Resource management method, resource recall method, device and electronic equipment

Abstract

This disclosure provides resource management methods, resource retrieval methods, devices, and electronic devices, relating to the field of computer technology, particularly artificial intelligence, natural language processing, large-scale models, and deep learning. The specific implementation scheme is as follows: Analyze the resource content used for resource binding to obtain entity information; based on the entity information, select multiple candidate resource nodes from multiple resource nodes in the resource tree; wherein each resource node has corresponding resource attribute information and hierarchical information; using the first large-scale language model (LLM), based on the entity information and the resource attribute and hierarchical information of each candidate resource node, determine the resource node to be bound from the multiple candidate resource nodes; establish a binding relationship between the resource content and the resource node to be bound to manage the resource content.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to the fields of artificial intelligence, natural language processing, large models, and deep learning, and especially to a resource management method, a resource retrieval method, a device, and an electronic device. Background Technology

[0002] In some application scenarios, such as intelligent customer service systems, knowledge graph construction, and big data analysis, there is often a need for resource retrieval. Resource retrieval refers to accurately and quickly identifying resources that meet specific conditions from a massive resource database to meet business processing or user query needs.

[0003] However, effective resource management is crucial before implementing resource recall operations. Due to the large quantity and diverse types of resources, without a scientific and reasonable management mechanism, resources will be scattered and disorganized. In this state, resources lack necessary connections and identification, making it difficult to quickly locate target resources during resource recall. Furthermore, information confusion can easily lead to recall errors, significantly reducing recall efficiency and accuracy.

[0004] Effective resource management makes the resource structure clearer and the relationships between resources more explicit and specific. This helps to quickly and accurately locate target resources during resource retrieval, thereby improving the efficiency and accuracy of resource retrieval and providing strong support for the efficient operation of various application scenarios. Summary of the Invention

[0005] This disclosure provides a resource management method, a resource recall method, an apparatus, and an electronic device.

[0006] According to one aspect of this disclosure, a resource management method is provided, the method comprising: analyzing resource content for resource binding to obtain entity information; selecting multiple candidate resource nodes from multiple resource nodes in a resource tree based on the entity information; wherein the resource nodes have corresponding resource attribute information and hierarchical information; using a first large language model LLM, determining a resource node to be bound from the multiple candidate resource nodes based on the entity information and the resource attribute information and hierarchical information of each candidate resource node; and establishing a binding relationship between the resource content and the resource node to be bound, so as to manage the resource content.

[0007] According to another aspect of this disclosure, a resource retrieval method is provided, the method comprising: parsing a resource retrieval task to obtain key resource information of the resource to be retrieved; selecting multiple reference resource nodes from multiple resource nodes in a resource tree based on the key resource information; wherein the resource nodes have corresponding resource attribute information and hierarchical information, and are pre-bound to corresponding resource content, the binding relationship being established using the method of one aspect of this disclosure; using a Large Language Model (LLM), based on the key resource information, and the resource node names and hierarchies of each of the reference resource nodes, determining the target resource node from the multiple reference resource nodes as the resource to be retrieved; and retrieving the resource content corresponding to the target resource node.

[0008] According to another aspect of this disclosure, a resource management device is provided, the device comprising: an analysis module for analyzing resource content for resource binding to obtain entity information; a selection module for selecting multiple candidate resource nodes from multiple resource nodes in a resource tree based on the entity information; wherein the resource nodes have corresponding resource attribute information and hierarchical information; a first determination module for determining a resource node to be bound from the multiple candidate resource nodes using a first large language model (LLM) based on the entity information, and the resource attribute information and hierarchical information of each candidate resource node; and an establishment module for establishing a binding relationship between the resource content and the resource node to be bound, so as to manage the resource content.

[0009] According to another aspect of this disclosure, a resource retrieval device is provided, the device comprising: a parsing module, configured to parse a resource retrieval task to obtain key resource information of the resource to be recalled; a selection module, configured to select multiple reference resource nodes from multiple resource nodes in a resource tree based on the key resource information; wherein the resource nodes have corresponding resource attribute information and hierarchical information, and are pre-bound to corresponding resource content, the binding relationship being established using a method of one aspect of this disclosure; a first determination module, configured to use a Large Language Model (LLM) to determine a target resource node from the multiple reference resource nodes as the resource to be recalled, based on the key resource information and the resource attribute information and hierarchical information of each of the reference resource nodes; and a retrieval module, configured to perform resource retrieval on the resource content corresponding to the target resource node.

[0010] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the resource management method or resource retrieval method proposed above in this disclosure.

[0011] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided that stores computer instructions for causing a computer to execute the resource management method or resource retrieval method proposed in this disclosure above.

[0012] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the resource management method or resource retrieval method proposed above.

[0013] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0014] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0015] Figure 1 This is a schematic diagram based on the first embodiment of the present disclosure; Figure 2 This is a schematic diagram according to the second embodiment of the present disclosure; Figure 3 This is a schematic diagram according to the third embodiment of the present disclosure; Figure 4 This is a schematic diagram according to the fourth embodiment of the present disclosure; Figure 5 This is a schematic diagram according to the fifth embodiment of the present disclosure; Figure 6 This is a schematic diagram according to the sixth embodiment of the present disclosure; Figure 7 This is a block diagram of an electronic device used to implement the resource management method or resource recall method of the embodiments of this disclosure. Detailed Implementation

[0016] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0017] When implementing systematic management of various resources, in order to improve resource utilization efficiency and synergistic effects, it is usually necessary to deeply associate and organically integrate resources that are scattered and isolated. Specifically, this involves using specific organizational forms to tightly combine resources with inherent connections, constructing a structured resource set. This process involves the application of relevant technical means, such as rule-based fuzzy matching methods and vector model schemes.

[0018] Among them, rule-based fuzzy matching methods mainly rely on a manually constructed grammatical rule system and a pre-defined synonym replacement library. This method has certain applicability in simple scenarios, but it has obvious shortcomings. Its flexibility is poor, and it is difficult to adapt to complex and ever-changing contexts. In particular, when dealing with texts with ambiguous semantics and diverse expressions, the accuracy and recall of matching are difficult to reach ideal levels.

[0019] The vector model approach maps the entire sentence into a low-dimensional and dense vector space, measuring semantic similarity by the distance between vectors, thus distributing semantically similar sentences closer together in the vector space. However, this distance-based method of measuring semantic similarity is not absolutely reliable. Its results are easily affected by various factors such as sentence length, semantic complexity, and contextual relevance, leading to biased judgments of similarity.

[0020] To address at least one of the aforementioned problems, this disclosure proposes a resource management method, a resource recall method, an apparatus, and an electronic device.

[0021] This disclosure proposes a resource management method, a resource recall method, an apparatus, and an electronic device.

[0022] Figure 1 This is a schematic diagram based on the first embodiment of the present disclosure. It should be noted that the resource management method of the present disclosure can be applied to a resource management device, which can be configured in an electronic device so that the electronic device can perform resource management functions.

