple edge weight parameters, where the number is the number of edges connected to the node. Then the aggregate weight value of the node is the sum of these edge weight parameters divided by the number of edges connected to the node.

[0040] The degree centrality parameter and the aggregation weight value are normalized and then weighted summed to generate the initial weight coefficient of each node. First, the degree centrality parameter and the aggregation weight value are normalized separately. Let the normalized degree centrality parameter and the normalized aggregation weight value be , and the normalization method can adopt the common minimum-maximum normalization method. Then, let the weight coefficients of the weighted sum be the degree centrality parameter weight coefficient and the aggregation weight value weight coefficient, respectively, and the sum of the degree centrality parameter weight coefficient and the aggregation weight value weight coefficient be 1. Then, the initial weight coefficient of the node is the degree centrality parameter weight coefficient multiplied by the normalized degree centrality parameter plus the aggregation weight value weight coefficient multiplied by the normalized aggregation weight value.

[0041] Based on the timestamp information of each requirement semantic unit in the requirement text sequence, an exponential decay function is used to generate the time decay factor of the requirement semantic unit. Let the time when the requirement semantic unit is generated be the generation time, the current time be the present time, and the time interval be the present time minus the generation time. The exponential decay function can be expressed as an exponential form, where the decay coefficient is a set value and the time decay factor decreases as the generation time interval increases.

[0042] The initial weight coefficient is dynamically adjusted according to the time decay factor to generate the semantic weight annotation of the requirement semantic unit in the requirement text sequence. Let the semantic weight annotation of the node be the semantic weight value, then the semantic weight value is the initial weight coefficient multiplied by the time decay factor.

[0043] Step S1215: extract the local structural features and global propagation path features of each node in the temporal dependency graph, and concatenate the local structural features with the global propagation path features to generate semantic association features.

[0044] Traverse each node in the temporal dependency graph and obtain the set of adjacent nodes and the corresponding edge weight parameters for each node. For each node, its adjacent node set is denoted as the adjacent node set, and the corresponding edge weight parameter set is denoted as multiple edge weight parameters. Calculate the ratio of the node's current degree centrality parameter to the average degree centrality parameter of its adjacent nodes to generate the local degree centrality feature.

[0045] Extract the maximum, minimum, and variance of all edge weight parameters in the set of adjacent nodes to generate a local weight distribution feature. Let the maximum value of the edge weight parameter set be the maximum edge weight, the minimum value be the minimum edge weight, and the variance be the weight variance. Then the local weight distribution feature is the set containing the maximum edge weight, the minimum edge weight, and the weight variance.

[0046] The local degree centrality feature and the local weight distribution feature are vectorized and spliced ​​to generate the local structural features of the node.

[0047] Traverse all node pairs in the temporal dependency graph, calculate the shortest path length between each pair of nodes, and extract the number of shortest paths passing through the current node to generate the path betweenness parameter for the current node. Let the path betweenness parameter of a node be the ratio of the number of shortest paths passing through the node to the number of all shortest paths.

[0048] Based on the shortest path length and path betweenness parameter, the inverse of the average shortest path length from the current node to all other nodes is calculated to generate the global communication influence parameter. Let the average shortest path length from the node to all other nodes be the average shortest path length, and the global communication influence parameter be 1 divided by the average shortest path length.

[0049] The path betweenness parameter and the global communication influence parameter are weighted and fused to generate the node's global communication path characteristics. Assuming the weight coefficients of the weighted fusion are the path betweenness parameter weight coefficient and the global communication influence parameter weight coefficient, and the sum of the path betweenness parameter weight coefficient and the global communication influence parameter weight coefficient is 1, then the node's global communication path characteristics are the path betweenness parameter weight coefficient multiplied by the path betweenness parameter plus the global communication influence parameter weight coefficient multiplied by the global communication influence parameter.

[0050] The local structural features of the same node are dimensionally aligned with the global propagation path features, and then concatenated in order of feature dimensions to generate semantic association features.

[0051] Step S1216: performing time series window division processing on the demand semantic units in the demand text sequence, extracting the demand change frequency characteristics and demand priority characteristics of the demand semantic units based on the divided time series windows, and generating demand time series characteristics.

[0052] The semantic units in the requirement text sequence are arranged in chronological order, and then divided into time series windows. Let the size of the time series window be the window size, and the semantic unit sequence is divided into multiple subsequences with the length of the window size.

[0053] Within each time series window, the demand change frequency characteristics of the demand semantic units are calculated. The demand change frequency characteristics reflect the changes in the demand semantic units within a time series window. This can be expressed by calculating the ratio of the number of different demand semantic units to the total number of demand semantic units. Let the number of different demand semantic units within the time series window be the number of different demand semantic units, and the total number of demand semantic units be the total number. The demand change frequency characteristics are the number of different demand semantic units divided by the total number.

[0054] At the same time, within each time window, the demand priority feature of the demand semantic unit is determined. This can be determined based on the importance and urgency of the demand semantic unit within the demand text sequence. For example, a comprehensive assessment can be performed based on factors such as the degree of relevance of the demand semantic unit to the core entrepreneurial goal and the timeliness of the demand. Let the demand priority feature of the demand semantic unit be the demand priority value.

[0055] The demand change frequency feature and the demand priority feature are combined to generate the demand timing feature.

[0056] Step S122: performing feedback feature mapping processing on the user feedback annotation information to obtain the target user's demand preference characteristics and resource adaptation constraints.

[0057] User feedback annotation information contains rich information. Through feedback feature mapping processing, the target user's demand preference characteristics and resource adaptation constraints can be extracted from it.

[0058] Step S1221: extracting resource satisfaction scores, resource usage frequency parameters, and resource relevance annotations from user feedback annotation information.

[0059] In the context of internet product innovation and entrepreneurship coaching, user feedback annotations include evaluations and usage of various resources. Resource satisfaction ratings quantify user satisfaction with the resources they use, using a 1-5 star rating scale, with higher scores indicating greater satisfaction. Resource usage frequency parameters record how frequently users use a resource, such as the number of times per week. Resource relevance annotations indicate the degree of relevance between the resource and the user's entrepreneurial project, such as high, moderate, or low correlation.

[0060] Step S1222: constructing a resource preference weight matrix of the target user based on the resource satisfaction score and the resource usage frequency parameter, wherein the rows of the resource preference weight matrix represent resource categories and the columns represent resource attribute dimensions.

[0061] Divide resources into different categories, such as course resources, consulting service resources, tool resources, etc. At the same time, determine the attribute dimensions of resources, such as content quality, practicality, ease of use, etc.

