Method and system for identifying and updating key content of power marketing policy file

Abstract

The invention discloses a method and a system for identifying and updating key content of a power marketing policy file. The method comprises the following steps: performing dependency syntactic analysis on an electricity marketing policy file, and extracting electricity marketing business characteristics in key business vocabularies; by applying a Word2Vec word vector model and combining the power marketing business characteristics, generating a business element preliminary word vector after scene adaptation; key business elements of the policy document are automatically extracted through BiLSTM and an attention mechanism; carrying out business element semantic coding by adopting a BERT pre-training model and a graph neural network to form a semantic network graph; establishing business element association between file versions through a cosine similarity algorithm, distributing version labels and change type labels, and generating a policy file dynamic association data model; and carrying out file marking and automatic updating and publishing. According to the invention, the problems of poor business element identification accuracy and low processing efficiency of the power marketing policy file are solved.

Description

Technical Field

[0001] This invention belongs to the field of information processing technology in the field of electricity marketing, and specifically relates to a method and system for identifying and updating key content of electricity marketing policy documents. Background Art

[0002] With the rapid development of the electricity market and the widespread adoption of smart grids, the complexity of electricity marketing has increased significantly. Frequent updates to various policy documents and increasingly complex policy provisions, coupled with different policy interpretation needs across various business scenarios, have made traditional document management and policy update methods unable to keep pace with evolving business demands.

[0003] In the current power marketing business, the following challenges are mainly faced: (1) A large number of policy documents and frequent updates: Various documents cover multiple aspects of power production, sales and management. The content of the documents often increases or changes at different times and in different scenarios, making it difficult to accurately respond to manual updates and manual management. This can easily lead to content delays or even inconsistent policy interpretations, affecting the compliance and efficiency of the business. (2) Complex content structure and high identification difficulty: Power marketing policy documents contain a large number of technical terms and rule details, and there are multiple levels of cross-references between different documents. Many key business elements in the documents are difficult to accurately identify and extract. Traditional document processing technology cannot effectively understand their deep semantics and business logic, resulting in the inability to systematize and structure the content, which in turn affects business decision-making and execution. (3) High difficulty in dynamic management and updates: Various policy documents in the power marketing business often involve a large number of business elements. When updating, it is necessary to accurately identify new elements or modify content and associate and replace them. Traditional management methods cannot efficiently and in real time handle changes in document content, making it difficult to achieve intelligent tracking and automatic updates of business content. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for identifying and updating key content in power marketing policy documents. The aim is to improve the processing efficiency and accuracy of power marketing business documents by intelligently extracting key content from policy documents and performing semantic abstraction, thereby achieving automatic identification, annotation, association modeling, and iterative updating of business elements in the documents.

[0005] To achieve the above-mentioned objectives, this invention provides a method and system for identifying and updating key content of electricity marketing policy documents, comprising:

[0006] Step S1: Perform dependency parsing on the electricity marketing policy document, use a business feature dictionary to parse the document's syntactic structure, extract electricity marketing business features from key business terms, generate a semantic vector with business feature annotations, and assign part-of-speech tags and business feature tags.

[0007] Step S2: Apply the Word2Vec word vector model and combine it with the characteristics of electricity marketing business to generate preliminary word vectors of business elements after scenario adaptation; through BiLSTM and attention mechanism, automatically extract key business elements of policy documents, assign business feature labels and attribute markers, construct a business element system and standard, and generate a set of quantitative expressions of business elements of policy documents.

[0008] Step S3: Use BERT pre-trained model and graph neural network (GNN) to perform semantic encoding of business elements and form a semantic network graph; use cosine similarity algorithm to establish the relationship between business elements between document versions, assign version labels and change type labels, and generate a dynamic association data model of policy documents. Combined with an automatic update mechanism, the structured management of the content of different version documents can be realized.

[0009] Step S4: Use adaptive difference detection to monitor and record changes in the content of policy documents, combine with multilayer perceptron for dynamic semantic adjustment, assign update markers and timestamps, generate updated iterative versions of the documents, and automatically publish them.

[0010] Furthermore, the specific process of step S1 includes:

[0011] Step S1.1: Use a dependency parser to process the sentences in the electricity marketing policy document to obtain structural information with syntactic dependency relations, and use a syntactic structure embedding model to form syntactic dependency vectors.

[0012] Step S1.2: Based on the syntactic dependency vector, extract the business features related to electricity marketing, perform semantic abstraction on the matched business features, and generate semantic vectors with business feature annotations.

[0013] Step S1.3: Assign part-of-speech tags and business feature tags to each word to ensure the accuracy of semantic information.

[0014] Furthermore, the specific process of step S1.1 includes:

[0015] Step S1.1.1: Use a dependency parser to perform lexical decomposition and dependency annotation on the sentences of the electricity marketing policy document to obtain the dependency structure between words in the sentence, including dependency relations and syntactic structure information;

[0016] Step S1.1.2: Using a syntactic structure embedding model, dependency relations and syntactic structure information are transformed into vector representations and semantically encoded at the lexical level to form syntactic dependency vectors;

[0017] Furthermore, the specific process of step S1.2 includes:

[0018] Step S1.2.1: Based on the syntactic dependency vector, the key business words in the sentence are matched and identified through the business feature dictionary to extract business features related to electricity marketing.