[0023] Among them, electronic devices can be any device with computing capabilities, such as personal computers (PCs), mobile terminals, servers, etc. Mobile terminals can be, for example, in-vehicle devices, mobile phones, tablets, personal digital assistants, wearable devices, smart speakers, servers, server clusters, and other hardware devices with various operating systems, touch screens and / or displays.

[0024] The resource management device can also be software within an electronic device, such as resource management software. In the following embodiments, an electronic device is used as an example for illustration.

[0025] like Figure 1 As shown, this resource management method may include the following steps: Step 101: Analyze the resource content used for resource binding to obtain entity information.

[0026] The resource content can be used to indicate the core components and / or specific presentation form of the relevant resources. Resources can be, for example, books, commodities, energy, etc., and this disclosure does not limit them.

[0027] Optionally, resource content includes, but is not limited to: resource display images (such as advertising images, promotional posters, etc.), resource display videos (such as advertising videos, product demonstration videos, etc.), resource description text, resource audio (such as background music, voice narration, etc.), resource names, etc. Among them, resource display images and resource display videos can be used to display relevant resources, resource description text can be used to describe relevant resources, and resource audio can be used, for example, to enhance the persuasiveness of the relevant resources through voice narration or background music.

[0028] Entity information can be used to indicate a corresponding entity, such as a person, object, event, or location. As an example, entity information may include a brand name, resource names under that brand, etc. It should be noted that the above examples of entity information are merely illustrative; other types are possible in practical applications, and this disclosure does not impose any limitations on them.

[0029] In this embodiment of the disclosure, the resource content used for resource binding can be analyzed to obtain entity information. As an example, a deep learning model can be used to analyze the resource content used for resource binding, thereby extracting entity information from the resource content.

[0030] Step 102: Select multiple candidate resource nodes from multiple resource nodes in the resource tree based on the entity information.

[0031] The resource tree can include multiple levels of resource nodes. The levels can indicate logical ownership relationships, scope granularity relationships, and inheritance relationships of management permissions between resources. It should be noted that this disclosure does not limit the number of resource trees; that is, there can be one or multiple. For example, for products under the same brand, a corresponding resource tree can be established for that brand. For instance, for a specific car brand, the car brand itself can be the first-level resource node, and its sub-brands can be the second-level resource nodes. Under each sub-brand's resource node, resource nodes corresponding to various product categories are attached, such as nodes for "new energy vehicles," "gasoline vehicles," and "plug-in hybrids." Under the resource nodes corresponding to product categories, resource nodes corresponding to the car brand's vehicle series are attached, and so on.

[0032] It should be noted that the above example of a resource tree is merely exemplary, and other examples may be used in practical applications. This disclosure does not impose any restrictions on such examples.

[0033] Resource nodes can have corresponding resource attribute information and hierarchical information. Resource attribute information may include, for example, the resource node name, creation time, modification time, binding time, etc., and this disclosure does not impose any restrictions on this. Hierarchical information can be used to indicate the level to which the corresponding node belongs in the resource tree.

[0034] In this embodiment of the disclosure, multiple candidate resource nodes can be selected from multiple resource nodes in the resource tree based on entity information.

[0035] Step 103: Using the first LLM (Large Language Model), based on entity information, as well as the resource attribute information and hierarchical information of each candidate resource node, determine the resource node to be bound from multiple candidate resource nodes.

[0036] As an example, entity information, resource attribute information, and hierarchical information of each candidate resource node are input into the first LLM, and the resource node to be bound is obtained in response to the output of the first LLM.

[0037] Step 104: Establish the binding relationship between resource content and the resource node to be bound in order to manage the resource content.

[0038] The binding relationship can be further categorized into a corresponding relationship, a mapping relationship, and so on.

[0039] In this embodiment of the disclosure, a binding relationship can be established between resource content for resource binding and resource nodes to be bound, thereby realizing the binding of resource content to resource tree.

[0040] The resource management method of this disclosure analyzes the resource content used for resource binding to obtain entity information; selects multiple candidate resource nodes from multiple resource nodes in a resource tree based on the entity information; wherein the resource nodes have corresponding resource attribute information and hierarchical information; uses a first large language model LLM to determine the resource node to be bound from multiple candidate resource nodes based on the entity information and the resource attribute information and hierarchical information of each candidate resource node; and establishes a binding relationship between the resource content and the resource node to be bound in order to manage the resource content. Therefore, by extracting entity information from resource content, resource content can be associated and anchored with resource nodes, improving the adaptability and accuracy of resource binding. Simultaneously, by employing a pre-screening mechanism for candidate resource nodes, the range of target nodes can be quickly narrowed, effectively reducing the computational load during subsequent processing by the large language model, thereby improving the overall efficiency of resource binding. Furthermore, leveraging the powerful semantic understanding and computational capabilities of the large language model, combined with entity information, resource attribute information, and hierarchical information for comprehensive analysis and decision-making, the potential relationships between resource content and candidate resource nodes can be fully explored, improving the accuracy and reliability of selecting resource nodes to be bound. This further contributes to improving the accuracy and effectiveness of resource binding, enabling effective resource management, and facilitating subsequent resource retrieval, reuse, and maintenance.

[0041] like Figure 2 As shown, this disclosure also proposes a resource management method. Figure 2 As shown in the schematic diagram of the second embodiment of this disclosure, the resource management method may include the following steps: Step 201: Analyze the resource content used for resource binding to obtain entity information.

[0042] It should be noted that the execution process of step 201 can refer to the execution process of any embodiment of this disclosure, and will not be repeated here.

[0043] In any embodiment disclosed herein, target text information corresponding to resource content can be obtained, and a second LLM can be used to determine entity information based on the target text information.

[0044] To obtain the target text information corresponding to the resource content, at least one of the following methods can be used as a possible approach: When the resource content includes images, the text information obtained after image recognition of the images can be used as the target text information. When the resource content includes audio, the text information obtained after speech recognition of the audio can be used as the target text information. When the resource content includes raw text information, the raw text information can be used as the target text information.

[0045] In this embodiment of the disclosure, after obtaining text through speech recognition of audio, in one example, stop word removal processing can be performed on the text. In this embodiment of the disclosure, when the image originates from a video, in one example, an image recognition algorithm can be used to identify the subtitles in the image to obtain the text to be processed. For example, an Optical Character Recognition (OCR) algorithm can be used to identify the subtitles in the image to obtain the text to be processed. In another example, the text to be processed can be generated based on the video using a video understanding model. This video understanding model can be a large model with the ability to deeply understand and analyze video content. In yet another example, the text obtained from image subtitle recognition of the video can be combined with the text generated from the video using a video understanding model to generate the text to be processed. Thus, through multimodal information fusion, the information richness and accuracy of the text can be improved.

[0046] It should be noted that there are no restrictions on the method of obtaining the target text information here.

[0047] Therefore, it can support multimodal text information acquisition methods to meet the text conversion and extraction solutions in different scenarios, fully adapt to the resource management needs of different fields, and improve the user experience.

[0048] In one possible implementation of this disclosure, the target text information can be tagged based on its source type. This facilitates the rapid and accurate determination of the data source in subsequent data processing.

[0049] Therefore, in this embodiment of the disclosure, a second LLM can be used to determine entity information based on target text information.

