Construction method and system of digital content identification system

Abstract

The invention provides a method for constructing a digital content identification system, which comprises the following steps of: generating training data required by model training by utilizing a pre-trained large language model; constructing an initial tree structure from different dimensions according to the training data to form an initial identification system; extracting identification data from the initial identification system, constructing a positive and negative sample pair, and finely adjusting a semantic embedding model by using the positive and negative sample pair; and performing semantic comparison on a newly appearing digital content identifier and an identifier in an identifier system by using the fine-tuned semantic embedding model, determining a peer node associated with the content of the new digital content identifier in the tree structure, and positioning the new digital content identifier to the peer node. The invention further provides a construction system of the digital content identification system, a storage medium and electronic equipment. Therefore, under the condition that a large amount of basic data is not needed, an identification system capable of accurately capturing the deep semantic relation of the words can be constructed, so that the identification set is structured and correlated, and analysis, display and continuous updating are facilitated.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for constructing a digital content identification system, a storage medium and an electronic device. Background Art

[0002] With the vigorous development of blockchain technology, the application of digital content on blockchain is increasing. However, due to the virtual, abstract and dynamic evolution of digital content, effective identification and management has become a major challenge. The identification system refers to an overall structure of the content identification set. The existing identification system is difficult to adapt to the special needs of blockchain digital content, resulting in difficulties in the organization, retrieval and supervision of these digital contents. Therefore, building a targeted digital content identification system has become a problem to be solved.

[0003] At present, the following technical methods are mainly used for structuring and relationship construction of word sets:

[0004] (1) Vector representation combined with clustering and classification algorithms:

[0005] Principle: Use word vector models to map words to high-dimensional space to capture the similarities between words. The word vector models include but are not limited to Word2Vec, GloVe (Global Vectors for Word Representation), BGE (BAAI General Embedding, vector model of Beijing Institute of Artificial Intelligence), etc. Then, clustering and classification algorithms are used to group and classify words.

[0006] Implementation: Without special fine-tuning, the word vector model mainly generates vector representations of words based on contextual co-occurrence relationships. Clustering algorithms such as K-means or hierarchical clustering divide words into different clusters based on these vector representations.

[0007] Analysis: This method can find the commonality of words to a certain extent, but it fails to accurately capture the deep semantics and nuances of words, resulting in inaccurate classification results. At the same time, clustering algorithms can only identify the similarity of words and cannot clearly describe the specific relationship between words.

[0008] (2) Topic Model:

[0009] Principle: Through statistical methods, potential topic distribution is extracted from a large amount of text data to represent the relationship between words and topics.

[0010] Implementation: Use topic models such as LDA (Latent Dirichlet Allocation) to model the probability distribution of words and divide the words into different topic categories.

[0011] Analysis: Topic models can capture the topic structure of text at a macro level, but the generated topics are often too abstract to correspond to actual concepts or categories. The result is only a classification of words, lacking a description of the specific relationship between words.

[0012] (3) Knowledge Graph:

[0013] Principle: By constructing a network of entities (words) and relationships, various relationships between words can be accurately described.

[0014] Implementation: Manually or automatically build a graph structure containing entities and relationships, and use standards such as RDF (Resource Description Framework) and OWL (Web Ontology Language) to describe terms and their relationships.

[0015] Analysis: Knowledge graphs can theoretically fully represent the relationships between words, but manual construction requires a lot of manpower and expertise, while automated construction requires complex algorithms and high-quality data support. Due to the lack of sufficient basic data, building high-quality knowledge graphs is complex and costly.

[0016] The above defects in the construction process of the existing blockchain digital content identification system are mainly due to the following reasons:

[0017] 1. Lack of accurate capture of the deep semantics of new words: The existing word vector model does not consider the semantic features specific to blockchain digital content, resulting in the inability to accurately map newly emerging words to the existing system.

[0018] 2. Unable to reflect the hierarchy and relationship between words: Clustering and topic models can only provide category division, lack of description of the specific relationship and hierarchy between words, and cannot form an organized system.

[0019] 3. High construction complexity and large data requirements: The construction of knowledge graphs requires complex algorithms and a large amount of high-quality data. The rapid development of blockchain digital content makes data updates frequent, making it difficult to maintain high-quality knowledge graphs.

[0020] In summary, the prior art obviously has inconveniences and defects in practical use, so it is necessary to improve it. Summary of the invention

[0021] In view of the above-mentioned defects, the purpose of the present invention is to provide a method and system for constructing a digital content identification system, a storage medium and an electronic device, which can construct an identification system that accurately captures the deep semantic relationship of words without the need for a large amount of basic data, so that the identification set is structured and relational, and is easy to analyze, display and continuously update.