[0062] Assume that the resource category set is a number of resource categories, and the resource attribute dimension set is a number of resource attribute dimensions. For each resource category and resource attribute dimension, determine its weight based on the resource satisfaction score and resource usage frequency parameter. Assume that the resource satisfaction score is the satisfaction value, and the resource usage frequency parameter is the usage frequency value. The weight of each position can be determined using a certain functional relationship. This constructs the resource preference weight matrix.

[0063] Step S1223: performing association path parsing on the resource association degree annotation to generate cross-category association path features of the target user between multiple resource categories.

[0064] Resource association annotations record the associations between different resource categories. These annotations are then processed for association path analysis to analyze the association paths between resource categories. For example, if user feedback indicates that course resources and consulting service resources are frequently used together and are significantly helpful for entrepreneurial projects, then it can be determined that an association path exists between these two resource categories.

[0065] By analyzing the associations between all resource categories, we generate cross-category association path features for target users across multiple resource categories. This can be represented as a graph, where nodes represent resource categories and edges represent association paths between resource categories. Edge weights are determined based on the strength of the associations.

[0066] Step S1224: perform independent feature encoding on the resource preference weight matrix and the cross-category association path feature respectively, concatenate the encoded feature vectors, and generate the demand preference feature; wherein, the independent feature encoding maps inputs of different dimensions to a unified dimension through a fully connected layer.

[0067] Perform independent feature encoding on the resource preference weight matrix. Input the resource preference weight matrix into a fully connected layer, which maps the matrix's different dimensional inputs to a unified dimension. Assuming the output dimension of the fully connected layer is a certain dimensional value, the eigenvector of the resource preference weight matrix obtained after encoding has the dimension of that dimensional value.

[0068] Independent feature encoding is also performed on cross-category association path features. The graph structure represented by the cross-category association path features is converted into a vector form and then input into another fully connected layer, which is also mapped to a feature vector with the dimension value of the dimension.

[0069] The encoded feature vectors are concatenated to generate demand preference features.

[0070] Step S1225: Generate resource adaptation constraints based on the resource conflict annotations and resource usage constraints in the user feedback annotation information.

[0071] User feedback annotations may include resource conflict annotations and resource usage constraints. Resource conflict annotations indicate that certain resources conflict with each other and cannot be used simultaneously or their combined effectiveness is poor. Resource usage constraints specify resource usage restrictions, such as usage time and usage permissions.

[0072] Based on this information, resource adaptation constraints are generated. These can be represented by a set of rules. For example, if resource 1 and resource 2 conflict, both resources cannot be selected simultaneously during resource matching. If resource 3 has a privileged user status, only target users who meet the privileged user status can use that resource.

[0073] Step S123: performing weighted aggregation processing on the semantic association features according to the semantic weight annotation to generate weighted semantic association features.

[0074] After obtaining the semantic association features and semantic weight annotations, the semantic association features are weighted and aggregated. Let the semantic association feature set be multiple semantic association features, and the corresponding semantic weight annotation set be multiple semantic weight annotations.

[0075] For each semantic association feature, multiply it by the corresponding semantic weight label, and then aggregate all the results. The aggregation method can be a summation method, that is, the weighted semantic association feature is the sum of all multiplication results.

[0076] Step S124: performing time series window splicing processing on the weighted semantic association features and the demand time series features to generate time series enhanced semantic features.

[0077] The weighted semantic association features and demand timing features are concatenated according to the timing windows. Let the dimension of the weighted semantic association features be the weighted dimension, the dimension of the demand timing features be the demand dimension, and the number of timing windows be the number of windows.

[0078] For each time series window, the weighted semantic association features and demand time series features within the window are spliced. Specifically, in a certain time series window, the weighted semantic association feature vector and the demand time series feature vector corresponding to the window are spliced ​​in the order of feature dimensions to obtain the spliced ​​feature vector of the time series window. The dimension of the spliced ​​feature vector is the sum of the weighted dimension and the demand dimension. By performing such a splicing operation on all time series windows, a time series enhanced semantic feature is generated. This time series enhanced semantic feature integrates semantic association information and time series information, and can more comprehensively reflect the demand characteristics of the target users at different time stages.

[0079] Step S125: performing preference screening processing on the temporal enhancement semantic features based on the demand preference features to generate demand semantic association features.

[0080] After obtaining the temporal enhanced semantic features, the demand preference features are used to perform preference screening. The demand preference features contain information such as the target user's preference tendencies and usage habits for different resources.

[0081] First, feature matching is performed between the temporal enhancement semantic features and the demand preference features. This can be achieved by calculating the similarity between each feature vector in the temporal enhancement semantic features and the demand preference feature vector. Let the temporal enhancement semantic feature vector set be multiple temporal enhancement semantic feature vectors, and let the demand preference feature vector be a specific demand preference feature vector. For each temporal enhancement semantic feature vector, the similarity between it and the demand preference feature vector is calculated. The similarity calculation method can be cosine similarity, for example.

[0082] Based on the calculated similarity, the temporal enhancement semantic feature vectors with high similarity to the demand preference feature are selected. A similarity threshold can be set. When the similarity between a temporal enhancement semantic feature vector and the demand preference feature vector exceeds the threshold, the vector is retained; otherwise, it is filtered out.

[0083] The retained temporal enhanced semantic feature vectors after screening are recombined to generate demand semantic association features. These demand semantic association features are more focused on the target user's demand preferences and can more accurately reflect the target user's core needs in innovation and entrepreneurship coaching scenarios.

[0084] Step S126: performing constraint matching processing on the demand semantic association features according to the resource adaptation constraint conditions to generate a resource adaptation feature set.

[0085] After obtaining the semantic association features of the requirements, the constraint matching process is performed based on the resource adaptation constraint conditions. The resource adaptation constraint conditions are a set of rules generated based on the resource conflict annotations and resource usage constraints in the user feedback annotation information.

[0086] Traverse each feature vector in the demand semantics association feature to check whether it satisfies the resource adaptation constraint. For example, if the resource adaptation constraint stipulates that certain resources cannot be used simultaneously, then in the demand semantics association feature, if a feature vector represents a demand involving these conflicting resources, the feature vector needs to be adjusted or filtered.

[0087] The feature vectors that meet the resource adaptation constraint conditions are retained and sorted to generate a resource adaptation feature set. The requirements represented by the feature vectors in the resource adaptation feature set are matched with the available resources.