[0050] As one possible implementation, the second prompt word can be determined based on the target text information; the second prompt word can be input into the second LLM, and entity information can be obtained from the model output.

[0051] To obtain the second prompt word, as an example, the target text information can be filled into the corresponding fill position in the second prompt template to obtain the second prompt word.

[0052] As another example, the professional knowledge constraint rules, sample examples, information extraction objects, and target text information can be filled into the corresponding fill positions in the second prompt template to obtain the second prompt word.

[0053] Among them, professional knowledge constraint rules can be used to indicate knowledge in specific fields such as industry, science, and technology. They can guide large models in defining the cognitive boundaries, terminology definitions, logical connections, and judgment criteria of related fields. For example, in the field of computer science, professional knowledge constraint rules can define "interface idempotency in API (Application Programming Interface) testing means that the result of multiple executions of the same request is consistent with the result of a single execution," and "RESTful API request methods include GET (query), POST (add), PUT (update), and DELETE (delete)." It should be noted that the above examples are merely illustrative; in practical applications, they can also be applied to other fields.

[0054] Among them, sample examples can be used to assist large models in analysis and judgment.

[0055] Among them, the information extraction object can be used to indicate the entity object to be extracted (such as a person, brand, etc.).

[0056] For example, the second prompt template includes slots for few samples, slots for professional knowledge constraint rules, slots for information extraction objects, and slots for text corresponding to the target text information. After filling the professional knowledge constraint rules, sample examples, information extraction objects, and target text information into the corresponding filling positions in the second prompt template, the resulting second prompt words are as follows:

[0057] It should be noted that the above description of each slot and the generated second prompt word is merely exemplary, and other options may be used in actual applications. This disclosure does not impose any restrictions on these options.

[0058] In summary, by leveraging the target text information corresponding to the resource content, an automated transformation link from resource content to entity information is established. Through deep semantic analysis of the target text information using the second LLM, the core entity information implicit in the resource text can be accurately extracted, which helps improve the data source quality for subsequent resource node matching. The second LLM has strong contextual understanding and generalization capabilities, and can adapt to different types and formats of text information, effectively solving the problems of incomplete and inaccurate entity recognition in unstructured text.

[0059] Step 202: Select multiple candidate resource nodes from multiple resource nodes in the resource tree based on the entity information.

[0060] It should be noted that the execution process of step 202 can refer to the execution process of any embodiment of this disclosure, and will not be described in detail here.

[0061] In this embodiment of the disclosure, the resource attribute information of a resource node may include the resource node name.

[0062] Step 203: Based on entity information, as well as resource attribute information and hierarchical information of each candidate resource node, determine the first prompt word.

[0063] As an example, entity information, resource attribute information of each candidate resource node, and hierarchical information can be filled into the corresponding fill positions in the first prompt template to obtain the first prompt word.

[0064] As another example, entity information, resource attribute information and hierarchical information of each candidate resource node, sample examples, and task guidance information can be filled into the corresponding fill positions in the first prompt template to obtain the first prompt word. For example, the first prompt template includes few-shot word slots, task guidance word slots, and hierarchical candidate noun slots. After filling the entity information, resource attribute information and hierarchical information of each candidate resource node, sample examples, and task guidance information into the corresponding fill positions in the first prompt template, the first prompt word is as follows:

[0065] Among them, sample examples can be used to assist large models in analysis and judgment.

[0066] It should be noted that the above-described slots and generated first prompt words are merely illustrative examples, and may be used in practice as well. This disclosure does not impose any restrictions on them.

[0067] Step 204: The first LLM is used to identify the resource node to be bound based on the first prompt word, and the identification result is obtained.

[0068] The identification results can include category labels, hierarchy labels, and name labels.

[0069] Among them, the classification label can be used to indicate whether entity information matches multiple candidate resource nodes. For example, it can be used to indicate whether the resource name in the entity information belongs to the resource node name of multiple candidate resource nodes.

[0070] The hierarchy label can be used to indicate the hierarchy to which the resource node to be bound belongs, and the name label can be used to indicate the name of the resource node to be bound.

[0071] As an example, the first prompt word can be input into the first LLM to identify the resource node to be bound, and the identification result can be obtained in response to the output of the first LLM.

[0072] Step 205: Establish the binding relationship between resource content and the resource node to be bound in order to manage the resource content.

[0073] It should be noted that the execution process of step 205 can refer to the execution process of any embodiment of this disclosure, and will not be described in detail here.

[0074] In any embodiment of this disclosure, when the classification label indicates that the entity information matches multiple candidate resource nodes, the resource node to be bound can be determined (or located) from the resource tree according to the hierarchical label and the name label, thereby establishing a binding relationship between the resource content and the resource node to be bound.

[0075] It is understandable that there may be situations where the classification label indicates that the entity information does not match multiple candidate resource nodes. Therefore, in one possible implementation of this disclosure, when the classification label indicates that the entity information does not match multiple candidate resource nodes, a prompt message can be generated and sent to indicate that the information is abnormal.

[0076] Therefore, differentiated processing strategies were designed for different matching results between entity information and candidate resource nodes. In scenarios where entity information matches multiple candidate resource nodes, the dual guidance of hierarchical and name tags allows for rapid location of the resource node to be bound from a complex resource tree, effectively resolving the filtering conflict problem of multiple candidate resource nodes and improving the accuracy of binding resource content to nodes. Conversely, in abnormal scenarios where entity information does not match multiple candidate resource nodes, the automatic generation and sending of prompts provides immediate feedback on resource matching anomalies, preventing invalid binding operations or erroneous associations and reducing the cost of subsequent anomaly troubleshooting. This not only improves the efficiency and accuracy of resource binding but also maintains the stability and reliability of the entire resource management process.

[0077] The resource management method of this disclosure determines a first prompt word based on entity information, resource attribute information, and hierarchical information of each candidate resource node. A first LLM (Local Language Model) is then used to identify the resource node to be bound based on the first prompt word, yielding an identification result. This identification result includes a category label, a hierarchical label, and a name label. The category label indicates whether the entity information matches multiple candidate resource nodes; the hierarchical label indicates the hierarchical level of the resource node to be bound; and the name label indicates the name of the resource node to be bound. Thus, by constructing a prompt word through the multi-dimensional fusion of entity information, resource attribute information, and hierarchical information, and combining the powerful semantic understanding and logical reasoning capabilities of a large language model, the identification and multi-label output of the resource node to be bound are achieved. On the one hand, the category label can quickly determine the matching relationship between entity information and candidate resource nodes, effectively avoiding the risk of misbinding unrelated resources and improving the accuracy and effectiveness of resource matching. On the other hand, the simultaneous output of the hierarchical label and name label can clearly define the hierarchical affiliation and identity of the resource node to be bound, helping to reduce the cost of screening and classification.

[0078] like Figure 3 As shown, this disclosure also proposes a resource management method. Figure 3 The resource management method, illustrated in the third embodiment of this disclosure, may include the following steps: Step 301: Analyze the resource content used for resource binding to obtain entity information.

[0079] It should be noted that the explanation of step 301 in any embodiment of this disclosure is also applicable to this embodiment, and will not be repeated here.