[0022] In order to solve the above-mentioned technical problems, the present invention is achieved as follows:

[0023] The present invention provides a method for constructing a digital content identification system, comprising:

[0024] Use the pre-trained large language model to generate the training data required for model training;

[0025] Constructing an initial tree structure from different dimensions according to the training data to form an initial identification system;

[0026] Extracting identification data from the initial identification system, constructing positive and negative sample pairs, and fine-tuning the semantic embedding model using the positive and negative sample pairs;

[0027] By using the fine-tuned semantic embedding model, a semantic comparison is performed between the newly appeared digital content identifier and the identifier in the identifier system, the content-associated sibling nodes of the new digital content identifier in the tree structure are determined, and the new digital content identifier is positioned at the sibling nodes.

[0028] According to the method for constructing a digital content identification system of the present invention, the step of using a pre-trained large language model to generate training data required for model training includes:

[0029] Collect existing digital content identifiers to form an initial identifier set;

[0030] Generate semantically relevant words and semantically irrelevant words of the digital content identifier using the large language model;

[0031] Combine manual review and model evaluation to eliminate worthless training data and retain valuable training data.

[0032] According to the method for constructing a digital content identification system of the present invention, after the step of using a pre-trained large language model to generate training data required for model training, the method further includes:

[0033] The training data is optimized using a data iteration optimization algorithm.

[0034] According to the method for constructing a digital content identification system of the present invention, the tree structure is used as an initial identification system to represent the hierarchy and relationship between identifications; and / or

[0035] The dimensions include content, subject and / or status.

[0036] According to the method for constructing a digital content identification system of the present invention, the step of fine-tuning the semantic embedding model using positive and negative sample pairs comprises:

[0037] The step of bringing semantically similar words closer and farther apart brings the vectors of semantically similar words closer and the vectors of semantically irrelevant words farther apart, preliminarily adjusting the semantic distance between words;

[0038] The aggregation and separation steps of nodes at the same level bring the word vectors of nodes at the same level closer for the first time, pull the samples with farther distances farther away, and strengthen the aggregation of words at the same level;

[0039] The optimization step of the hierarchical structure brings the word vectors of nodes at the same level closer for the second time, and pulls the child node vectors of the sibling nodes of their parent nodes farther apart, clarifying the relationship between the hierarchical structure and words.

[0040] According to the method for constructing a digital content identification system of the present invention, the step of bringing semantically similar words closer and further apart comprises:

[0041] Select words with similar semantics as the first positive sample pair;

[0042] Select semantically irrelevant words as the first negative sample pair;

[0043] By adjusting the model parameters, the vector distance of the first positive sample pair is reduced, and the vector distance of the first negative sample pair is increased;

[0044] The steps of aggregating and separating the peer nodes include:

[0045] Select the node words of the same level as the second positive sample pair;

[0046] Select the remote node words as the second negative sample pair;

[0047] The word vectors of nodes at the same level are brought closer for the first time, while the vector distances of word pairs that are farther apart are pulled farther apart;

[0048] The optimization steps of the hierarchical structure include:

[0049] Select the node words of the same level as the third positive sample pair;

[0050] Select nearby node words as the third negative sample pair;

[0051] The word vector distance of the nodes at the same level is shortened for the second time, and the word vector distance of the child node of the sibling node of its parent node is increased.

[0052] According to the method for constructing a digital content identification system of the present invention, the steps of using the fine-tuned semantic embedding model to semantically compare the newly appeared new digital content identification with the identification in the identification system, determining the content-related peer nodes of the new digital content identification in the tree structure, and locating the new digital content identification at the peer nodes include:

[0053] Using the fine-tuned semantic embedding model, obtaining a vector representation of the newly appeared new digital content identifier;

[0054] Calculate the similarity between the vector of the new digital content identifier and the vector of the original word in the identifier system;

[0055] Determining, based on the result of the similarity calculation, a word node associated with the content of the new digital content identifier;

[0056] The new digital content identifier is inserted into the corresponding node of the tree structure as a node of the same level as the word node associated with the content.

[0057] In a second aspect, an embodiment of the present invention provides a system for constructing a digital content identification system based on any of the above methods, including:

[0058] The data generation module is used to generate the training data required for model training using the pre-trained large language model;

[0059] A system construction module, used to construct an initial tree structure from different dimensions according to the training data to form an initial identification system;

[0060] A system fine-tuning module, used to extract identification data from the initial identification system, construct positive and negative sample pairs, and use the positive and negative sample pairs to fine-tune the semantic embedding model;

[0061] The system update module is used to use the fine-tuned semantic embedding model to semantically compare the newly appeared new digital content identifier with the identifier in the identifier system, determine the content-related peer node of the new digital content identifier in the tree structure, and locate the new digital content identifier at the peer node.