[0088] Step S130: performing multi-level policy matching processing on the demand semantic association features and the resource adaptation feature set based on the preset policy adaptation network to generate a resource adaptation policy set for the target user.

[0089] The preset policy adaptation network is a network model specially designed for generating resource matching strategies. It consists of multiple layers, each with different functions. Through multi-level processing, it can achieve fine matching of demand semantic association features and resource adaptation feature sets.

[0090] Step S131: inputting the demand semantic association features into the demand parsing layer of the policy adaptation network to generate the core demand features and auxiliary demand features of the target user.

[0091] The demand parsing layer is the first layer of the strategy adaptation network. Its main function is to parse the semantic association features of the demand and extract the core demand features and auxiliary demand features of the target users.

[0092] The requirements parsing layer can use a neural network structure, such as a fully connected layer or a convolutional layer. The semantically related feature vectors of the requirements are input to the input nodes of the requirements parsing layer. After calculation and processing by the network, two feature vector sets are output: the core requirement feature set and the auxiliary requirement feature set.

[0093] Core demand characteristics represent the key needs of target users in innovation and entrepreneurship coaching scenarios. For example, for internet product startups, core needs may include the product's technical research and development direction and market positioning. Auxiliary demand characteristics supplement and support core needs, such as team management training and marketing channel expansion.

[0094] Step S132: inputting the resource adaptation feature set into the resource screening layer of the policy adaptation network to generate static resource attribute features and dynamic resource availability features of the candidate resource set.

[0095] After receiving the resource adaptation feature set, the resource screening layer is used to select candidate resources that match the target user's needs from the huge resource library.

[0096] In the context of internet product innovation and entrepreneurship coaching, the resource library may contain a rich variety of resources, including online courses, professional consulting services, and entrepreneurial tools and software. Each feature vector in the resource adaptation feature set represents a certain demand or preference of the target user for a resource.

[0097] First, the resource screening layer will conduct a detailed analysis of the resource adaptation feature set. It can identify the key information contained in the feature vector, such as resource type preferences (such as technical courses and market consulting), resource quality requirements (such as the qualification level of the teaching staff and the functional completeness of the tool software), etc.

[0098] The resource screening layer then compares this key information with the metadata of each resource in the resource library. Metadata describes the basic properties of a resource, including its name, type, description, and applicable objects. Through this comparison, resources that match the resource adaptation feature set are selected to form a candidate resource set.

[0099] Next, for the candidate resource set, the resource screening layer extracts its static resource attribute features and dynamic resource availability features respectively.

[0100] Static resource attribute characteristics are relatively fixed attributes of resources. For online course resources, static resource attribute characteristics may include the course subject (such as the application of artificial intelligence in internet products, internet marketing strategies), course duration, the background and qualifications of the instructor (such as whether they have work experience at well-known companies in the industry, whether they hold an advanced degree in a related major), the course content outline (the scope of knowledge covered, the depth of teaching), etc. For consulting service resources, static resource attribute characteristics may include the scale and reputation of the consulting organization, the professional field and experience of the consultant, the method of consulting services (such as face-to-face consultation, online video consultation), etc. For entrepreneurial tool software resources, static resource attribute characteristics may include the software's functional modules (such as project management, data analysis, user feedback collection), the difficulty of software operation, and the frequency of software updates.

[0101] To extract these static resource attribute features, the resource screening layer directly obtains relevant information from the resource's metadata and organizes it into feature vectors. For example, for course topics, text encoding can be used to convert them into numerical vectors; for instructor qualifications, they can be quantified based on pre-set levels.

[0102] Dynamic resource availability characteristics reflect the availability and usage of resources at different times and in different environments. For online course resources, dynamic resource availability characteristics may include the current number of course enrollments, remaining spots, and course start times. For consulting service resources, dynamic resource availability characteristics may include the consultant's appointment status and the time periods when services are available. For entrepreneurial tool software resources, dynamic resource availability characteristics may include the software's concurrent use limit and the current number of online users.

[0103] To extract dynamic resource availability features, the resource screening layer queries resource status information in real time. This may require data exchange with resource providers to obtain the latest resource usage information. For example, it can query the number of course registrations and remaining spots by invoking the online course platform's API; or query the consulting service agency's reservation system to find out the consultant's available appointment times. This dynamic information is organized into feature vectors, which, combined with static resource attribute features, form a complete feature representation of candidate resources.

[0104] Step S133: Call the policy matching layer of the policy adaptation network to perform multi-level attention matching processing on the core demand features, auxiliary demand features, static resource attribute features and dynamic resource availability features to generate a resource adaptation policy set; wherein, the multi-level attention matching processing includes demand level attention calculation, resource level attention calculation and cross-level policy fusion calculation.

[0105] The strategy matching layer can accurately match the target user's demand characteristics with the characteristics of candidate resources through multi-level attention matching processing, thereby generating an effective set of resource adaptation strategies.

[0106] Step S1331: In the demand-level attention calculation stage, the demand attention weight is generated according to the semantic correlation between the core demand feature and the auxiliary demand feature, and the auxiliary demand features are weightedly fused based on the demand attention weight to obtain the optimized core demand feature.

[0107] In the demand-level attention calculation stage, core demand features represent the target user's key needs in innovation and entrepreneurship coaching, while auxiliary demand features supplement and support the core needs. To better integrate this demand information, it is necessary to calculate the semantic relevance between them and generate demand attention weights based on this.

[0108] First, for each feature vector in the core and auxiliary requirement feature sets, a semantic similarity calculation method is used to determine the semantic association between them. Semantic similarity calculation can be based on the word embedding model, converting the text information in the feature vector into a numerical vector representation. The degree of semantic similarity is then measured by calculating the cosine similarity between the vectors. For example, the core requirement feature vector represents "user growth strategy for internet products" and the auxiliary requirement feature vector represents "social media marketing promotion methods." After converting these two feature vectors into numerical vectors using the word embedding model, their cosine similarity is calculated to obtain the semantic association value.

[0109] Assume that the core requirement feature vector set is a plurality of core requirement feature vectors, and the auxiliary requirement feature vector set is a plurality of auxiliary requirement feature vectors. For each core requirement feature vector and each auxiliary requirement feature vector, calculate the semantic association between them.

[0110] Next, demand attention weights are generated based on the semantic relevance. To ensure the rationality and comparability of the weights, a normalization method is used to convert the semantic relevance into weight values. Let the demand attention weight matrix be the weight matrix, where each element is the semantic relevance divided by the sum of all semantic relevances. Thus, each core demand feature vector corresponds to a demand attention weight vector, which represents the weight of the degree of association between each auxiliary demand feature vector and the core demand feature vector.