[0080] In this embodiment of the disclosure, entity information may include predicted resource objects. These predicted resource objects can be resource objects obtained by analyzing resource content used for resource binding. The resource object can be used to identify an entity of a type of data resource; for example, the resource object can be a brand, region, project, etc., and this disclosure does not impose any limitations on this.

[0081] Step 302: Based on the resource objects corresponding to multiple resource trees, determine the target resource tree that matches the predicted resource object from the multiple resource trees.

[0082] It should be noted that the explanation of step 302 in any embodiment of this disclosure is also applicable to this embodiment, and will not be repeated here.

[0083] In this embodiment of the disclosure, there can be multiple resource trees, and each resource tree has a corresponding resource object. For example, each resource tree has a corresponding brand, discipline, technical field, etc.

[0084] As an example, a resource tree whose corresponding resource object matches (or is the same as) the predicted resource object can be selected as the target resource tree. For instance, suppose there are three resource trees, and the resource objects corresponding to these three resource trees are subject a, subject b, and subject c, respectively. The predicted resource object is subject x. Then, by querying the resource objects of the above resource trees, the resource tree whose subject name is the same as the subject name of the predicted resource object is determined as the target resource tree.

[0085] Understandably, there may be situations where the predicted resource object differs from the resource objects corresponding to multiple resource trees. In such cases, as a possible implementation paradigm, the target resource tree matching the predicted resource object can be determined from multiple resource trees based on the vector similarity between the resource objects corresponding to each resource tree and the predicted resource object. For example, the resource tree with the highest vector similarity can be used as the target resource tree.

[0086] Step 303: Based on the matching degree information between entity information and resource nodes in the target resource tree, determine multiple candidate resource nodes.

[0087] The matching degree information can be used to indicate the degree of matching between entity information and resource nodes in the target resource tree, including but not limited to: matching degree value, matching degree level, etc.

[0088] One approach is to use a scoring system, or matching degree numerical value, to represent the matching degree between entity information and resource nodes in the target resource tree. In other words, the matching degree information can be quantified into a specific numerical value with a certain range of values, such as 0 to 100 or 0 to 10.

[0089] One approach is to use a rating system, or matching degree level, to represent the matching degree between entity information and resource nodes in the target resource tree. In other words, in the rating system, the matching degree information can be divided into several discrete levels, such as "high", "medium", "low", or numerical levels such as 1, 2, 3, etc., or letter levels such as "A+", "A", "A-", "B+", etc.

[0090] In order to obtain matching information, in one possible implementation of this disclosure, a vector generation model is used based on entity information and resource attribute information of each resource node to generate target vectors corresponding to entity information and node vectors corresponding to each resource node; for any resource node, the matching information between entity information and resource node is determined according to the similarity score between the target vector and the node vector corresponding to the resource node.

[0091] The vector generation model, such as an LLM, has the function of generating vectors. This disclosure does not restrict the vector generation model.

[0092] As an example, entity information can be input into a vector generation model to generate a target vector corresponding to the entity information, and the resource attribute information of any resource node can be input into a vector generation model to generate a node vector of the corresponding resource node.

[0093] As another example, when the entity information includes the predicted resource name, the predicted resource name in the entity information can be input into the vector generation model to generate the target vector corresponding to the entity information through the vector generation model. The resource node name in the resource attribute information of any resource node can be input into the vector generation model to generate the node vector of the corresponding resource node through the vector generation model.

[0094] Furthermore, in this disclosure, for any resource node, the similarity score between the target vector and the node vector corresponding to the resource node can be determined. For example, a similarity algorithm (such as cosine similarity algorithm, Euclidean distance algorithm, etc.) can be used to determine the similarity score between the target vector and the node vector corresponding to the resource node. Further, the matching degree information between entity information and resource node can be determined based on the similarity score between the target vector and the node vector corresponding to the resource node.

[0095] Therefore, by using a vector generation model, abstract entity information and resource attribute information of resource nodes are transformed into target vectors and node vectors in a high-dimensional space, achieving accurate transformation of unstructured information into structured vector data. Furthermore, by calculating vector similarity, the semantic association between entities and resource nodes is deeply mined, effectively avoiding matching deviations caused by keyword ambiguity and differences in information expression, and improving the accuracy and reliability of matching results.

[0096] In this embodiment of the disclosure, multiple candidate resource nodes can be determined based on the matching degree information between entity information and resource nodes in the target resource tree.

[0097] As one possible implementation, when the matching degree information includes matching degree values, multiple resource nodes can be sorted in descending order of matching degree values ​​to obtain a sorted sequence; based on the sorted sequence, multiple candidate resource nodes can be determined from the multiple resource nodes.

[0098] As another possible implementation, if the matching degree information includes the matching degree level, multiple resource nodes can be sorted in descending order of matching degree level to obtain a sorted sequence; based on the sorted sequence, multiple candidate resource nodes can be determined from the multiple resource nodes.

[0099] To determine multiple candidate resource nodes from multiple resource nodes based on a sorting sequence, as an example, a set number of resource nodes ranked first in the sorting sequence can be identified as candidate resource nodes; or, a set proportion of resource nodes ranked first in the sorting sequence can be identified as candidate resource nodes.

[0100] The quantity can be preset, such as 20, 30, etc., and can be set as needed. This disclosure does not impose any restrictions on this.

[0101] The set ratio can be preset, such as 10%, 20%, etc., and can be set as needed. This disclosure does not impose any restrictions on this.

[0102] It should be noted that the quantitative screening mechanism based on matching degree numerical ranking transforms the matching relationship between entities and resource nodes into a quantifiable priority ranking sequence. This enables priority screening of resource nodes with the strongest correlation to entity information, effectively eliminating irrelevant nodes with low matching degree, and improving the quality and relevance of the selected candidate resource nodes.

[0103] Step 304: Using the first major language model LLM, based on entity information, as well as the resource attribute information and hierarchical information of each candidate resource node, determine the resource node to be bound from multiple candidate resource nodes.

[0104] Step 305: Establish the binding relationship between resource content and the resource node to be bound in order to manage the resource content.

[0105] It should be noted that the execution process of steps 304 to 305 can refer to the execution process of any embodiment of this disclosure, and will not be repeated here.

[0106] The resource management method of this disclosure determines a target resource tree that matches a predicted resource object from multiple resource trees based on the resource objects corresponding to those resource trees. It then determines multiple candidate resource nodes based on the matching degree information between entity information and resource nodes in the target resource tree. Thus, by first relying on the matching logic between the predicted resource object and the corresponding resource object in the resource tree, the target resource tree is quickly located from multiple resource trees, narrowing the retrieval scope for subsequent node matching and effectively solving the problem of low efficiency in global traversal node search. Then, based on the matching degree information between entity information and nodes within the target resource tree, multiple candidate resource nodes that meet the conditions are filtered out, avoiding interference from irrelevant nodes and improving the targeting and effectiveness of node matching. This layered and progressive processing method reduces computational overhead and retrieval time in large-scale resource tree scenarios while maintaining the quality of candidate resource nodes, providing a high-quality data foundation for subsequent resource binding and further improving the efficiency and accuracy of resource management.