[0019] Step S1.2.2: Semantically abstract the matched business features to generate semantic vectors labeled with business features, thereby enhancing the high-level semantic expression of the file content.

[0020] Furthermore, the specific process of step S2 includes:

[0021] Step S2.1: Use the Word2Vec word vector model to embed the text content of the electricity marketing policy document to generate preliminary word vectors of business elements. Align the preliminary word vectors with the electricity marketing business feature dictionary to form preliminary word vectors of business elements after scenario adaptation.

[0022] Step S2.2: Using BiLSTM combined with an attention mechanism, feature enhancement is performed on the initial word vectors of the business elements after scenario adaptation, and key business elements of the power marketing policy document are automatically extracted.

[0023] Step S2.3: Assign corresponding business feature tags and attribute markers to each key business element to construct a business element system and standards;

[0024] Step S2.4: Using the constructed business element system and standards, all business elements of the electricity marketing policy documents are transformed into quantitative expressions of business elements, generating a set of quantitative expressions of business elements of policy documents.

[0025] Furthermore, the specific process of step S2.1 is as follows:

[0026] Step S2.1.1: Based on the pre-trained Word2Vec word vector model, each word in the text data of the electricity marketing policy document is vectorized to obtain the preliminary word vectors of the business elements.

[0027] Step S2.1.2: Through business scenario adaptive optimization, the preliminary word vectors of business elements are aligned with the power marketing business feature dictionary to ensure that the word vector expression adapts to the semantic features in the power marketing business scenario, thus forming the preliminary word vectors of business elements after scenario adaptation.

[0028] The specific process of step S2.2 is as follows:

[0029] Step S2.2.1: Input the initial word vectors of the business elements after scene adaptation into BiLSTM, capture the contextual semantic information of the business elements in the file through bidirectional encoding, and generate a vector representation of the business elements containing global dependencies.

[0030] Step S2.2.2: Based on the output of BiLSTM, an attention mechanism is applied to weight the features of each business element in order to strengthen the key elements related to electricity marketing business and automatically extract the key business elements of electricity marketing policy documents.

[0031] The specific process of step S2.3 is as follows:

[0032] Step S2.3.1: Based on the element types in the electricity marketing business characteristic dictionary, assign corresponding business characteristic tags to each key business element to enrich the business characteristic information of the element;

[0033] Step S2.3.2: Through quantitative processing, the attribute characteristics of key business elements are expressed in a numerical way, the attribute tags of each key business element are marked, and the business element system and standards are constructed.

[0034] Furthermore, the specific process of step S3 is as follows:

[0035] Step S3.1: Use the BERT pre-trained model to perform deep semantic vector encoding on each business element, and combine it with a graph neural network to connect the semantic vector nodes of different business elements into a semantic network graph.

[0036] Step S3.2: Using the cosine similarity algorithm, identify and associate similar or related business elements in the old and new version files, establish the association between business elements between versions, and ensure the consistency of business content in file iteration;

[0037] Step S3.3: Assign version tags and change type tags to each associated business element to record element changes during the document version update process;

[0038] Step S3.4: Construct a dynamic relational data model for policy documents based on key semantic relationships, and combine it with an automatic update mechanism to achieve structured management and consistent maintenance of the content of different versions of documents.

[0039] Furthermore, the specific process of step S3.1 is as follows:

[0040] Step S3.1.1: Use the BERT pre-trained model to encode the deep semantic vector of each business element and generate a high-dimensional semantic vector representation to capture the deep semantic features of the business elements.

[0041] Step S3.1.2: Construct a semantic relationship graph between business elements using a graph neural network, and connect the semantic vector nodes of different business elements into a semantic network graph;

[0042] The specific process of step S3.2 is as follows:

[0043] Step S3.2.1: Based on the cosine similarity algorithm, calculate the semantic similarity of different semantic vector nodes in the semantic network graph to identify business elements with similar or related semantics in the old and new versions of the file;

[0044] Step S3.2.2: Associate similar or related business elements in the old and new version files to establish business element associations between versions;

[0045] The specific process of step S3.3 is as follows:

[0046] Step S3.3.1: Assign a corresponding file version tag to each associated business element to identify the file version in which the business element is located;

[0047] Step S3.3.2: Assign a change type label to each business element based on the changes in the business elements;

[0048] The specific process of step S3.4 is as follows:

[0049] Step S3.4.1: Based on the semantic relationship of key business elements, construct a dynamic relational data model for electricity marketing policy documents to form a structured business content relationship, so as to support semantic consistency management between different versions of documents;

[0050] Step S3.4.2: Establish an automatic update mechanism for document association relationships to ensure that the association model is dynamically updated when business elements change, and maintain high consistency and accuracy of document content.

[0051] Furthermore, the specific process of step S4 is as follows:

[0052] Step S4.1: Use adaptive difference detection to monitor and record changes in the content of electricity marketing policy documents in order to monitor changes in business elements in real time;

[0053] Step S4.2: Based on the relationship between electricity marketing policy documents, use a multilayer perceptron to dynamically adjust the semantics of the changed business elements to ensure the consistency of the document content;

[0054] Step S4.3: Assign update markers and timestamps to each updated electricity marketing business element to track the update history of document content and support iterative management;

[0055] Step S4.4: Integrate the changed business elements, generate an automatically updated iterative version of the document, and realize the automated management and iterative release of the document content.