[0111] Finally, the auxiliary demand features are weighted and fused based on the demand attention weights. Each core demand feature vector is multiplied by the corresponding demand attention weight vector. All auxiliary demand feature vectors are then weighted and concatenated to produce the optimized core demand feature vector. This process of applying this approach to all core demand feature vectors yields an optimized core demand feature set that integrates information from both core and auxiliary demands, more accurately reflecting the needs of target users.

[0112] Step S1332: In the resource-level attention calculation stage, the static resource attribute features and the dynamic resource availability features are spliced ​​to obtain the resource comprehensive features, and the resource attention weight is generated according to the priority parameters of each resource attribute in the resource comprehensive features.

[0113] In the resource-level attention calculation stage, it is necessary to integrate static resource attribute features and dynamic resource availability features, and generate resource attention weights based on the priority parameters of resource attributes.

[0114] First, the static resource attribute feature vector and the dynamic resource availability feature vector are concatenated in the order of their feature dimensions to obtain a comprehensive resource feature vector. Let the static resource attribute feature vector be the static feature vector and the dynamic resource availability feature vector be the dynamic feature vector. The comprehensive resource feature vector is the concatenation of the static and dynamic feature vectors. This concatenation operation integrates the fixed attributes and dynamic state information of a resource, forming a more comprehensive representation of the resource's characteristics.

[0115] Next, determine the priority parameters for each resource attribute within the resource's comprehensive characteristics. These priority parameters are determined based on factors such as the resource's importance, scarcity, and the degree to which it satisfies the target user's needs. For example, for an internet product startup, the qualifications of the instructors teaching technical courses may be a key attribute, with a relatively high priority parameter; while the course duration may be less important, with a lower priority parameter.

[0116] Let the resource attribute set be multiple resource attributes, and the corresponding priority parameter set be multiple priority parameters. Generate resource attention weights based on the priority parameters. Similarly, normalization is used to convert the priority parameters into weight values. Let the resource attention weight vector be a weight vector, where each element is the priority parameter divided by the sum of all priority parameters. Thus, each resource attribute is associated with a resource attention weight that represents the importance of that attribute in resource matching.

[0117] Step S1333: In the cross-level strategy fusion calculation stage, the optimized core demand features and resource comprehensive features are projected separately to align the dimensions, and the projected features are subjected to cross-level feature interaction to generate a demand-resource interaction matrix. A resource adaptation strategy set is generated based on the matching parameters in the demand-resource interaction matrix.

[0118] In the cross-level strategy fusion calculation stage, it is necessary to effectively integrate the optimized core demand characteristics and resource comprehensive characteristics to generate a set of resource adaptation strategies.

[0119] First, feature projection is performed on the optimized core requirement feature vector and resource comprehensive feature vector. Since the optimized core requirement features and resource comprehensive features may have different dimensions, they need to be projected into the same dimensional space to facilitate feature interaction. Feature projection can be achieved using a fully connected layer. A fully connected layer is a neural network layer that multiplies the input feature vector by a set of weight matrices, adds a bias term, and then performs a nonlinear transformation using an activation function, outputting the projected feature vector.

[0120] Assume that the optimized core demand feature vector set is a plurality of optimized core demand feature vectors, and the resource comprehensive feature vector set is a plurality of resource comprehensive feature vectors. After feature projection processing, the projected core demand feature vector set and the projected resource comprehensive feature vector set are obtained.

[0121] The projected features are then subjected to cross-level feature interaction. The purpose of cross-level feature interaction is to calculate the degree of match between the optimized core demand features and the comprehensive resource features. This can be achieved by calculating the similarity between the projected core demand feature vector and the projected comprehensive resource feature vector. Let the demand-resource interaction matrix be a matrix, where each element represents the similarity between the projected core demand feature vector and the projected comprehensive resource feature vector. Similarity calculations can use methods such as cosine similarity and Euclidean distance.

[0122] Finally, a set of resource adaptation strategies is generated based on the matching parameters in the demand-resource interaction matrix. A matching threshold can be set. When an element in the demand-resource interaction matrix exceeds this threshold, it indicates a high degree of match between the core demand characteristics and the comprehensive resource characteristics. A resource adaptation strategy is generated, allocating the corresponding resource to the corresponding demand. By performing this judgment on all elements in the demand-resource interaction matrix, a complete set of resource adaptation strategies is generated. Furthermore, resource adaptation strategies can be sorted based on matching, prioritizing resources with high matching to improve the accuracy and effectiveness of resource matching.

[0123] In the above embodiment, the construction of the policy adaptation network includes several necessary modules: the demand analysis layer, the resource screening layer, and the policy matching layer. The demand analysis layer mainly analyzes and processes the input demand semantic association features. The demand analysis layer can adopt a multi-layer fully connected neural network structure. The input layer receives the demand semantic association feature vector, performs nonlinear transformation and feature extraction on the features through multiple hidden layers, and finally outputs the core demand features and auxiliary demand features. The number of neurons in the hidden layer can be adjusted according to actual needs. For example, it can be set to 128, 256, etc. The activation function can use the ReLU function, which can effectively alleviate the gradient vanishing problem and improve the training efficiency of the policy adaptation network.

[0124] The resource screening layer is responsible for screening and extracting the resource adaptation feature set, generating static resource attribute features and dynamic resource availability features for the candidate resource set. This layer can be implemented by combining a convolutional neural network (CNN) with a fully connected layer. The CNN portion is used to extract local features from the resource adaptation features. Parameters such as the convolution kernel size, number, and stride need to be optimized based on data characteristics. For example, the convolution kernel size can be set to 3x3, and the number of convolution kernels can be 64 or 128. After CNN processing, the feature vector is input into the fully connected layer for further feature fusion and transformation, outputting the required static and dynamic resource features.

[0125] The policy matching layer is the core component, performing multi-level attention matching. It consists of several submodules: demand-level attention calculation, resource-level attention calculation, and cross-level policy fusion calculation. Each submodule can be implemented using fully connected layers and attention mechanisms. The attention mechanism allows the model to focus more on important feature information, improving matching accuracy.

[0126] The hierarchical connection relationship of the policy adaptation network is as follows: the output of the demand analysis layer serves as the input of the demand-level attention calculation in the policy matching layer; the output of the resource screening layer serves as the input of the resource-level attention calculation in the policy matching layer; the results of the demand-level attention calculation and the resource-level attention calculation are jointly input into the cross-level policy fusion calculation module, and finally the resource adaptation policy set is output.