[0107] The above embodiments illustrate the method of resource binding. The following describes the method of retrieving resources in the resource tree, i.e., the resource retrieval method.

[0108] like Figure 4 As shown, this disclosure also proposes a resource recall method. Figure 4 As shown in the schematic diagram of the fourth embodiment of this disclosure, the resource recall method may include the following steps: Step 401: Analyze the resource recall task to obtain key resource information of the resources to be recalled.

[0109] The resource recall task can be used to recall relevant resources. It should be noted that the above explanation of resources also applies to this embodiment, and will not be repeated here.

[0110] Key resource information may include, but is not limited to, entity and time information.

[0111] In any embodiment of this disclosure, a resource retrieval task can be obtained by acquiring a task instruction for resource retrieval.

[0112] The task instruction can be a natural language expression representing the resource recall request. For example, the task instruction could be "Please help me find the price information for product xx".

[0113] In this embodiment of the disclosure, the resource recall task can be parsed to obtain key resource information of the resources to be recalled.

[0114] In any embodiment of this disclosure, LLM can be used to determine the key resource information of the resources to be recalled based on the resource recall task.

[0115] As an example, a third prompt word can be determined based on the resource retrieval task, the third prompt word can be input into the LLM, and the key resource information of the resource to be recalled can be obtained from the model output.

[0116] Step 402: Select multiple reference resource nodes from multiple resource nodes in the resource tree based on the key resource information.

[0117] The resource node can have corresponding resource attribute information, and the resource node has been pre-bound to the corresponding resource content; the resource attribute information may include, for example, the resource node name, and the binding relationship can be established using any possible implementation method of the above resource management methods.

[0118] It should be noted that the explanations of resource trees, resource nodes, resource attribute information, resource content, etc. in any of the above embodiments also apply to this embodiment, and will not be repeated here.

[0119] In any embodiment of this disclosure, when there are multiple resource trees, each resource tree has a corresponding resource object, and the key resource information includes the target resource object, a target resource tree matching the target resource object can be determined from the multiple resource trees based on the resource objects corresponding to the multiple resource trees; and multiple reference resource nodes can be determined based on the matching degree information between the key resource information and the resource nodes in the target resource tree.

[0120] It should be noted that the method for selecting reference resource nodes in this disclosure is similar to the method for selecting candidate resource nodes in the above-mentioned resource management methods, and will not be elaborated here.

[0121] Step 403: Using LLM, based on key resource information, as well as resource attribute information and hierarchical information of each reference resource node, determine the target resource node from multiple reference resource nodes to determine the resource to be recalled.

[0122] As an example, when the resource attribute information includes the resource node name, a fourth prompt word can be determined based on the key resource information, as well as the resource attribute information and hierarchical information of each reference resource node. The fourth prompt word is then input into the LLM, thereby responding to the model output and obtaining the recognition result. The recognition result can include a hierarchical label and a name label. The hierarchical label can be used to indicate the hierarchical level to which the target resource node belongs, and the name label can be used to indicate the resource node name of the target resource node.

[0123] Furthermore, based on hierarchical and name tags, the target resource node can be located in the resource tree, and the resource content that has established a binding relationship with the target resource node can be used as the resource to be recalled.

[0124] Step 404: Retrieve the resource content corresponding to the target resource node.

[0125] As an application scenario, after retrieving relevant resource content, the corresponding page filling mode can be determined according to the scenario type to which the resource retrieval task belongs; target information matching the resource retrieval task can be extracted from the retrieved resource content, and the target information can be filled into the corresponding filling position in the page filling mode to obtain push materials (such as advertising materials); then, the push materials can be pushed accordingly.

[0126] The scenario type, such as e-commerce scenario, science popularization scenario, promotion scenario, etc., can be set as needed, and this disclosure does not impose any restrictions on it.

[0127] To determine the scenario type to which a resource recall task belongs, one possible approach is to obtain scenario parameters when acquiring the task instruction for resource recall, and then determine the scenario type to which the resource recall task belongs based on the scenario parameters.

[0128] The scenario parameters may include, but are not limited to: industry attributes, task consultation channels, time information, user information, etc.

[0129] The page filling mode can include, but is not limited to: product manual mode, tweet mode, note mode, detail page mode, etc., and can be set as needed. This disclosure does not impose any restrictions on this.

[0130] In this embodiment of the disclosure, a correspondence between scene type and page filling mode can be established in advance and the correspondence can be saved. After determining the scene type to which the resource recall task belongs, the above correspondence can be queried to determine the corresponding page filling mode.

[0131] In this embodiment of the disclosure, target information matching the resource recall task can be extracted from the recalled resource content. For example, a large language model can be used to extract target information that matches the resource recall task and the fill type corresponding to the corresponding fill position in the page fill mode from the recalled resource content.

[0132] The fill type can include, but is not limited to, fill name type, fill image type, fill video type, fill description information type, etc., and can be set as needed. This disclosure does not impose any restrictions on this. Among them, the fill name type can be used to indicate that a name is filled in the corresponding fill position; the fill image type can be used to indicate that an image is filled in the corresponding fill position; the fill video type can be used to indicate that a video is filled in the corresponding fill position; and the fill description information type can be used to indicate that a description is filled in the corresponding fill position.

[0133] Furthermore, in this embodiment of the disclosure, the target information can be filled into the corresponding fill position in the page fill mode to obtain the push material. For example, the information in the target information that corresponds to the fill type can be filled into the corresponding fill position in the page fill mode to obtain the push material.

[0134] Furthermore, in this embodiment of the disclosure, the pushed materials can be pushed accordingly, such as to the terminal of the corresponding user.

[0135] Thus, by accurately recalling resources and precisely matching scenario types with page filling modes, the system efficiently transforms recalled resources into usable push materials. Customized page filling modes address the differentiated display needs of various scenarios, presenting target information in a way that better suits user habits, avoiding the clutter and lack of focus that often result from generic displays. Simultaneously, by selectively extracting task-matching target information from recalled resources, removing redundant content, and accurately filling it into the corresponding positions, the system maintains the information density and readability of push materials while reducing the information filtering cost for users. This push mechanism, combining scenario-based adaptation with precise filling, not only fully realizes the value of resource recall but also improves the user experience through "on-demand display and precise reach."

[0136] The resource retrieval method of this disclosure analyzes the resource retrieval task to obtain key resource information of the resource to be recalled. Based on the key resource information, multiple reference resource nodes are selected from multiple resource nodes in the resource tree. Each resource node has corresponding resource attribute information and hierarchical information, and is pre-bound to corresponding resource content using the resource management method described above. A Large Language Model (LLM) is used to determine the target resource node from the multiple reference resource nodes based on the key resource information and the resource attribute and hierarchical information of each reference resource node. Resource retrieval is then performed on the resource content corresponding to the target resource node. Thus, by analyzing the resource retrieval task to extract key resource information, clear guidance is provided for subsequent node selection. Furthermore, by selecting reference resource nodes from the resource tree based on this key information, the resource waste and inefficiency caused by full-domain scanning are avoided, and the retrieval scope is narrowed. Meanwhile, by leveraging LLM's deep semantic fusion and logical reasoning of key resource information and reference resource node resource attribute and hierarchical information, it can accurately capture the potential correlations and semantic fits between resource needs and node information. This effectively solves the recall bias problem caused by differences in information expression and single attribute dimensions, improving the accuracy and targeting of target resource node identification. Furthermore, since the binding relationship is established based on the aforementioned precise matching resource management method, the recall process can seamlessly adapt to multiple types of resource content. The integration of hierarchical information makes the recall results more consistent with the structural logic of the resource system, improving not only the efficiency and quality of resource recall but also the relevance, accuracy, and usability of recalled resources, thus enhancing the user experience.