[0062] In a third aspect, an embodiment of the present invention provides a storage medium for storing a computer program for executing any one of the methods for constructing a digital content identification system.

[0063] In a fourth aspect, an embodiment of the present invention provides a computer device, comprising a storage medium, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements any one of the methods for constructing a digital content identification system when executing the computer program.

[0064] The construction technology of the digital content identification system of the present invention includes: (1) data generation: using a pre-trained large language model to generate training data required for model training; and using a data iteration optimization algorithm to gradually improve data quality, that is, without a large amount of basic data support, rich training samples can be obtained. (2) Constructing a multi-dimensional tree structure: constructing an initial tree structure from different dimensions according to the training data to form an initial identification system. (3) Fine-tuning the semantic embedding model: using positive and negative sample pairs, fine-tuning the semantic embedding model, preferably polling and fine-tuning the semantic embedding model in three stages to accurately capture the deep semantic relationship of words. (4) Dynamic expansion of the system architecture: semantically comparing the newly emerging new digital content identification with the logo of the logo system, determining the content-related nodes of the new digital content identification in the tree structure, and locating the new digital content identification to the nodes of the same level, automatically expanding the tree structure, and keeping the system updated and perfect. The present invention is particularly suitable for digital content identification on the blockchain. The present invention is aimed at the characteristics of digital content on the blockchain that are virtual, abstract, and dynamically evolving. The existing technology cannot effectively construct a structured identification system containing specific relationships between words. The present invention can construct an identification system that can accurately capture the deep semantic relationship of words without requiring a large amount of basic data, making the identification set structured and relational, and convenient for analysis, display and continuous updating. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a flowchart of a method for constructing a digital content identification system provided in the first embodiment of the present invention;

[0066] Figure 2 It is a flowchart of a method for constructing a digital content identification system provided in Embodiment 2 of the present invention;

[0067] Figure 3 is a schematic diagram of the integration process of the new identifier provided in the third embodiment of the present invention;

[0068] Figure 4 It is a structural diagram of a system for constructing a digital content identification system provided in the first embodiment of the present invention;

[0069] Figure 5 It is a structural diagram of a system for constructing a digital content identification system provided in Embodiment 2 of the present invention;

[0070] Figure 6It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0072] It should be noted that references to "one embodiment", "embodiment", "example embodiment", etc. in this specification refer to the embodiment described, which may include specific features, structures or characteristics, but not every embodiment must include these specific features, structures or characteristics. In addition, such expressions do not refer to the same embodiment. Furthermore, when describing specific features, structures or characteristics in conjunction with an embodiment, whether or not there is an explicit description, it has been shown that it is within the knowledge of those skilled in the art to combine such features, structures or characteristics into other embodiments.

[0073] In addition, certain words are used in the specification and subsequent claims to refer to specific components or parts. Those with ordinary knowledge in the relevant field should understand that manufacturers can use different nouns or terms to refer to the same component or part. This specification and subsequent claims do not use differences in names as a way to distinguish components or parts, but use differences in the functions of components or parts as the criteria for distinction. "Including" and "including" mentioned throughout the specification and subsequent claims are open-ended terms and should be interpreted as "including but not limited to". In addition, the word "connected" here includes any direct and indirect electrical connection means. Indirect electrical connection means include connection through other devices.

[0074] The following describes in detail the method for constructing a digital content identification system provided by the embodiment of the present invention through specific embodiments and application scenarios in conjunction with the accompanying drawings.

[0075] Figure 1 : is a flow chart of a method for constructing a digital content identification system provided in Embodiment 1 of the present invention. The present invention aims to transform an unstructured identification set into a structured system containing specific relationships between words. The method at least comprises the following steps:

[0076] Step S101, training data generation: using a pre-trained large language model to generate training data required for model training.

[0077] Preferably, a data iteration optimization algorithm is used to gradually improve the data quality. The large language models include but are not limited to GPT (Generative Pre-trained Transformer) and Gemini (Google's artificial intelligence model). The data iteration optimization of the present invention includes: 1. The steps of bringing semantically similar words closer and farther apart; 2. The steps of aggregating and separating the nodes at the same level; 3. The step of optimizing the hierarchical structure. That is, iterative training from simple to difficult.

[0078] Step S102, constructing a multi-dimensional tree structure: constructing an initial tree structure from different dimensions according to the training data to form an initial identification system.