[0127] When training a strategy adaptation network, the first step is to prepare training data. This training data comes from a large collection of target users' demand interaction data in innovation and entrepreneurship coaching scenarios, including historical coaching session records and corresponding user feedback annotations. This training data is preprocessed through operations such as semantic decomposition and feature extraction to obtain the input data for training: a set of demand semantic association features and resource adaptation features, as well as the corresponding labeled data, representing the actual set of resource adaptation strategies.

[0128] The specific steps of training are as follows:

[0129] The first step is to initialize the model parameters. Random initial values ​​are assigned to the weights and biases of each layer of the policy adaptation network.

[0130] The second step is forward propagation. The training data is input into the policy adaptation network, which passes through the demand analysis layer, resource screening layer, and policy matching layer in sequence to obtain the resource adaptation policy set output by the policy adaptation network.

[0131] The third step is to calculate the loss. Use an appropriate loss function to measure the difference between the policy adaptation network output and the true label. The cross-entropy loss function can be used, which performs well in classification and matching problems.

[0132] The fourth step is backpropagation. Based on the gradient of the loss function, an optimization algorithm (such as stochastic gradient descent, Adam, etc.) is used to update the model parameters so that the value of the loss function gradually decreases.

[0133] Step 5: Repeat the above steps and continue iterating the training process until the loss function converges or the preset number of training rounds is reached.

[0134] Essential training parameters include the learning rate, number of training epochs, and batch size. The learning rate controls the step size for parameter updates. Excessively large learning rates can prevent the model from converging, while excessively small ones can slow down training. The number of training epochs determines the number of times the model is trained, while the batch size specifies the number of data samples used in each training session. These parameters need to be adjusted and optimized based on the specific dataset and model structure.

[0135] In the specific context of internet product innovation and entrepreneurship coaching, the integration of the strategic adaptation network and the scenario is reflected in the following aspects. Regarding input data, both the demand semantic association feature and the resource adaptation feature set are constructed around the specific needs and resource characteristics of internet product startups. The demand semantic association feature may include demand information on functional design, user experience, and marketing promotion for internet products; the resource adaptation feature set encompasses the characteristics of resources related to internet products, such as online courses, technical consulting, and market research.

[0136] The output data is a collection of resource adaptation strategies. These resource adaptation strategies directly target the needs of internet product startups and recommend appropriate coaching resources for target users. For example, for a startup team focused on developing social internet products, the strategy adaptation network might recommend resources such as online courses on social network algorithm optimization and consulting services for social product marketing.

[0137] Step S140: Dynamically adjust the resource adaptation policy set according to the user feedback annotation information to obtain an optimized resource adaptation policy set.

[0138] After generating a set of resource adaptation strategies, it is necessary to dynamically adjust them based on user feedback annotation information to improve the accuracy and effectiveness of resource matching.

[0139] Step S141: extracting resource matching accuracy, resource response timeliness, and resource usage conversion rate from user feedback annotation information.

[0140] In the context of internet product innovation and entrepreneurship coaching, user feedback annotations contain rich information about resource matching effectiveness. Resource matching accuracy, resource response timeliness, and resource utilization conversion rate are important indicators for evaluating the effectiveness of resource adaptation strategies.

[0141] To determine resource matching accuracy, we need to conduct a detailed analysis of user feedback regarding resource-demand matching. First, we need to identify the requirements for each resource allocation and then determine whether the resource truly meets those requirements. This can be accomplished by setting a series of matching rules. For example, for a technical consultation resource, if the consultation is highly relevant to the user's technical problem and provides an effective solution, the resource is considered a successful match; otherwise, the match fails. The ratio of successful matches to the total number of resource allocations is calculated, and the resource matching accuracy is the ratio of the two.

[0142] Extracting resource response timeliness relies on recording the time from resource request to resource provision. When a user requests a resource, the system records the request time; when the resource is actually provided to the user, the provisioning time is also recorded. The difference between these two time points is calculated to obtain the response time for each resource. Statistical analysis of the response times of all resources yields data related to resource response timeliness.

[0143] Determining resource usage conversion rates requires focusing on users' actual use of the allocated resources. Once resources are allocated to users, it's important to track whether and to what extent they actually use them. This can be determined by monitoring user behavior, such as whether they open online courses or use consulting services. The resource usage conversion rate is calculated by calculating the ratio of users using a resource to the number of users allocated the resource.

[0144] Step S142: Generate a first effect score based on the deviation value between the resource matching accuracy and the preset benchmark accuracy, generate a second effect score based on the difference between the resource response timeliness and the preset timeliness threshold, and generate a third effect score based on the ratio between the resource usage conversion rate and the historical average conversion rate.

[0145] The preset benchmark accuracy is a target value set based on historical data, industry standards, or experience, and is used to measure the accuracy of current resource matching. The deviation between the resource matching accuracy and the benchmark accuracy is calculated, that is, the absolute value of the difference between the resource matching accuracy and the benchmark accuracy.

[0146] To generate the first effect score, a monotonically decreasing function is used to reflect the relationship between the deviation value and the score. A linear function can be used. For example, the first effect score function is a linear function with pre-set coefficients. When the deviation value is 0, the resource matching accuracy reaches the baseline accuracy, and the first effect score reaches its maximum value. The larger the deviation value, the lower the first effect score.

[0147] The preset timeliness threshold is the upper limit of the specified resource response time and is used to measure the timeliness of resource responses. The difference between the resource response timeliness and the timeliness threshold is calculated as the resource response timeliness minus the timeliness threshold.

[0148] A monotonically decreasing function is also used to generate the second effect score. The second effect score function is assumed to be a linear function with coefficients that are pre-set values. When the resource response timeliness is less than or equal to the timeliness threshold, the difference is negative or 0, indicating a higher second effect score. A larger difference indicates a lower second effect score.

[0149] The historical average conversion rate is the average of the resource utilization conversion rates over a period of time, reflecting the historical efficiency of resource utilization. Calculate the ratio of the resource utilization conversion rate to the historical average conversion rate: the resource utilization conversion rate divided by the historical average conversion rate.

[0150] A monotonically increasing function is used to generate the third effect score. The third effect score function is assumed to be a linear function with a pre-set coefficient. When the ratio is greater than 1, it indicates that the current resource utilization conversion rate is higher than the historical average, and the third effect score is high; the smaller the ratio, the lower the third effect score.

[0151] Step S143: performing weighted summation on the first effect score, the second effect score, and the third effect score to generate a strategy effect score set.