[0056] Furthermore, the specific process of step S4.1 is as follows:

[0057] Step S4.1.1: Use an adaptive difference detection algorithm to periodically scan the content of the electricity marketing policy document and automatically identify the differences in business elements between the old and new versions of the document, including newly added, modified or deleted content.

[0058] Step S4.1.2: For the identified differences, record the change type and change location for further association and update operations;

[0059] The specific process of step S4.2 is as follows:

[0060] Step S4.2.1: Input the changed business elements into the multilayer perceptron to generate an updated semantic vector representation, so that it is semantically consistent with the associated existing business elements;

[0061] Step S4.2.2: Reassign business feature tags and attribute tags to the updated business elements to ensure that the updated business elements have the same semantic attributes as the existing elements in the file.

[0062] The specific process of step S4.3 is as follows:

[0063] Step S4.3.1: Assign a unique update tag to each updated business element to identify its latest version in the file, which will facilitate subsequent tracking and management.

[0064] Step S4.3.2: Attach a timestamp record to each update operation to accurately mark the time of each update, supporting version control and historical traceability of business elements;

[0065] The specific process of step S4.4 is as follows:

[0066] Step S4.4.1: Integrate the updated business elements into the current version of the document to generate an automatically updated version of the document to reflect the latest business content and policy requirements;

[0067] Step S4.4.2: Establish an automated document release mechanism so that documents can be automatically released after being changed or updated, ensuring that business departments can obtain the latest document version in real time.

[0068] Secondly, the present invention provides a system for identifying and updating key content of electricity marketing policy documents, which includes a document feature identification module, a business element quantification module, a semantic association modeling module, and an automatic update and iteration module.

[0069] Document Feature Recognition Module: Used to perform dependency parsing on power marketing policy documents, parse the document's syntactic structure using a business feature dictionary, extract power marketing business features from key business terms, generate semantic vectors with business feature annotations, and assign part-of-speech tags and business feature tags.

[0070] Business Element Quantification Module: Applying the Word2Vec word vector model and combining it with the characteristics of electricity marketing business, it generates preliminary word vectors of business elements after scenario adaptation; through BiLSTM and attention mechanism, it automatically extracts key business elements of policy documents, assigns business feature labels and attribute markers, constructs a business element system and standards, and generates a set of quantitative expressions of business elements in policy documents.

[0071] Semantic association modeling module: It uses BERT pre-trained model and graph neural network to encode business elements semantically and form a semantic network graph; through cosine similarity algorithm, it establishes business element association between document versions, assigns version labels and change type labels, and generates a dynamic association data model of policy documents. Combined with an automatic update mechanism, it realizes the structured management of the content of different version documents.

[0072] Automatic update and iteration module: It uses adaptive difference detection to monitor and record changes in the content of policy documents, combines multilayer perceptrons to perform dynamic semantic adjustments, assigns update tags and timestamps, generates updated and iterative versions of the documents, and automatically publishes them.

[0073] Compared with the prior art, the present invention has the following technical effects:

[0074] Precise Content Recognition: This invention can automatically parse the syntactic structure and business features of power policy documents, achieving accurate identification of key content and high-level semantic expression. Compared with traditional manual annotation methods, this invention improves the efficiency and accuracy of document content extraction, laying a data foundation for subsequent quantification and management of business elements.

[0075] Precise Quantization and Feature Annotation: In the quantification and identification of business elements, a word vector model and BiLSTM combined with an attention mechanism are used to quantify and annotate business elements in the document, ensuring structured management of the content. Compared with traditional static text processing methods, this invention achieves automated quantification and feature annotation of business elements, providing support for automated document analysis and management.

[0076] Enhanced Semantic Consistency and Relevance: Through key semantic association modeling, this invention establishes semantic relationships between business elements across different versions of files, ensuring consistency of file content during iteration. The application of deep semantic matching and graph neural networks significantly improves the accuracy of business element associations and semantic consistency, thereby ensuring the integrity and accuracy of the file content update process.

[0077] Real-time updates and automatic publishing: In automatic update and iteration management, this invention ensures that changes in business elements are updated in real time and automatically trigger the publishing process through adaptive difference detection, multilayer perceptron (MLP), and an automated publishing mechanism. Compared with the traditional manual update mode, this invention significantly improves the timeliness and automation of document management, making the version update and publishing process of power policy documents more efficient and reliable, and effectively supporting the flexible response of power business and policy compliance management. Attached Figure Description

[0078] Figure 1 This is a flowchart of a method for identifying and updating key content in electricity marketing policy documents according to the present invention;

[0079] Figure 2 This is a diagram illustrating the composition of a system for identifying and updating key content in electricity marketing policy documents according to the present invention. Detailed Implementation

[0080] The technical solutions of the embodiments of the present invention will be explained and described below with reference to the accompanying drawings. However, the following embodiments are only preferred embodiments of the present invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments in the implementation methods without creative effort are all within the protection scope of the present invention.