[0137] To clearly illustrate the resource management and resource retrieval methods disclosed herein, the following explanation is provided with examples.

[0138] As an example, the resource management method of this disclosure is applied to the resource preprocessing process to establish a one-to-one mapping relationship (referred to as a binding relationship in this disclosure) between resource content and SPU nodes (referred to as resource nodes in this disclosure) in the SPU (Standard Product Unit) tree (referred to as the resource tree in this disclosure). The resource preprocessing process may include the following steps: Step 1: Preprocess the resource content and extract the entity information.

[0139] The entity information may include, for example, the name of the predicted resource and the object of the predicted resource.

[0140] It should be noted that before mounting resources for a brand's content, a preprocessing step is performed. Entity information is extracted from the large model for subsequent matching. Specifically, the core slots to be filled can include at least one of the following: a) Knowledge slots (referred to as professional knowledge constraint rule slots in this disclosure): specific knowledge content in the industry to which the current brand belongs; b) Few-shot slots: Some examples help large models make judgments; c) Brand name and resource name (referred to as information extraction object slots in this disclosure): the brand entity and corresponding resource name involved in each round of judgment.

[0141] This preprocessing step allows us to identify truly reasonable and easily identifiable entity information.

[0142] Step 2: Perform coarse screening of relevance using a vector model (referred to as the vector generation model in this disclosure).

[0143] At this point, to find the most appropriate SPU node in the SPU tree, it is necessary to determine the relevance of the entity information to each node in the SPU. Considering the timeliness requirements, a coarse-grained search is needed, which involves filtering out relatively irrelevant SPU nodes in the SPU tree.

[0144] When there are multiple SPU trees, a target SPU tree matching the predicted resource object can be determined from the multiple SPU trees based on the resource objects corresponding to the multiple SPU trees. A target vector of the predicted resource name in the entity information is obtained through vector modeling. Subsequently, a node vector corresponding to each SPU node in the target SPU tree is obtained through vector modeling. The correlation between the target vector and any node vector is calculated to determine the distance between the vectors (referred to as the similarity score in this disclosure). After calculation, the SPU nodes corresponding to the top-K most similar node vectors are taken as the coarse recruitment result.

[0145] Step 3: Input the coarse search results and entity information into the large model for precise matching.

[0146] The coarse recruitment results and entity information can be input into a large model, which will then determine the most matching SPU node (referred to as the resource node to be bound in this disclosure). Specifically, the core slots to be filled may include at least one of the following: a) few-shot slots: Some examples help large models make judgments. b) Brand name and resource name (referred to as task guidance information slots in this disclosure): the brand entity and corresponding resource name involved in each round of judgment.

[0147] c) Hierarchical candidate names (referred to as hierarchical candidate noun slots in this disclosure): Candidate SPU nodes (referred to as candidate resource nodes in this disclosure) in the coarse recruitment results are input hierarchically according to their level in the SPU tree.

[0148] Using hierarchical candidate SPU nodes as input to the large model can enhance its understanding and instruction-following capabilities, leading to instruction optimization. During relevance assessment, the position of a candidate SPU node in the SPU tree indicates its level; for example, a brand node corresponds to the first level, a category node to the third level, and so on. Therefore, when inputting the model, slots can be pre-defined as: first-level candidate names, second-level candidate names, third-level candidate names, and so on.

[0149] The model output may include at least the following: a) The is_in field: indicates whether the predicted resource name is contained in the candidate SPU node name; b) The level field: indicates the level corresponding to the most matching SPU node; c) The name field: indicates the name of the node that best matches the SPU node.

[0150] If the is_in field is 0, it means that the predicted resource name is not included in the candidate SPU node name. In this case, it means that the predicted resource name cannot correspond to the SPU node and needs to be manually checked and processed again. The level and name fields can be used to locate the best matching SPU node.

[0151] After resource preprocessing, the resource retrieval method of this disclosure can be implemented. The resource retrieval method may include: parsing a query command (referred to as a task command in this disclosure) to obtain key resource information of the resource to be retrieved; selecting multiple reference SPU nodes (referred to as reference resource nodes in this disclosure) from multiple SPU nodes in the SPU tree based on the key resource information; wherein, the SPU nodes have been pre-mapped with the corresponding resource content, and the mapping relationship is established using the method described above in the resource preprocessing process; using LLM, based on the key resource information and the node name and level of each reference SPU node, determining the target SPU node for determining the resource to be retrieved from multiple reference SPU nodes; and retrieving the resource content corresponding to the target SPU node.

[0152] In one possible implementation of this disclosure, the corresponding page filling mode is determined according to the scenario type corresponding to the query command; target information matching the query command is extracted from the recalled resource content; the target information is filled into the corresponding filling position in the page filling mode to obtain the push material; and the push material is pushed.

[0153] In summary, the method disclosed herein has at least the following advantages: 1. High Accuracy: A rigorous three-step processing mechanism is employed in the resource matching process. First, a large-scale model is used for preprocessing to initially screen and structure massive amounts of data. Next, a vector model is used for coarse screening to quickly locate potential matching ranges. Finally, the large-scale model is used for precise matching to ensure that each resource can accurately establish a correspondence with an SPU node. This approach allows for efficient retrieval of all resources associated with a given SPU node when pushing materials to relevant users, simply by determining the SPU node corresponding to the user's query. This improves the accuracy of resource retrieval and effectively enhances the push performance.

[0154] 2. Strong Explainability: Every step from user needs to resource matching is traceable, resulting in extremely strong overall explainability. This architecture not only facilitates understanding the logical process of resource matching but also provides data support for subsequent fulfillment of various user resource needs (such as personalized push notifications and resource analysis), enabling continuous expansion and optimization of system functions.

[0155] To implement the above embodiments, this disclosure also provides a resource management device. For example... Figure 5 As shown, Figure 5 This is a schematic diagram according to the fifth embodiment of the present disclosure. The resource management device 500 may include: an analysis module 501, a selection module 502, a first determination module 503, and an establishment module 504.

[0156] The analysis module 501 is used to analyze the resource content used for resource binding to obtain entity information.

[0157] Selection module 502 is used to select multiple candidate resource nodes from multiple resource nodes in the resource tree based on entity information; wherein, the resource nodes have corresponding resource attribute information and hierarchical information.

[0158] The first determination module 503 is used to determine the resource node to be bound from multiple candidate resource nodes by using the first large language model LLM, based on entity information, as well as the resource attribute information and hierarchical information of each candidate resource node.