[0152] In order to comprehensively consider the impact of resource matching accuracy, resource response timeliness and resource usage conversion rate on resource adaptation strategy, it is necessary to perform weighted summation of the first effect score, the second effect score and the third effect score.

[0153] Assume that the weighting coefficient for the first effect score is the first weighting coefficient, the weighting coefficient for the second effect score is the second weighting coefficient, and the weighting coefficient for the third effect score is the third weighting coefficient, and the sum of the first, second, and third weighting coefficients is 1. The values ​​of these weighting coefficients are determined based on the importance of each indicator in the overall evaluation. For example, if the accuracy of resource matching is more important, the value of the first weighting coefficient can be appropriately increased.

[0154] For each resource adaptation strategy, calculate its strategy effect score, that is, the first weighting coefficient multiplied by the first effect score plus the second weighting coefficient multiplied by the second effect score plus the third weighting coefficient multiplied by the third effect score. The strategy effect scores of all resource adaptation strategies are organized into a set to obtain a strategy effect score set.

[0155] Step S144: determining the adjustment priority and adjustment range parameters of each resource adaptation policy in the resource adaptation policy set according to the policy effect score set.

[0156] Sort the resource adaptation policies in the resource adaptation policy set based on the policy effectiveness score set. Resource adaptation policies with lower policy effectiveness scores have poor matching effectiveness and require priority adjustment. You can sort the policies in descending order, arranging them from lowest to highest effectiveness scores. Policies with lower scores are ranked higher and receive higher adjustment priority.

[0157] After determining the adjustment priority, the adjustment range parameter is further determined. The adjustment range parameter is determined based on the difference between the strategy effectiveness score and the average strategy effectiveness score. In other words, the absolute value of the difference between the strategy effectiveness score and the average strategy effectiveness score.

[0158] A monotonically increasing function is used to determine the adjustment amplitude parameter. Assume that the adjustment amplitude parameter function is a certain function form. The larger the gap, the larger the adjustment amplitude parameter, indicating that the strategy needs to be adjusted more significantly.

[0159] Step S145: iteratively optimize the policy parameters in the resource adaptation policy set based on the adjustment priority and adjustment amplitude parameters until the policy effect score set meets a preset convergence condition.

[0160] Iteratively optimize the policy parameters in the resource adaptation policy set according to the adjustment priority. In each iteration, the policy parameters are adjusted according to the adjustment amplitude parameter.

[0161] Parameters of a resource adaptation strategy may include resource allocation ratios, resource recommendation order, and resource-demand matching rules. For example, if a resource adaptation strategy allocates resources to demand, adjusting the strategy parameters may include adjusting the resource allocation quantity, changing the resource recommendation priority, and optimizing the resource-demand matching rules.

[0162] After each iteration, the policy effectiveness score set is recalculated. The policy effectiveness score set is checked to see if it meets the preset convergence criteria. Convergence criteria can be set in a variety of ways, such as when the change in the policy effectiveness score is less than a certain threshold, meaning the difference between the average values ​​of the policy effectiveness scores between two consecutive iterations is less than a pre-set decimal value; or when the number of iterations reaches a pre-set maximum value.

[0163] If the convergence condition is not met, the next iteration is continued, and the policy parameters are adjusted again according to the adjustment priority and adjustment amplitude parameters; if the convergence condition is met, the iteration is stopped.

[0164] Step S146: taking the policy parameter set that meets the convergence condition as the optimized resource adaptation policy set.

[0165] When the strategy effect score set meets the preset convergence conditions, the strategy parameter set becomes the optimized resource adaptation strategy set. The resource adaptation strategies in this set have undergone multiple adjustments and optimizations to more accurately meet the needs of target users, improving resource matching accuracy, response timeliness, and usage conversion rate, thereby enhancing the performance of the entire innovation and entrepreneurship coaching resource matching system.

[0166] Step S150: Send the optimized resource adaptation policy set to the resource scheduling engine to trigger the coaching resource matching operation.

[0167] After obtaining the optimized resource adaptation policy set, it is sent to the resource scheduling engine, which performs a matching operation on the coaching resources according to the policy set.

[0168] Step S151: parsing the resource matching rules and resource scheduling priorities in the optimized resource adaptation policy set.

[0169] The optimized resource adaptation policy set includes resource matching rules and resource scheduling priority information. Resource matching rules specify which resources should be allocated to which needs, while resource scheduling priorities determine the order in which resources are allocated.

[0170] Parse the optimized resource adaptation policy set to extract resource matching rules and resource scheduling priorities. This information can be stored in a data structure, such as a list or dictionary, for subsequent use.

[0171] Step S152: Send the resource matching rule to the corresponding resource service node according to the resource scheduling priority.

[0172] According to the resource scheduling priority, the resource matching rules are sent to the corresponding resource service nodes in the order of priority. Resource service nodes are entities that provide tutoring resources, such as online course platforms, consulting service agencies, etc.

[0173] For each resource matching rule, the corresponding resource service node is found and the resource matching rule is sent to the node. After receiving the resource matching rule, the resource service node will prepare and allocate resources according to the resource matching rule.

[0174] Step S153: Execute resource matching verification processing in the resource service node to obtain resource availability status and resource conflict detection results.

[0175] After receiving the resource matching rules, the resource service node performs resource matching verification. First, it checks the availability of the required resource to determine whether the resource exists and is available. For example, if the resource is an online course, it will check whether the course has started and whether there are any remaining seats.

[0176] At the same time, resource conflict detection is performed to check whether the current resource matching rules conflict with other rules being executed, such as whether the resource usage time conflicts or whether the resource quantity exceeds the limit.

[0177] Obtain resource availability status and resource conflict detection results, and feed these results back to the resource scheduling engine.

[0178] Step S154: When the resource conflict detection result meets the preset conflict resolution condition, a resource scheduling instruction is generated and sent to the resource execution terminal to complete the tutoring resource matching operation.

[0179] The default conflict resolution conditions are a set of pre-defined rules used to determine whether a resource conflict can be resolved. If the resource conflict detection results meet the conflict resolution conditions, it means that resources can be allocated appropriately to resolve the conflict.

[0180] At this point, the resource scheduling engine generates resource scheduling instructions based on resource matching rules and resource availability status. The resource scheduling instructions contain resource allocation information, such as resource type, quantity, and allocation object.

[0181] The resource scheduling instructions are sent to the resource execution terminal, which allocates and provides resources according to the instructions to complete the tutoring resource matching operation.