[0081] Example 1

[0082] This embodiment provides a method for identifying and updating key content in electricity marketing policy documents. (See also...) Figure 1 As shown, the following steps are included:

[0083] Step S1, Policy Document Feature Identification: This step involves performing dependency parsing on electricity marketing policy documents. It uses a business feature dictionary to parse the document's syntactic structure, extracts electricity marketing business features from key business terms, generates semantic vectors with business feature annotations, and assigns part-of-speech tags and business feature tags. Specifically:

[0084] Step S1.1 involves processing the sentences in the electricity marketing policy document using a dependency parser to obtain structural information with syntactic dependency relations. A syntactic structure embedding model is then used to form syntactic dependency vectors. Specifically:

[0085] Step S1.1.1: Use a dependency parser to perform lexical decomposition and dependency annotation on the sentences of the electricity marketing policy document to obtain the dependency structure between words in the sentence, including dependency relations and syntactic structure information;

[0086] Given a sentence S = {w1, w2, ..., w i ,…,w n}, where w iLet w represent the i-th word. The dependency model applies to each word w. i Assign a dependency relation d(w) i This relation describes the i-th word w. i Dependence on other words in the sentence.

[0087] Step S1.1.2: Using a syntactic structure embedding model, dependency relations and syntactic structure information are transformed into vector representations and semantically encoded at the lexical level to form syntactic dependency vectors;

[0088] Syntactic Dependency Vector V syntax (S) can be expressed by the following formula:

[0089]

[0090] Where, α i It is the word w i The weights are determined by context information; f(d(w) i ), w i ) is an embedding function that transforms dependency relations and dependency syntax information into vectors.

[0091] Step S1.2: Based on the syntactic dependency vector, extract the business features related to electricity marketing, perform semantic abstraction on the matched business features, and generate semantic vectors with business feature annotations.

[0092] Specifically:

[0093] Step S1.2.1: Based on the syntactic dependency vector, the key business words in the sentence are matched and identified through the business feature dictionary to extract business features related to electricity marketing.

[0094] Let T be the business feature dictionary, if w i ∈T, then w i With business characteristic label t(w i ),Right now:

[0095]

[0096] Step S1.2.2: Semantically abstract the matched business features to generate semantic vectors labeled with business features, thereby enhancing the high-level semantic expression of the file content.

[0097] For each business feature label t(w) i Using semantic vector V semantic (t(w i ))express:

[0098] V semantic (t(w i))=g(t(w i ),V(w i ))

[0099] Where g is a semantic abstraction function, which abstracts the business feature label t(w) i ) and vocabulary w i The dependency vector V(w) i These can be combined to generate higher-level semantic vectors.

[0100] Step S1.3: Assign part-of-speech tags and business feature tags to each word to ensure the accuracy of semantic information.

[0101] Part-of-speech tag POS(w i ) and business feature tags t(w i The allocation function of ) can be expressed as:

[0102] POS(w i )=h(w i )

[0103] Here, h represents the part-of-speech tagging model. This model enriches the contextual semantic information and optimizes the relevance of sentences in semantic expression.

[0104] Step S2, Quantitative Identification of Business Elements in Policy Documents: Applying the Word2Vec word vector model and combining it with the characteristics of electricity marketing business, preliminary word vectors for business elements are generated after scenario adaptation. Through BiLSTM and an attention mechanism, key business elements of the policy documents are automatically extracted, business feature labels and attribute markers are assigned, a business element system and standards are constructed, and a set of quantitative expressions of business elements in the policy documents is generated. Specifically, as follows:

[0105] Step S2.1 involves embedding the text content of the electricity marketing policy document using the Word2Vec word vector model to generate preliminary word vectors for business elements. These preliminary word vectors are then aligned with the electricity marketing business feature dictionary to form scenario-adapted preliminary word vectors for the business elements. Specifically:

[0106] Step S2.1.1: Based on the pre-trained Word2Vec word vector model, each word in the text data of the electricity marketing policy document is vectorized to obtain the preliminary word vectors of the business elements.

[0107] Given vocabulary w i Its initial word vector representation is as follows:

[0108] V initial (w i =Word2Vec(w i )

[0109] Among them, Vinitial (w i ) represents the vocabulary w obtained through the pre-trained Word2Vec word vector model. i Preliminary word vectors.

[0110] Step S2.1.2: Through business scenario adaptive optimization, the preliminary word vectors of business elements are aligned with the power marketing business feature dictionary to ensure that the word vector expression adapts to the semantic features in the power marketing business scenario, thus forming the preliminary word vectors of business elements after scenario adaptation.

[0111] For example, for the word w i Preliminary word vector V initial (w i ), and pass it through the adaptive function f adapt Adjustments are made to generate preliminary word vectors V for business elements adapted to the specific scenario. adapt (w i ):

[0112] V adapt (w i )=f adapt (V initial (w i ), T)

[0113] Where T is the dictionary of characteristics of electricity marketing business, and the adaptive function f adapt The initial word vectors are optimized to generate semantic representations that better suit the business scenario, i.e., initial word vectors for business elements.