[0079] From different perspectives, tree structures are constructed as the initial identification system to represent the hierarchy and relationship between identifications. The advantages of building a tree structure include: (1) simple structure and data-friendly; (1) clear structure and easy management.

[0080] Step S103, fine-tuning the semantic embedding model: extracting identification data from the initial identification system, constructing positive and negative sample pairs, and using the positive and negative sample pairs to fine-tune the semantic embedding model.

[0081] Preferably, the pre-trained semantic embedding model is fine-tuned in three stages using positive and negative sample pairs, so that it can accurately capture the deep semantic relationship of words.

[0082] Step S104, integration of new identifiers and system expansion: using the fine-tuned semantic embedding model, semantically compare the newly appeared new digital content identifier with the identifier (i.e., word) of the identification system, determine the content-related nodes of the new digital content identifier in the tree structure, and locate the new digital content identifier at the node of the same level.

[0083] For newly emerging blockchain digital content identifiers, a fine-tuned semantic embedding model is used to perform semantic comparisons between the new identifiers and the words in the identification system to determine their position in the tree structure and achieve dynamic expansion of the system.

[0084] The method for constructing a digital content identification system provided by the embodiment of the present invention is based on the characteristics of virtual, abstract, and dynamically evolving digital content on the blockchain. The existing technology cannot effectively construct a structured identification system that includes specific relationships between words. The present invention can construct an identification system that can accurately capture the deep semantic relationship between words without requiring a large amount of basic data, making the identification set structured and relational, and convenient for analysis, display, and continuous updating.

[0085] Figure 2: is a flow chart of a method for constructing a digital content identification system provided in Embodiment 2 of the present invention, wherein the method comprises the following steps:

[0086] Step S201, identification data collection: collecting existing digital content identifications to form an initial identification set.

[0087] Step S202, identification data generation: using a large language model to generate semantically relevant words and semantically irrelevant words for digital content identification.

[0088] Step S203, identification data optimization: combining manual review and model evaluation to eliminate worthless training data and retain valuable training data.

[0089] The steps S201 to S203 are steps for generating training data, that is, using a pre-trained large language model to generate the training data required for model training, without requiring a large amount of basic data.

[0090] Preferably, after step S203, the following steps are further included:

[0091] The data iteration optimization algorithm is used to optimize the training data.

[0092] Step S204, constructing a multi-dimensional tree structure: constructing an initial tree structure from different dimensions according to the training data to form an initial identification system.

[0093] Preferably, the tree structure is used as an initial identification system to represent the hierarchy and relationship between identifications. And / or

[0094] The dimensions include content, subject and / or status, etc.

[0095] Step S205, fine-tuning the semantic embedding model: extracting identification data from the initial identification system, constructing positive and negative sample pairs, and using the positive and negative sample pairs to fine-tune the semantic embedding model. Specifically, the model can be fine-tuned in three stages by using positive and negative sample pairs. A staged fine-tuning strategy (gradually increasing the difficulty of training) is adopted to optimize the semantic embedding model (such as Word2Vec, GloVe, BGE Embedding, etc.), specifically including:

[0096] In the first stage, semantically similar words are brought closer or further apart. The vectors of semantically similar words are brought closer together, and the vectors of semantically irrelevant words are moved further apart, preliminarily adjusting the semantic distance between words.

[0097] Preferably, the steps of bringing semantically similar words closer and further apart include:

[0098] (11) Positive sample pair selection: Select words with similar semantics as the first positive sample pair.

[0099] (12) Negative sample pair selection: Select semantically unrelated words as the first negative sample pair.

[0100] (13) Model fine-tuning: By adjusting the model parameters, the vector distance of the first positive sample pair is reduced and the vector distance of the first negative sample pair is increased.

[0101] In the second stage, nodes of the same level are aggregated and separated. The distance between word vectors of nodes of the same level is shortened for the first time, and samples with longer distances are pulled apart to strengthen the aggregation of words of the same level.

[0102] Preferably, the steps of aggregating and separating the peer nodes include:

[0103] (21) Positive sample pair selection: Select words at the same level as the second positive sample pair.

[0104] (22) Negative sample pair selection: Select the remote node words as the second negative sample pair.

[0105] (23) Model fine-tuning: The distance between word vectors of nodes at the same level is further shortened, while the distance between vectors of word pairs that are far apart (for example, the distance is greater than a preset threshold) is further extended.

[0106] The third stage is the optimization of the hierarchical structure. The distance between the word vectors of the nodes at the same level is shortened for the second time, and the child node vectors of the sibling nodes of their parent nodes are pulled apart to clarify the relationship between the hierarchical structure and words.