[0159] Module 504 is used to establish a binding relationship between resource content and resource nodes to be bound, in order to manage the resource content.

[0160] In one possible implementation of this disclosure, the first determining module 503 is configured to: determine a first prompt word based on entity information, resource attribute information and hierarchical information of each candidate resource node; and use a first LLM to identify the resource node to be bound based on the first prompt word to obtain an identification result; wherein the identification result includes a category label, a hierarchical label and a name label, the category label is used to indicate whether the entity information matches multiple candidate resource nodes, the hierarchical label is used to indicate the hierarchical level to which the resource node to be bound belongs, and the name label is used to indicate the resource node name of the resource node to be bound.

[0161] In one possible implementation of this disclosure, the establishment module 504 is configured to: in response to the classification label indicating that entity information matches with multiple candidate resource nodes, determine the resource node to be bound from the resource tree according to the hierarchical label and the name label; and establish a binding relationship between the resource content and the resource node to be bound.

[0162] In one possible implementation of this disclosure, the resource management device 500 may further include: The processing module is used to generate and send a prompt message in response to the classification label indicating that the entity information does not match multiple candidate resource nodes, so as to indicate that the information is abnormal.

[0163] In one possible implementation of this disclosure, there are multiple resource trees, each resource tree has a corresponding resource object, and the entity information includes the predicted resource object; the selection module 502 is used to: determine a target resource tree that matches the predicted resource object from the multiple resource trees according to the resource objects corresponding to the multiple resource trees; and determine multiple candidate resource nodes based on the matching degree information between the entity information and the resource nodes in the target resource tree.

[0164] In one possible implementation of this disclosure, the resource management device 500 may further include: The generation module is used to generate target vectors corresponding to entity information and node vectors corresponding to each resource node based on entity information and resource attribute information of each resource node using a vector generation model.

[0165] The second determining module is used to determine the matching degree information between entity information and resource node based on the similarity score between the target vector and the node vector corresponding to the resource node for any resource node.

[0166] In one possible implementation of this disclosure, the matching degree information includes a matching degree value; the selection module 502 is used to: sort multiple resource nodes in descending order of matching degree values ​​to obtain a sorting sequence; and determine multiple candidate resource nodes from the multiple resource nodes based on the sorting sequence.

[0167] In one possible implementation of this disclosure, the selection module 502 is used to: determine a set number of resource nodes that are ranked first in the sorting sequence as candidate resource nodes; or, determine a set proportion of resource nodes that are ranked first in the sorting sequence as candidate resource nodes.

[0168] In one possible implementation of this disclosure, the analysis module 501 is used to: obtain target text information corresponding to the resource content; and use a second LLM to determine entity information based on the target text information.

[0169] In one possible implementation of this disclosure, the analysis module 501 is configured to perform at least one of the following: in response to the resource content including an image, using text information obtained after image recognition of the image as target text information; in response to the resource content including audio, using text information obtained after speech recognition of the audio as target text information; in response to the resource content including original text information, using original text information as target text information.

[0170] It should be noted that the resource management device provided in this embodiment can implement all the method steps implemented in any of the above resource management method embodiments and can achieve the same technical effect. Therefore, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail here.

[0171] The resource management device of this disclosure analyzes the resource content used for resource binding to obtain entity information; selects multiple candidate resource nodes from multiple resource nodes in the resource tree based on the entity information; wherein the resource nodes have corresponding resource attribute information and hierarchical information; adopts the first large language model LLM, and determines the resource node to be bound from multiple candidate resource nodes based on the entity information, as well as the resource attribute information and hierarchical information of each candidate resource node; establishes a binding relationship between the resource content and the resource node to be bound in order to manage the resource content. Therefore, by extracting entity information from resource content, resource content can be associated and anchored with resource nodes, improving the adaptability and accuracy of resource binding. Simultaneously, by employing a pre-screening mechanism for candidate resource nodes, the range of target nodes can be quickly narrowed, effectively reducing the computational load during subsequent processing by the large language model, thereby improving the overall efficiency of resource binding. Furthermore, leveraging the powerful semantic understanding and computational capabilities of the large language model, combined with entity information, resource attribute information, and hierarchical information for comprehensive analysis and decision-making, the potential relationships between resource content and candidate resource nodes can be fully explored, improving the accuracy and reliability of selecting resource nodes to be bound. This further contributes to improving the accuracy and effectiveness of resource binding, enabling effective resource management, and facilitating subsequent resource retrieval, reuse, and maintenance.

[0172] To implement the above embodiments, this disclosure also provides a resource recall device. For example... Figure 6 As shown, Figure 6 This is a schematic diagram according to the sixth embodiment of the present disclosure. The resource recall device 600 may include: a parsing module 601, a selection module 602, a first determination module 603, and a recall module 604.

[0173] The parsing module 601 is used to parse the resource recall task to obtain the key resource information of the resource to be recalled.

[0174] Selection module 602 is used to select multiple reference resource nodes from multiple resource nodes in the resource tree based on key resource information; wherein, the resource nodes have corresponding resource attribute information and hierarchical information, and are pre-bound to the corresponding resource content, and the binding relationship is established by any possible implementation method of the resource management method disclosed herein.

[0175] The first determination module 603 is used to determine the target resource node as the resource to be recalled from multiple reference resource nodes by using a large language model (LLM) based on key resource information, as well as resource attribute information and hierarchical information of each reference resource node.

[0176] The recall module 604 is used to recall the resource content corresponding to the target resource node.

[0177] In one possible implementation of this disclosure, the resource recall device 600 may further include: The second determining module is used to determine the corresponding page filling mode based on the scenario type to which the resource recall task belongs.

[0178] The extraction module is used to extract target information that matches the resource recall task from the recalled resource content.

[0179] The fill module is used to fill the target information into the corresponding fill position in the page fill mode to obtain the pushed material.

[0180] The push module is used to push materials.

[0181] The resource retrieval device of this embodiment parses the resource retrieval task to obtain key resource information of the resource to be recalled; based on the key resource information, it selects multiple reference resource nodes from multiple resource nodes in the resource tree; wherein, each resource node has corresponding resource attribute information and hierarchical information, and is pre-bound to the corresponding resource content, the binding relationship being established using any possible implementation of the resource management method described above; using a Large Language Model (LLM), based on the key resource information and the resource attribute information and hierarchical information of each reference resource node, it determines the target resource node as the resource to be recalled from the multiple reference resource nodes; and performs resource retrieval on the resource content corresponding to the target resource node. Thus, by parsing the resource retrieval task to extract key resource information, it provides clear guidance for subsequent node selection; and by selecting reference resource nodes from the resource tree based on this key information, it avoids the resource waste and inefficiency problems caused by full-domain scanning, and narrows the retrieval scope. Meanwhile, by leveraging LLM's deep semantic fusion and logical reasoning of key resource information and reference resource node resource attribute and hierarchical information, it can accurately capture the potential correlations and semantic fits between resource needs and node information. This effectively solves the recall bias problem caused by differences in information expression and single attribute dimensions, improving the accuracy and targeting of target resource node identification. Furthermore, since the binding relationship is established based on the aforementioned precise matching resource management method, the recall process can seamlessly adapt to multiple types of resource content. The integration of hierarchical information makes the recall results more consistent with the structural logic of the resource system, improving not only the efficiency and quality of resource recall but also the relevance, accuracy, and usability of recalled resources, thus enhancing the user experience.