[0114] Step S2.2 involves using BiLSTM combined with an attention mechanism to enhance the features of the initial word vectors of business elements after scenario adaptation, automatically extracting key business elements from electricity marketing policy documents. Specifically:

[0115] Step S2.2.1: Input the initial word vectors of the business elements after scene adaptation into BiLSTM, capture the contextual semantic information of the business elements in the file through bidirectional encoding, and generate a vector representation of the business elements containing global dependencies.

[0116] The bidirectional encoding formula is as follows:

[0117]

[0118] in, and The hidden layer vectors from the forward and backward LSTM outputs are combined to form the business element vector representation H. i .

[0119] Step S2.2.2: Based on the BiLSTM output, an attention mechanism is applied to weight the features of each business element to strengthen the key elements related to electricity marketing business, and to automatically extract the key business element A from the electricity marketing policy document. i .

[0120] For each business element vector representation H i Its key business element A i The calculation formula is as follows:

[0121]

[0122] Where, α i The attention weight is based on the feature score (A) of the content related to the power business. i ) Perform calculations. Key business element A i The semantic importance of key business content was emphasized.

[0123] Step S2.3 involves assigning corresponding business feature tags and attribute markers to each key business element to construct a business element system and standards. Specifically:

[0124] Step S2.3.1: Based on the element types in the electricity marketing business characteristic dictionary, identify each key business element H... i Assign corresponding business characteristic tags (such as "electricity price discount", "policy applicable objects", etc.) to enrich the business characteristic information of the elements; for example, if A i If the expression is "electricity price discount", then assign the "policy item" feature label to complete its semantic information.

[0125] Step S2.3.2: Through quantitative processing, the attribute characteristics of key business elements are expressed in a numerical way, the attribute tags of each key business element are marked, and the business element system and standards are constructed.

[0126] For example, business feature label t(A) i ) and attribute tag p(A i ) are respectively encoded into numerical form encode(t(A) i )) and encode(p(A i )):

[0127] FA i =[A i encode(t(A) i )); encode(p(A i ))]

[0128] Among them, FA i Quantify the representation of business elements that include features and attributes.

[0129] Step S2.4: Using the constructed business element system and standards, all business elements of the electricity marketing policy documents are transformed into quantitative expressions of business elements, generating a set of quantitative expressions of business elements of policy documents.

[0130] Quantify all quantified business elements into FA i The set is a quantitative expression of the business elements of a document, V. file V file ={FA1, FA2, ..., FA n}, where n represents the number of business elements in the file. This quantitative expression can be used for automated management and subsequent analysis of file content.

[0131] Step 3: Identify the relationships between different versions of policy documents: A BERT pre-trained model and graph neural network are used to semantically encode business elements, forming a semantic network graph. Through a cosine similarity algorithm, relationships between business elements across document versions are established, version labels and change type labels are assigned, and a dynamic relationship data model of policy documents is generated. Combined with an automatic update mechanism, this achieves structured management of the content of different version documents. Specifically:

[0132] Step S3.1 involves using a BERT pre-trained model to perform deep semantic vector encoding on each business element, and then connecting the semantic vector nodes of different business elements into a semantic network graph using a graph neural network; specifically:

[0133] Step S3.1.1: Use the BERT pre-trained model to encode the deep semantic vector of each business element and generate a high-dimensional semantic vector representation to capture the deep semantic features of the business elements.

[0134] Let the business element be a i Then its deep semantic vector V BERT (a i ) is represented as:

[0135] V BERT (a i ) = BERT(a i )

[0136] The BERT model extracts a deep semantic vector for each element, which is then used for subsequent business element matching.

[0137] Step S3.1.2: Construct a semantic relationship graph between business elements using a graph neural network, and connect the semantic vector nodes of different business elements into a semantic network graph.

[0138] In a semantic network graph model, each node represents a business element, and each edge represents the semantic relationship between two elements. Assuming G = (V, E) is a semantic graph, where V is the set of nodes and E is the set of edges, then node v... i The update formula is:

[0139] v′ i =σ(∑ j∈N(i) f(V BERT (a i V BERT (a i ))·W)

[0140] Where N(i) is the set of neighboring nodes of node i, f is the node similarity function, W is the weight matrix, and σ is the activation function.

[0141] This semantic network graph model helps capture potential relationships between business elements to support the construction of semantic associations between policy document versions.

[0142] Step S3.2 involves using a cosine similarity algorithm to identify and associate similar or related business elements between old and new versions of the file, establishing relationships between business elements across versions to ensure consistency of business content during file iteration; specifically:

[0143] Step S3.2.1: Based on the cosine similarity algorithm, calculate the semantic similarity of different semantic vector nodes in the semantic network graph to identify business elements with similar or related semantics in the old and new versions of the file;

[0144] Given two business elements a i and b j The formula for calculating semantic similarity is as follows:

[0145]

[0146] This similarity measure measures the semantic proximity of similar business elements in the old and new versions of the file, and is used for initial business element matching.

[0147] Step S3.2.2: Associate similar or related business elements in the old and new version files to establish inter-version business element associations. If the similarity between two business elements is higher than a preset threshold θ, they are considered a match and associated in the old and new version files.