[0107] Preferably, the step of optimizing the hierarchical structure comprises:

[0108] (31) Positive sample pair selection: Select words at the same level as the third positive sample pair.

[0109] (32) Negative sample pair selection: Select nearby node words as the third negative sample pair.

[0110] (33) Model fine-tuning: The word vectors of nodes at the same level are shortened for the second time, and the word vectors of the child nodes of the sibling nodes of their parent nodes are further apart to strengthen the hierarchical structure.

[0111] Steps S206 to S207 are steps for integrating new identifiers and expanding the system, i.e., integrating new identifiers and expanding the system: using the fine-tuned semantic embedding model, semantically compare the newly appeared new digital content identifier with the identifier of the identifier system, determine the content-related nodes of the new digital content identifier in the tree structure, and locate the new digital content identifier at the nodes of the same level. Specifically, when a new identifier appears, the following steps are performed:

[0112] Step S206, semantic representation of the new identifier: Use the fine-tuned semantic embedding model to obtain the vector representation of the newly emerged new digital content identifier.

[0113] Step S207, similarity calculation: Calculate the similarity between the vector of the new digital content identifier and the vectors of the original words in the semantic embedding model. Cosine similarity calculation is preferably used.

[0114] Step S208, determination of the most relevant node: Determine the word node associated with the content of the new digital content identifier according to the result of the similarity calculation.

[0115] Step S209, position insertion: Insert the new digital content identifier as a sibling node of the content-associated word node at the corresponding sibling node of the tree structure.

[0116] System update: Complete the update and expansion of the identifier system, thereby realizing the dynamic update of the identifier system.

[0117] The following takes the newly emerged word "A" as an example to illustrate the integration process of the new identifier, as Figure 3 shown. The steps of the embodiment include:

[0118] I. Obtain the semantic representation of the new identifier

[0119] Use the fine-tuned semantic embedding model to process "A" to obtain its vector representation.

[0120] II. Calculate the similarity

[0121] Calculate the similarity between the vector of "A" and the vectors of all words in the identifier system.

[0122] III. Determine the most relevant word

[0123] The calculation result shows that the similarity between "A" and "B" is the highest, indicating that in the semantic space, the two are the closest and have a high semantic similarity.

[0124] IV. Insert into the identifier system

[0125] Insert "A" as a sibling node of "B" into the system tree, indicating that the two have a strong semantic correlation.

[0126] V. System update

[0127] The tree structure of the identifier system is updated, and a new node "A" is added, expanding the system.

[0128] The beneficial effects of this embodiment include:

[0129] 1. Multi-dimensional tree structure with clear relationships. By building a tree structure from different angles, the unstructured logo set is transformed into an organized system. This structure makes the relationship and hierarchy between logos clear at a glance.

[0130] 2. Large models generate data without the need for a large amount of basic data. Use pre-trained large language models to generate semantically related and semantically unrelated word data. Through the generation capability of large models, rich training samples can be obtained without the support of a large amount of basic data. Use data iteration optimization methods to retain key data samples, gradually improve data quality and model performance, and ensure that the model can effectively capture the semantic characteristics of words.

[0131] 3. Accurate semantic capture and efficient integration of new logos.

[0132] The semantic embedding model based on fine-tuning can accurately capture the deep semantic relationship of words. By using positive and negative sample pairs, the model is fine-tuned in three stages:

[0133] The first stage: bring the vectors of semantically similar words closer together and the vectors of semantically unrelated words farther apart, and preliminarily adjust the semantic distance between words.

[0134] The second stage: further shorten the distance between word vectors of nodes at the same level, and move samples with longer distances apart to strengthen the aggregation of words at the same level.

[0135] The third stage: continue to shorten the distance between word vectors of nodes at the same level, and further apart the child node vectors of the sibling nodes of their parent nodes, to clarify the relationship between the hierarchical structure and words.

[0136] Through such fine-tuning strategies, newly emerging logos can be accurately positioned in the appropriate position of the logo system, significantly improving the efficiency and accuracy of new logo integration.

[0137] 4. Dynamic evolution of the system and continuous updating.

[0138] The present invention adapts to the dynamic evolution of blockchain digital content. The tree structure can be continuously updated and expanded with the emergence of new identifiers, maintaining the vitality and timeliness of the identification system. Compared with the knowledge graph construction method that requires a lot of manpower and data support, the present invention has low complexity, is easy to implement, and can respond to changes in digital content in real time.

[0139] 5. Improve analysis and management efficiency.