[0182] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision, and disclosure of any type of information, such as user personal information, are all carried out with the user's consent and comply with relevant laws and regulations, and do not violate public order and good morals.

[0183] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0184] Figure 7 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0185] like Figure 7 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0186] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0187] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as resource management methods or resource retrieval methods. For example, in some embodiments, the resource management method or resource retrieval method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the resource management method or resource retrieval method described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to perform a resource management method or a resource recall method by any other suitable means (e.g., by means of firmware).

[0188] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0189] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0190] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0191] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0192] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0193] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0194] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0195] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A resource management method, wherein, The method includes: The resource content used for resource binding is analyzed to obtain entity information; Based on the entity information, multiple candidate resource nodes are selected from multiple resource nodes in the resource tree; wherein, the resource nodes have corresponding resource attribute information and hierarchical information; Using the first major language model, LLM, based on the entity information, as well as the resource attribute information and hierarchical information of each candidate resource node, the resource node to be bound is determined from the multiple candidate resource nodes; Establish a binding relationship between the resource content and the resource node to be bound in order to manage the resource content.

2. The method according to claim 1, wherein, The resource attribute information includes the resource node name; the method of using the first large language model LLM, based on the entity information, and the resource attribute information and hierarchical information of each candidate resource node, to determine the resource node to be bound from the plurality of candidate resource nodes includes: Based on the entity information, as well as the resource attribute information and hierarchical information of each candidate resource node, the first prompt word is determined; The first LLM is used to identify the resource node to be bound based on the first prompt word, and the identification result is obtained; The identification result includes a category label, a hierarchy label, and a name label. The category label indicates whether the entity information matches the multiple candidate resource nodes. The hierarchy label indicates the hierarchy to which the resource node to be bound belongs. The name label indicates the name of the resource node to be bound.

3. The method according to claim 2, wherein, Establishing the binding relationship between the resource content and the resource node to be bound includes: In response to the classification label indicating that the entity information matches the plurality of candidate resource nodes, the resource node to be bound is determined from the resource tree based on the hierarchical label and the name label; Establish a binding relationship between the resource content and the resource node to be bound.

4. The method according to claim 3, wherein, The method further includes: In response to the classification label indicating that the entity information does not match the multiple candidate resource nodes, a prompt message is generated and sent to indicate an anomaly.

5. The method according to claim 1, wherein, There are multiple resource trees, each of which has a corresponding resource object. The entity information includes the predicted resource object. The step of selecting multiple candidate resource nodes from the multiple resource nodes of the resource trees based on the entity information includes: Based on the resource objects corresponding to the multiple resource trees, determine the target resource tree that matches the predicted resource object from the multiple resource trees; Based on the matching degree information between the entity information and the resource nodes in the target resource tree, the plurality of candidate resource nodes are determined.

6. The method according to claim 5, wherein, The process of obtaining the matching degree information includes the following steps: Based on the entity information and the resource attribute information of each resource node, a vector generation model is used to generate the target vector corresponding to the entity information and the node vector corresponding to each resource node. For any of the resource nodes, the matching degree information between the entity information and the resource node is determined based on the similarity score between the target vector and the node vector corresponding to the resource node.

7. The method according to claim 5, wherein, The matching degree information includes a matching degree value; determining the plurality of candidate resource nodes based on the matching degree information between the entity information and the resource nodes in the target resource tree includes: The multiple resource nodes are sorted in descending order of their matching degree values ​​to obtain a sorted sequence; Based on the sorting sequence, the plurality of candidate resource nodes are determined from the plurality of resource nodes.

8. The method according to claim 7, wherein, The step of determining the plurality of candidate resource nodes from the plurality of resource nodes based on the sorting sequence includes: The resource nodes that rank first in the sorted sequence are determined as the candidate resource nodes; or, A predetermined proportion of the resource nodes that appear first in the sorting sequence are identified as candidate resource nodes.

9. The method according to claim 1, wherein, The analysis of the resource content used for resource binding to obtain entity information includes: Obtain the target text information corresponding to the resource content; The second LLM is used to determine the entity information based on the target text information.

10. The method according to claim 9, wherein, The acquisition of the target text information corresponding to the resource content includes at least one of the following: In response to the fact that the resource content includes an image, the text information obtained after image recognition of the image is used as the target text information; In response to the fact that the resource content includes audio, the text information obtained after performing speech recognition on the audio is used as the target text information; In response to the resource content including original text information, the original text information is used as the target text information.

11. A resource retrieval method, wherein, The method includes: The resource recall task is analyzed to obtain key resource information of the resources to be recalled; Based on the key resource information, multiple reference resource nodes are selected from multiple resource nodes in the resource tree; wherein, the resource node has corresponding resource attribute information and hierarchical information, and is pre-bound to the corresponding resource content, and the binding relationship is established by the method described in any one of claims 1-10; Using a large language model (LLM), based on the key resource information, as well as the resource attribute information and hierarchical information of each of the reference resource nodes, a target resource node for determining the resource to be recalled is determined from the plurality of reference resource nodes; Resource retrieval is performed on the resource content corresponding to the target resource node.

12. The method according to claim 11, wherein, The method further includes: Determine the corresponding page filling mode based on the scenario type to which the resource recall task belongs; Extract target information that matches the resource recall task from the recalled resource content; The target information is filled into the corresponding fill position in the page fill mode to obtain the push material; The push material is pushed.

13. A resource management device, wherein, The device includes: The analysis module is used to analyze the resource content used for resource binding to obtain entity information; The selection module is used to select multiple candidate resource nodes from multiple resource nodes in the resource tree based on the entity information; wherein the resource nodes have corresponding resource attribute information and hierarchical information; The first determining module is used to determine the resource node to be bound from the plurality of candidate resource nodes by using the first large language model LLM, based on the entity information, as well as the resource attribute information and hierarchical information of each candidate resource node; A module is established to create a binding relationship between the resource content and the resource node to be bound, so as to manage the resource content.

14. A resource recall device, wherein, The device includes: The parsing module is used to parse the resource recall task to obtain the key resource information of the resource to be recalled; The selection module is used to select multiple reference resource nodes from multiple resource nodes in the resource tree based on the key resource information; wherein the resource node has corresponding resource attribute information and hierarchical information, and is pre-bound to the corresponding resource content, and the binding relationship is established by the method described in any one of claims 1-10; The first determining module is used to use a large language model (LLM) to determine the target resource node as the resource to be recalled from the plurality of reference resource nodes based on the key resource information, as well as the resource attribute information and hierarchical information of each of the reference resource nodes. The recall module is used to recall the resource content corresponding to the target resource node.

15. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 10, or the method of any one of claims 11-12.

16. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 10, or the method according to any one of claims 11-12.

17. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 10, or the method according to any one of claims 11-12.

Images