[0182] Step S155: If the conflict resolution condition is not met, a candidate resource set is regenerated based on a preset alternative resource matching rule, and a resource matching verification process is triggered.

[0183] If the resource conflict detection result does not meet the preset conflict resolution conditions, it means that the current resource matching rules have unresolvable conflicts. In this case, it is necessary to regenerate the candidate resource set based on the preset alternative resource matching rules.

[0184] The preset alternative resource matching rules are a set of pre-defined rules used to provide alternative resources when resource conflicts occur. Based on these rules, suitable alternative resources are screened from the resource library to generate a new set of candidate resources.

[0185] The new candidate resource set is sent to the resource service node again to trigger the resource matching verification process, and the above steps are repeated until the resource matching is successful or the preset upper limit of the number of attempts is reached.

[0186] Figure 2 A schematic diagram illustrates exemplary hardware and software components of an intelligent content resource matching system 100 for innovation and entrepreneurship coaching, provided in some embodiments of the present application, that can implement the concepts of the present application. For example, a processor 120 can be used in the intelligent content resource matching system 100 for innovation and entrepreneurship coaching and perform the functions of the present application.

[0187] The intelligent content resource matching system 100 for innovation and entrepreneurship guidance needs can be a general-purpose server or a special-purpose server, both of which can be used to implement the intelligent content resource matching method for innovation and entrepreneurship guidance needs of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0188] For example, the intelligent content resource matching system 100 for innovation and entrepreneurship counseling needs may include a network port 110 connected to the network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the intelligent content resource matching system 100 for innovation and entrepreneurship counseling needs may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application can be implemented according to these program instructions. The intelligent content resource matching system 100 for innovation and entrepreneurship counseling needs also includes an I / O interface 150 between the computer and other input and output devices.

[0189] For ease of explanation, only one processor is described in the intelligent content resource matching system 100 for innovation and entrepreneurship counseling needs. However, it should be noted that the intelligent content resource matching system 100 for innovation and entrepreneurship counseling needs in this application may also include multiple processors, so the steps performed by one processor described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the intelligent content resource matching system 100 for innovation and entrepreneurship counseling needs executes step A and step B, it should be understood that step A and step B may also be performed jointly by two different processors or individually in one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor execute steps A and B together.

[0190] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When a processor executes the computer-executable instructions, the above-mentioned intelligent content resource matching method for innovation and entrepreneurship coaching needs is implemented.

[0191] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. An intelligent content resource matching method for innovation and entrepreneurship guidance needs, characterized by: The method comprises: Obtaining a target user's demand interaction data set in an innovation and entrepreneurship coaching scenario, wherein the demand interaction data set includes multiple historical coaching session records and corresponding user feedback annotation information; Performing tutoring demand feature extraction processing on the demand interaction data set to obtain the target user's demand semantic association features and resource adaptation feature set; Performing multi-level policy matching processing on the demand semantic association features and the resource adaptation feature set based on a preset policy adaptation network to generate a resource adaptation policy set for the target user; Performing dynamic policy adjustment processing on the resource adaptation policy set according to the user feedback annotation information to obtain an optimized resource adaptation policy set; Sending the optimized resource adaptation policy set to the resource scheduling engine to trigger the tutoring resource matching operation; The extracting and processing of the demand interaction data set to obtain the target user's demand semantic association features and resource adaptation feature set includes: Performing semantic decomposition processing on the demand text sequence in the historical tutoring session record to obtain multiple demand semantic units and corresponding semantic weight annotations, and performing context association analysis processing on each demand semantic unit to generate semantic association features and demand temporal features between the demand semantic units; Performing feedback feature mapping processing on the user feedback annotation information to obtain the demand preference characteristics and resource adaptation constraints of the target user; Performing weighted aggregation processing on the semantic association features according to the semantic weight annotation to generate weighted semantic association features; Performing time-series window splicing processing on the weighted semantic association feature and the demand time-series feature to generate a time-series enhanced semantic feature; Performing preference screening on the temporal enhancement semantic feature based on the demand preference feature to generate the demand semantic association feature; The resource adaptation feature set is generated by performing constraint matching processing on the demand semantic association features according to the resource adaptation constraint conditions.

2. The intelligent content resource matching method for innovation and entrepreneurship guidance needs according to claim 1 is characterized in that: The semantic decomposition processing is performed on the demand text sequence in the historical tutoring session record to obtain multiple demand semantic units and corresponding semantic weight annotations, and the context association analysis processing is performed on each demand semantic unit to generate semantic association features and demand time sequence features between the demand semantic units, including: Performing a co-occurrence relationship analysis on adjacent demand semantic units in the demand text sequence in the historical tutoring session record to determine the contextual co-occurrence frequency and semantic similarity parameters between the adjacent demand semantic units; Calculating an edge weight parameter between the adjacent demand semantic units according to the context co-occurrence frequency and the semantic similarity parameter, wherein the edge weight parameter is positively correlated with the context co-occurrence frequency and the semantic similarity parameter; Traversing all the requirement semantic units in the requirement text sequence, taking the requirement semantic units as nodes and constructing edges according to the edge weight parameters, to generate a temporal dependency graph, wherein the nodes of the temporal dependency graph represent the requirement semantic units, and the edges represent the contextual association strength between adjacent requirement semantic units; Determining the semantic weight labeling of the requirement semantic unit in the requirement text sequence according to the degree centrality parameter and the edge weight parameter of each node in the temporal dependency graph; Extracting local structural features and global propagation path features of each node in the temporal dependency graph, and splicing the local structural features with the global propagation path features to generate the semantic association features; The demand semantic units in the demand text sequence are divided into time series windows, and the demand change frequency characteristics and demand priority characteristics of the demand semantic units are extracted based on the divided time series windows to generate the demand time series characteristics.

3. The intelligent content resource matching method for innovation and entrepreneurship guidance needs according to claim 2 is characterized in that: The determining, based on the degree centrality parameter and the edge weight parameter of each node in the temporal dependency graph, the semantic weight annotation of the requirement semantic unit in the requirement text sequence includes: Calculating a degree centrality parameter for each node in the timing dependency graph, where the degree centrality parameter is a ratio of the number of direct connection edges between the current node and other nodes to the maximum possible number of connection edges in the graph; Performing weighted aggregation processing on all edge weight parameters corresponding to each node to obtain an aggregated weight value of each node, wherein the weighted aggregation processing adopts the arithmetic mean of the edge weight parameters; Normalizing the degree centrality parameter and the aggregation weight value and performing weighted summation to generate an initial weight coefficient for each node; Based on the timestamp information of each demand semantic unit in the demand text sequence, an exponential decay function is used to generate a time decay factor of the demand semantic unit, wherein the time decay factor decreases as the generation time interval increases; The initial weight coefficient is dynamically adjusted according to the time decay factor to generate a semantic weight annotation of the requirement semantic unit in the requirement text sequence.