[0140] By structuring the identification set, the present invention greatly improves the efficiency of analysis and management of blockchain digital content. Users can view and analyze identification from different angles, quickly understand the relationship between identifications, and assist in decision-making and supervision.

[0141] 6. The technology has strong applicability and high promotion value.

[0142] The method of the present invention does not require a large amount of basic data support and is applicable to the construction of blockchain digital content identification systems of various sizes and types. Its high efficiency and accuracy make it have a wide range of application prospects and can provide strong support for the organization and management of digital content.

[0143] In summary, the present invention proposes an innovative method in the field of identification system construction, which solves the problem that the existing technology cannot effectively build a blockchain digital content identification system. Through a multi-dimensional tree structure and a semantic embedding model based on fine-tuning, the present invention realizes the structured and relational management of identification sets, improves the efficiency of analysis and display, and has significant technical and application value.

[0144] It should be noted that the execution subject of the method for constructing a digital content identification system provided in the embodiment of the present invention may be an electronic device, a system for constructing a digital content identification system, or a control module in the system for constructing a digital content identification system for executing the method for constructing a digital content identification system. In the embodiment of the present invention, the system for constructing a digital content identification system provided in the embodiment of the present invention is described by taking the execution of the method for constructing a digital content identification system by the system for constructing a digital content identification system as an example.

[0145] Figure 4 1 is a schematic diagram of the structure of a digital content identification system construction system provided in the first embodiment of the present invention. The digital content identification system construction system 100 includes a data generation module 10, a system construction module 20, a system fine-tuning module 30 and a system update module 40, wherein:

[0146] The data generation module 10 is used to generate training data required for model training using a pre-trained large language model.

[0147] The system construction module 20 is used to construct an initial tree structure from different dimensions according to the training data to form an initial identification system.

[0148] The system fine-tuning module 30 is used to extract identification data from the initial identification system, construct positive and negative sample pairs, and use the positive and negative sample pairs to fine-tune the semantic embedding model.

[0149] The system update module 40 is used to compare the newly appeared new digital content identifier with the identifier of the identification system, determine the content-related node of the new digital content identifier in the tree structure, and locate the new digital content identifier at the node of the same level.

[0150] Figure 51 is a schematic diagram of a digital content identification system construction system according to a second embodiment of the present invention. The digital content identification system construction system 100 includes a data generation module 10, a system construction module 20, a system fine-tuning module 30 and a system update module 40, wherein:

[0151] The data generation module 10 is used to generate training data required for model training using a pre-trained large language model. Preferably, the data generation module 10 further includes:

[0152] The identifier collection submodule 11 is used to collect existing digital content identifiers to form an initial identifier set.

[0153] The word generation submodule 12 generates semantically relevant words and semantically irrelevant words of the digital content identifier using a large language model.

[0154] The data screening submodule 13 is used to eliminate worthless training data and retain valuable training data by combining manual review and model evaluation.

[0155] The data optimization submodule 14 is used to optimize the training data using a data iteration optimization algorithm.

[0156] The system construction module 20 is used to construct an initial tree structure from different dimensions according to the training data to form an initial identification system.

[0157] Preferably, the tree structure is used as an initial identification system to represent the hierarchy and relationship between identifications. And / or

[0158] The dimensions include content, subject and / or status.

[0159] The system fine-tuning module 30 is used to extract identification data from the initial identification system, construct positive and negative sample pairs, and use the positive and negative sample pairs to fine-tune the semantic embedding model. Preferably, the system fine-tuning module 30 further includes:

[0160] The first fine-tuning submodule 31 is used to bring the vectors of semantically similar words closer together and the vectors of semantically irrelevant words farther apart, thereby preliminarily adjusting the semantic distance between words. Preferably, the first fine-tuning submodule 31 is used to select semantically similar words as the first positive sample pair; select semantically irrelevant words as the first negative sample pair; and by adjusting the model parameters, reduce the vector distance of the first positive sample pair and increase the vector distance of the first negative sample pair.

[0161] The second fine-tuning submodule 32 is used to bring the word vectors of nodes at the same level closer for the first time, and to pull the samples with a longer distance away, so as to strengthen the aggregation of words at the same level. Preferably, the second fine-tuning submodule 32 is used to select words at the same level nodes as the second positive sample pair; select words at the remote node as the second negative sample pair; bring the word vectors of nodes at the same level closer for the first time, and at the same time pull the vector distance of the word pairs with a longer distance away.