[0148] Step S3.3: Assign version tags and change type tags to each associated business element to record element changes during the document version update process; specifically:

[0149] Step S3.3.1, assign corresponding file version tags to each associated business element to identify the file version where the business element is located; for example, if the business element comes from a file of version 1, assign the tag Version(a i ) = V1; if it comes from version 2, assign the tag Version(a i ) = V2.

[0150] Step S3.3.2, according to the change situation of the business element (such as addition, modification or deletion), assign change type tags to each business element; if the element is newly added in the new version, assign the tag "New"; if the content is changed, assign the tag "Modified"; if it is deleted in the old version, assign the tag "Deleted", so as to record the evolution track of the file content.

[0151] Step S3.4, build a dynamic association data model of the policy document based on the key semantic association relationship, and combine with the automatic update mechanism to achieve structured management and consistency maintenance of the content of different version files. Specifically:

[0152] Step S3.4.1, based on the key business element semantic association relationship, build a dynamic association data model of the power marketing policy document to form a structured business content relationship to support semantic consistency management between different version files; let the dynamic association data model be M, which contains all associated business elements F(a i ), and its expression is as follows:

[0153] M = {F(a i ) | a i ∈ G version1 , G version2 , …, G versionN}

[0154] Among them, G versionN represents the business element set of each version.

[0155] Step S3.4.2, establish an automatic update mechanism for file association relationships to ensure that the association model is dynamically updated when business elements change. Whenever a business element changes, the association update is automatically triggered, so that the model always reflects the latest file content relationship and maintains high consistency and accuracy of the file content.

[0156] Step S4, automatic update iteration management: use adaptive difference detection to monitor the content changes of the policy document and record them, combine with a multi-layer perceptron for dynamic semantic adjustment, and assign update marks and timestamps to generate an updated iteration version of the file and automatically publish it.

[0157] Step S4.1, use adaptive difference detection to monitor the content changes of the power marketing policy document and record them to monitor the changes of business elements in real time; specifically:

[0158] Step S4.1.1: Use an adaptive difference detection algorithm to periodically scan the content of the electricity marketing policy document, and automatically identify the differences in business elements between the old and new versions of the document, including newly added, modified or deleted content.

[0159] Given the old and new versions of file F new and F old The adaptive difference detection algorithm calculates the content difference D as follows:

[0160]

[0161] Among them, a i As a new element, b j For deleted elements, the difference set D captures the changes in the document content.

[0162] Step S4.1.2: For the identified differences, record the change type (addition, modification, or deletion) and the change location for further association and update operations.

[0163] The change record for each element can be represented as a triple (a i (type, position), where type is the change type and position is the position index in the file.

[0164] Step S4.2, based on the correlation of electricity marketing policy documents, uses a multilayer perceptron (MLP) to dynamically adjust the semantics of changed business elements to ensure the consistency of document content; specifically:

[0165] Step S4.2.1: Input the changed business elements into the multilayer perceptron to generate an updated semantic vector representation, so that it is semantically consistent with the associated existing business elements;

[0166] Assume a′ i For the changed business elements, their updated semantic vector V MLP (a′ i The following is represented:

[0167] V MLP (a′ i ) = MLP(V BERTLP (a′ i ), C)

[0168] Where C represents the contextual information of the currently associated elements, and the MLP model dynamically adjusts the semantics of changing elements to ensure consistency; V BERTLP (a′ i ) represents the changed business element (a′) iSemantic vector representation generated by BERT model and linear projection.

[0169] Step S4.2.2: Reassign business feature labels and attribute labels to the updated business elements.

[0170] Assume a′ i The business feature label is t(a′) i ), the attribute tag is p(a′ i If the label is reallocated, then the reassignment is represented as:

[0171] t(a′ i )=t new ,p(a′ i ) = p new

[0172] This step ensures that the updated business elements are semantically and business-attribute consistent with the existing elements in the document.

[0173] Step S4.3: Assign an update tag and timestamp to each updated electricity marketing business element to track the update history of the document content and support iterative management; specifically:

[0174] Step S4.3.1: Assign a unique update tag to each updated business element to identify its latest version in the file;

[0175] For example, if a′ i For the updated business element, its update tag (a') can be represented as:

[0176] tag(a') = UUID(a')

[0177] UUID is a unique identifier that facilitates the tracking and management of versions of business elements.

[0178] Step S4.3.2: Attach a timestamp record to each update operation to accurately mark the time of each update.

[0179] Let the timestamp be T(a′) i If ), then the updated record is represented as (a′ i ,tag,T(a′ i This record helps to achieve version control and historical traceability of business elements.

[0180] Step S4.4: Integrate the changed business elements, generate an automatically updated version of the document, and realize automated management and iterative release of the document content; specifically:

[0181] Step S4.4.1: Integrate the updated business elements into the current version of the document to generate an automatically updated version of the document to reflect the latest business content and policy requirements;

[0182] Update version F updated Represented as:

[0183] F updated ={a′ i |a′ i ∈D and a′ i is updated}∪{a′ i |a′ i ∈F current and unchanged}

[0184] Among them, F current For the current file version, F updated This indicates the latest version of the file.

[0185] Step S4.4.2: Establish an automated document release mechanism so that documents can be automatically released after being changed or updated, ensuring that business departments can obtain the latest document version in real time.