4. The intelligent content resource matching method for innovation and entrepreneurship guidance needs according to claim 2 is characterized in that: The extracting of local structural features and global propagation path features of each node in the temporal dependency graph, and splicing the local structural features with the global propagation path features to generate the semantic association features includes: Traversing each node in the temporal dependency graph, obtaining the adjacent node set and corresponding edge weight parameters of each node, calculating the ratio of the current degree centrality parameter of the node to the average degree centrality parameter of the adjacent nodes, and generating a local degree centrality feature; Extracting the maximum value, minimum value and variance of all edge weight parameters in the adjacent node set to generate a local weight distribution feature; Vectorizing and concatenating the local degree centrality feature and the local weight distribution feature to generate a local structural feature of the node; Traversing all node pairs in the timing dependency graph, calculating the shortest path length between each pair of nodes, extracting the number of shortest paths passing through the current node, and generating a path betweenness parameter of the current node; Calculate the inverse of the average shortest path length from the current node to all other nodes based on the shortest path length and the path betweenness parameter to generate a global communication influence parameter; Performing weighted fusion on the path intermediacy parameter and the global propagation influence parameter to generate a global propagation path feature of the node; The local structural features of the same node are dimensionally aligned with the global propagation path features, and are spliced ​​in the order of feature dimensions to generate the semantic association features.

5. The intelligent content resource matching method for innovation and entrepreneurship guidance needs according to claim 1 is characterized in that: The performing feedback feature mapping processing on the user feedback annotation information to obtain the target user's demand preference characteristics and resource adaptation constraint conditions includes: Extracting resource satisfaction scores, resource usage frequency parameters, and resource relevance annotations from the user feedback annotation information; Constructing a resource preference weight matrix of the target user according to the resource satisfaction score and the resource usage frequency parameter, wherein the rows of the resource preference weight matrix represent resource categories and the columns represent resource attribute dimensions; Performing association path parsing on the resource association degree annotation to generate cross-category association path features of the target user between multiple resource categories; Independent feature encoding is performed on the resource preference weight matrix and the cross-category association path feature, and the encoded feature vectors are concatenated to generate the demand preference feature; wherein the independent feature encoding maps inputs of different dimensions to a unified dimension through a fully connected layer; The resource adaptation constraint condition is generated based on the resource conflict annotation and the resource usage constraint condition in the user feedback annotation information.

6. The intelligent content resource matching method for innovation and entrepreneurship guidance needs according to claim 1 is characterized in that: The preset policy adaptation network performs multi-level policy matching processing on the demand semantic association feature and the resource adaptation feature set to generate the resource adaptation policy set for the target user, including: Inputting the demand semantic association features into the demand parsing layer of the strategy adaptation network to generate the core demand features and auxiliary demand features of the target user; Inputting the resource adaptation feature set into the resource screening layer of the policy adaptation network to generate static resource attribute features and dynamic resource availability features of the candidate resource set; Calling the policy matching layer of the policy adaptation network to perform multi-level attention matching processing on the core demand characteristics, the auxiliary demand characteristics, the static resource attribute characteristics, and the dynamic resource availability characteristics to generate the resource adaptation policy set; wherein the multi-level attention matching processing includes demand-level attention calculation, resource-level attention calculation, and cross-level policy fusion calculation, specifically including: In the demand-level attention calculation stage, a demand attention weight is generated according to the semantic association between the core demand feature and the auxiliary demand feature, and the auxiliary demand features are weightedly fused based on the demand attention weight to obtain the optimized core demand feature; In the resource-level attention calculation stage, the static resource attribute features and the dynamic resource availability features are concatenated to obtain a comprehensive resource feature, and a resource attention weight is generated according to the priority parameters of each resource attribute in the comprehensive resource feature; In the cross-level strategy fusion calculation stage, the optimized core demand features and the resource comprehensive features are respectively subjected to feature projection processing to align dimensions, and the projected features are subjected to cross-level feature interaction to generate a demand-resource interaction matrix. The resource adaptation strategy set is generated according to the matching parameters in the demand-resource interaction matrix.

7. The intelligent content resource matching method for innovation and entrepreneurship guidance needs according to claim 1 is characterized in that: The dynamically adjusting the resource adaptation strategy set according to the user feedback annotation information to obtain an optimized resource adaptation strategy set includes: Extracting resource matching accuracy, resource response timeliness, and resource utilization conversion rate from the user feedback annotation information; A first effect score is generated based on a deviation between the resource matching accuracy and a preset benchmark accuracy, a second effect score is generated based on a difference between the resource response timeliness and a preset timeliness threshold, and a third effect score is generated based on a ratio between the resource usage conversion rate and a historical average conversion rate; Performing a weighted summation on the first effect score, the second effect score, and the third effect score to generate a strategy effect score set; Determining the adjustment priority and adjustment range parameters of each resource adaptation strategy in the resource adaptation strategy set according to the strategy effect score set; Iteratively optimizing the policy parameters in the resource adaptation policy set based on the adjustment priority and the adjustment amplitude parameter until the policy effect score set meets a preset convergence condition; The policy parameter set that meets the convergence condition is used as the optimized resource adaptation policy set.

8. The intelligent content resource matching method for innovation and entrepreneurship guidance needs according to claim 1 is characterized in that: The step of sending the optimized resource adaptation policy set to the resource scheduling engine to trigger the tutoring resource matching operation includes: Analyzing resource matching rules and resource scheduling priorities in the optimized resource adaptation strategy set; Sending the resource matching rule to the corresponding resource service node according to the resource scheduling priority; Performing resource matching verification processing in the resource service node to obtain resource availability status and resource conflict detection results; When the resource conflict detection result meets the preset conflict resolution condition, a resource scheduling instruction is generated and sent to the resource execution terminal to complete the tutoring resource matching operation; If the conflict resolution condition is not met, a candidate resource set is regenerated based on a preset alternative resource matching rule, and a resource matching verification process is triggered.

9. An intelligent content resource matching system for innovation and entrepreneurship guidance needs, characterized by: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the intelligent content resource matching method for innovation and entrepreneurship guidance needs as described in any one of claims 1 to 8 above.

Images