[0162] The third fine-tuning submodule 33 is used to bring the word vectors of the nodes at the same level closer for the second time, and to pull the child node vectors of the sibling nodes of the parent node farther away, so as to clarify the relationship between the hierarchical structure and the words. Preferably, the third fine-tuning submodule 33 is used to select the words of the nodes at the same level as the third positive sample pair; select the words of the nearby nodes as the third negative sample pair; bring the word vectors of the nodes at the same level closer for the second time, and to pull the word vectors of the child nodes of the sibling nodes of the parent node farther away.

[0163] The system update module 40 is used to use the fine-tuned semantic embedding model to semantically compare the newly appeared new digital content identifier with the identifier of the identifier system, determine the content-related peer nodes of the new digital content identifier in the tree structure, and locate the new digital content identifier at the peer node.

[0164] Preferably, the system updating module 40 includes:

[0165] An acquisition submodule 41 is used to acquire a vector representation of a newly appeared new digital content identifier using the fine-tuned semantic embedding model;

[0166] A calculation submodule 42, for calculating the similarity between the vector of the new digital content identifier and the vector of the original word in the semantic embedding model;

[0167] The association submodule 43 is used to determine the word node associated with the content of the new digital content identifier according to the result of the similarity calculation;

[0168] The positioning submodule 44 is used to insert the new digital content identifier as a node at the same level of the content-related word node into the corresponding node at the same level of the tree structure.

[0169] The digital content identification system construction system 100 provided in the embodiment of the present invention can realize Figure 1 to Figure 6 To avoid repetition, the various processes implemented in the embodiment of the method for constructing the digital content identification system are not described here in detail.

[0170] The digital content identification system construction system provided by the embodiment of the present invention is based on the characteristics of virtual, abstract and dynamic evolution of digital content on the blockchain. The existing technology cannot effectively construct a structured identification system containing specific relationships between words. The present invention can construct an identification system that can accurately capture the deep semantic relationship between words without a large amount of basic data, making the identification set structured and relational, and convenient for analysis, display and continuous updating.

[0171] The present invention also provides a storage medium for storing Figure 1 to Figure 3 A computer program for the method of constructing any of the digital content identification systems. For example, a computer program instruction, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer, and can achieve the same technical effect. To avoid repetition, it will not be repeated here. The program instructions for calling the method of the present invention may be stored in a fixed or removable storage medium, and / or transmitted through a data stream in a broadcast or other signal-bearing medium and / or stored in a storage medium of a computer device that runs according to the program instructions.

[0172] According to one embodiment of the present invention, the present invention also provides a Figure 6 The electronic device 400 shown, the electronic device 400 optionally includes a storage medium 200 for storing a computer program and a processor 300 for executing the computer program, wherein when the computer program is executed by the processor 300, any of the above-mentioned methods for constructing a digital content identification system is implemented, triggering the electronic device 400 to execute the methods and / or technical solutions based on the aforementioned multiple embodiments, and can achieve the same technical effect, to avoid repetition, it is not repeated here. It should be noted that the electronic devices in the embodiments of the present invention include mobile electronic devices and non-mobile electronic devices. Exemplarily, the mobile electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, a wearable device, a super mobile personal computer, a netbook or a personal digital assistant, etc., and the non-mobile electronic device can be a server, a network attached storage (Network Attached Storage, NAS), a personal computer (personal computer, PC), a television (television, TV), a teller machine or a self-service machine, etc., which is not specifically limited in the embodiments of the present invention.

[0173] It should be noted that the present invention can be implemented in software and / or a combination of software and hardware, for example, can be implemented using an application specific integrated circuit (ASIC), a general purpose computer or any other similar hardware device. In one embodiment, the software program of the present invention can be executed by a processor to implement the above steps or functions. Similarly, the software program of the present invention (including related data structures) can be stored in a computer-readable recording medium, for example, a RAM memory, a magnetic or optical drive or a floppy disk and the like. In addition, some steps or functions of the present invention can be implemented using hardware, for example, as a circuit that cooperates with a processor to perform each step or function.

[0174] The present invention can be implemented on a computer as a computer-implemented method, or implemented in dedicated hardware, or implemented in a combination of the two. The executable code or part thereof for the method according to the present invention can be stored on a computer program product. Examples of computer program products include memory devices, optical storage devices, integrated circuits, servers, online software, etc. Optionally, the computer program product includes a non-temporary program code component stored on a computer-readable medium so as to perform the method according to the present invention when the program product is executed on a computer.

[0175] In an alternative embodiment, the computer program comprises computer program code means adapted to perform all the steps of the method according to the invention when the computer program is run on a computer. Optionally, the computer program is embodied on a computer readable medium.