[0186] The automatic publishing mechanism is represented as:

[0187] Publish(F updated )

[0188] An automated publishing mechanism ensures the timeliness and consistency of file management.

[0189] Example 2

[0190] This embodiment provides a system for identifying and updating key content of electricity marketing policy documents, such as... Figure 2 As shown, it includes a file feature recognition module, a business element quantification module, a semantic association modeling module, and an automatic update and iteration module.

[0191] The document feature recognition module is used to parse key content in power policy documents. It employs dependency parsing and business feature dictionary matching techniques to achieve hierarchical parsing of document content, extracting business-related semantic information and generating structured high-level semantic expressions. This module implements the function of step 1 in Example 1, and will not be elaborated further here.

[0192] The business element quantification module is used to quantify and label the business elements in the file. Through a word vector model, BiLSTM, and attention mechanism, it transforms the business elements into quantified representations and assigns corresponding feature labels and attribute tags. This module ensures the structured management and traceability of business elements across different file versions, specifically implementing the function of step 2 in Example 1, which will not be elaborated further here.

[0193] The semantic association modeling module is used to construct the semantic relationships between business elements in different versions of files. It utilizes a BERT pre-trained model and GNN to generate a semantic graph between business elements, and identifies and associates similar or related business elements in old and new files through a deep semantic matching algorithm, ensuring semantic consistency and relevance of file content across version iterations. This module implements the function of step 3 in Example 1, and will not be elaborated further here.

[0194] The automatic update and iteration module monitors changes to file content. It uses an adaptive difference detection algorithm to detect changes in business elements within the file in real time and employs an MLP model for semantic adjustment to ensure that the updated business elements remain consistent with existing file elements. This module can also automatically generate the latest iteration of the file and trigger the release process, ensuring that the power marketing department receives the latest file version in real time. This module implements the function in step 4 of Example 1, and will not be elaborated further here.

[0195] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for identifying and updating key content in electricity marketing policy documents, characterized in that, include: Step S1: Perform dependency parsing on the electricity marketing policy document, use a business feature dictionary to parse the document's syntactic structure, extract electricity marketing business features from key business terms, generate a semantic vector with business feature annotations, and assign part-of-speech tags and business feature tags. Step S2: Apply the Word2Vec word vector model and combine it with the characteristics of electricity marketing business to generate preliminary word vectors of business elements after scenario adaptation; through BiLSTM and attention mechanism, automatically extract key business elements of policy documents, assign business feature labels and attribute markers, construct a business element system and standard, and generate a set of quantitative expressions of business elements of policy documents. Step S3: Use BERT pre-trained model and graph neural network to perform semantic encoding of business elements and form a semantic network graph; use cosine similarity algorithm to establish business element association between document versions, assign version labels and change type labels, and generate a dynamic association data model of policy documents. Combined with an automatic update mechanism, achieve structured management of the content of different version documents. Step S4: Use adaptive difference detection to monitor and record changes in the content of policy documents, combine with multilayer perceptron for dynamic semantic adjustment, assign update markers and timestamps, generate updated iterative versions of the documents, and automatically publish them.

2. The method for identifying and updating key content of electricity marketing policy documents according to claim 1, characterized in that, The specific process of step S1 includes: Step S1.1: Use a dependency parser to process the sentences in the electricity marketing policy document to obtain structural information with syntactic dependency relations, and use a syntactic structure embedding model to form syntactic dependency vectors. Step S1.2: Based on the syntactic dependency vector, extract the business features related to electricity marketing, perform semantic abstraction on the matched business features, and generate semantic vectors with business feature annotations. Step S1.3: Assign part-of-speech tags and business feature tags to each word.

3. The method for identifying and updating key content of electricity marketing policy documents according to claim 2, characterized in that, The specific process of step S1.1 includes: Step S1.1.1: Use a dependency parser to perform lexical decomposition and dependency annotation on the sentences of the electricity marketing policy document to obtain the dependency structure between words in the sentence, including dependency relations and syntactic structure information; Step S1.1.2: Using a syntactic structure embedding model, dependency relations and syntactic structure information are transformed into vector representations and semantically encoded at the lexical level to form syntactic dependency vectors; The specific process of step S1.2 includes: Step S1.2.1: Based on the syntactic dependency vector, the key business words in the sentence are matched and identified through the business feature dictionary to extract business features related to electricity marketing. Step S1.2.2: Semantically abstract the matched business features to generate semantic vectors with business feature annotations.

4. The method for identifying and updating key content of electricity marketing policy documents according to claim 1, characterized in that, The specific process of step S2 includes: Step S2.1: Use the Word2Vec word vector model to embed the text content of the electricity marketing policy document to generate preliminary word vectors of business elements. Align the preliminary word vectors with the electricity marketing business feature dictionary to form preliminary word vectors of business elements after scenario adaptation. Step S2.2: Using BiLSTM combined with an attention mechanism, feature enhancement is performed on the initial word vectors of the business elements after scenario adaptation, and key business elements of the power marketing policy document are automatically extracted. Step S2.3: Assign corresponding business feature tags and attribute markers to each key business element to construct a business element system and standards; Step S2.4: Using the constructed business element system and standards, all business elements of the electricity marketing policy documents are transformed into quantitative expressions of business elements, generating a set of quantitative expressions of business elements of policy documents.