[0176] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the method and device in the embodiment of the present invention is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0177] Of course, the present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for constructing a digital content identification system, characterized in that: include: Use the pre-trained large language model to generate the training data required for model training; Constructing an initial tree structure from different dimensions according to the training data to form an initial identification system; Extracting identification data from the initial identification system, constructing positive and negative sample pairs, and fine-tuning the semantic embedding model using the positive and negative sample pairs; By using the fine-tuned semantic embedding model, a semantic comparison is performed between the newly appeared digital content identifier and the identifier in the identifier system, the content-associated sibling nodes of the new digital content identifier in the tree structure are determined, and the new digital content identifier is positioned at the sibling nodes.

2. The method for constructing a digital content identification system according to claim 1, characterized in that: The step of using the pre-trained large language model to generate training data required for model training includes: Collect existing digital content identifiers to form an initial identifier set; Generate semantically relevant words and semantically irrelevant words of the digital content identifier using the large language model; Combine manual review and model evaluation to eliminate worthless training data and retain valuable training data.

3. The method for constructing a digital content identification system according to claim 1, characterized in that: The step of using the pre-trained large language model to generate training data required for model training also includes: The training data is optimized using a data iteration optimization algorithm.

4. The method for constructing a digital content identification system according to claim 1, characterized in that: The tree structure is used as an initial identification system to represent the hierarchy and relationship between identifications; and / or The dimensions include content, subject and / or status.

5. The method for constructing a digital content identification system according to claim 1, characterized in that: The step of fine-tuning the semantic embedding model using positive and negative sample pairs comprises: The step of bringing semantically similar words closer and farther apart brings the vectors of semantically similar words closer and the vectors of semantically irrelevant words farther apart, preliminarily adjusting the semantic distance between words; The aggregation and separation steps of nodes at the same level bring the word vectors of nodes at the same level closer for the first time, pull the samples with farther distances farther away, and strengthen the aggregation of words at the same level; The optimization step of the hierarchical structure brings the word vectors of nodes at the same level closer for the second time, and pulls the child node vectors of the sibling nodes of their parent nodes farther apart, clarifying the relationship between the hierarchical structure and words.

6. The method for constructing a digital content identification system according to claim 5, characterized in that: The steps of bringing semantically similar words closer and further apart include: Select words with similar semantics as the first positive sample pair; Select semantically irrelevant words as the first negative sample pair; By adjusting the model parameters, the vector distance of the first positive sample pair is reduced, and the vector distance of the first negative sample pair is increased; The steps of aggregating and separating the peer nodes include: Select the node words of the same level as the second positive sample pair; Select the remote node words as the second negative sample pair; The word vectors of nodes at the same level are brought closer for the first time, while the vector distances of word pairs that are farther apart are pulled farther apart; The optimization steps of the hierarchical structure include: Select the node words of the same level as the third positive sample pair; Select nearby node words as the third negative sample pair; The word vector distance of the nodes at the same level is shortened for the second time, and the word vector distance of the child node of the sibling node of its parent node is increased.

7. The method for constructing a digital content identification system according to claim 1, characterized in that: The step of using the fine-tuned semantic embedding model to semantically compare the newly appeared new digital content identifier with the identifier in the identifier system, determining the content-related peer nodes of the new digital content identifier in the tree structure, and locating the new digital content identifier at the peer nodes includes: Using the fine-tuned semantic embedding model, obtaining a vector representation of the newly appeared new digital content identifier; Calculate the similarity between the vector of the new digital content identifier and the vector of the original word in the identifier system; Determining, based on the result of the similarity calculation, a word node associated with the content of the new digital content identifier; The new digital content identifier is inserted into the corresponding node of the tree structure as a node of the same level as the word node associated with the content.

8. A system for constructing a digital content identification system based on the method according to any one of claims 1 to 7, characterized in that: include: The data generation module is used to generate the training data required for model training using the pre-trained large language model; A system construction module, used to construct an initial tree structure from different dimensions according to the training data to form an initial identification system; A system fine-tuning module, used to extract identification data from the initial identification system, construct positive and negative sample pairs, and use the positive and negative sample pairs to fine-tune the semantic embedding model; The system update module is used to use the fine-tuned semantic embedding model to semantically compare the newly appeared digital content identifier with the identifier in the identifier system, determine the content-related peer node of the new digital content identifier in the tree structure, and locate the new digital content identifier at the peer node.

9. A storage medium, characterized in that: A computer program for storing a method for executing the method for constructing a digital content identification system according to any one of claims 1 to 7.

10. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, the method for constructing a digital content identification system according to any one of claims 1 to 7 is implemented.

Images