5. The method for identifying and updating key content of electricity marketing policy documents according to claim 4, characterized in that, The specific process of step S2.1 is as follows: Step S2.1.1: Based on the pre-trained Word2Vec word vector model, each word in the text data of the electricity marketing policy document is vectorized to obtain the preliminary word vectors of the business elements. Step S2.1.2: Through business scenario adaptive optimization, the preliminary word vectors of business elements are aligned with the power marketing business feature dictionary to form preliminary word vectors of business elements after scenario adaptation. The specific process of step S2.2 is as follows: Step S2.2.1: Input the initial word vectors of the business elements after scene adaptation into BiLSTM, capture the contextual semantic information of the business elements in the file through bidirectional encoding, and generate a vector representation of the business elements containing global dependencies. Step S2.2.2: Based on the output of BiLSTM, an attention mechanism is applied to weight the features of each business element, and key business elements of the electricity marketing policy document are automatically extracted. The specific process of step S2.3 is as follows: Step S2.3.1: Assign corresponding business feature labels to each key business element according to the element type in the electricity marketing business feature dictionary; Step S2.3.2: Through quantitative processing, the attribute characteristics of key business elements are expressed in a numerical way, the attribute tags of each key business element are marked, and the business element system and standards are constructed.

6. The method for identifying and updating key content of electricity marketing policy documents according to claim 1, characterized in that, The specific process of step S3 is as follows: Step S3.1: Use the BERT pre-trained model to perform deep semantic vector encoding on each business element, and combine it with a graph neural network to connect the semantic vector nodes of different business elements into a semantic network graph. Step S3.2: Using the cosine similarity algorithm, identify and associate similar or related business elements in the old and new version files to establish the association between business elements between versions; Step S3.3: Assign version tags and change type tags to each associated business element to record element changes during the document version update process; Step S3.4: Construct a dynamic relational data model for policy documents based on key semantic relationships, and combine it with an automatic update mechanism to achieve structured management and consistent maintenance of the content of different versions of documents.

7. The method for identifying and updating key content of electricity marketing policy documents according to claim 6, characterized in that, The specific process of step S3.1 is as follows: Step S3.1.1: Use the BERT pre-trained model to perform deep semantic vector encoding on each business element to generate a high-dimensional semantic vector representation; Step S3.1.2: Construct a semantic relationship graph between business elements using a graph neural network, and connect the semantic vector nodes of different business elements into a semantic network graph; The specific process of step S3.2 is as follows: Step S3.2.1: Based on the cosine similarity algorithm, calculate the semantic similarity of different semantic vector nodes in the semantic network graph to identify business elements with similar or related semantics in the old and new versions of the file; Step S3.2.2: Associate similar or related business elements in the old and new version files to establish business element associations between versions; The specific process of step S3.3 is as follows: Step S3.3.1: Assign a corresponding file version tag to each associated business element to identify the file version in which the business element is located; Step S3.3.2: Assign a change type label to each business element based on the changes in the business elements; The specific process of step S3.4 is as follows: Step S3.4.1: Based on the semantic relationship of key business elements, construct a dynamic relational data model of electricity marketing policy documents to form a structured business content relationship; Step S3.4.2: Establish an automatic update mechanism for file association relationships to ensure that the association model is dynamically updated when business elements change.

8. The method for identifying and updating key content of electricity marketing policy documents according to claim 1, characterized in that, The specific process of step S4 is as follows: Step S4.1: Monitor and record changes in the content of electricity marketing policy documents using adaptive difference detection; Step S4.2: Based on the relationship between electricity marketing policy documents, use a multilayer perceptron to dynamically adjust the semantics of the changed business elements; Step S4.3: Assign update markers and timestamps to each updated electricity marketing business element to track the update history of document content and support iterative management; Step S4.4: Integrate the changed business elements, generate an automatically updated iterative version of the document, and realize the automated management and iterative release of the document content.

9. The method for identifying and updating key content of electricity marketing policy documents according to claim 8, characterized in that, The specific process of step S4.1 is as follows: Step S4.1.1: Use an adaptive difference detection algorithm to periodically scan the content of the electricity marketing policy document and automatically identify the differences in business elements between the old and new versions of the document, including newly added, modified or deleted content. Step S4.1.2: For the identified differences, record the change type and change location; The specific process of step S4.2 is as follows: Step S4.2.1: Input the changed business elements into the multilayer perceptron to generate an updated semantic vector representation, so that it is semantically consistent with the associated existing business elements; Step S4.2.2: Reassign business feature tags and attribute tags to the updated business elements to ensure that the updated business elements have the same semantic attributes as the existing elements in the file. The specific process of step S4.3 is as follows: Step S4.3.1: Assign a unique update tag to each updated business element to identify its latest version in the file; Step S4.3.2: Attach a timestamp record to each update operation to accurately mark the time of each update, supporting version control and historical traceability of business elements; The specific process of step S4.4 is as follows: Step S4.4.1: Integrate the updated business elements into the current version of the file to generate an automatically updated iterative version of the file; Step S4.4.2: Establish an automated file publishing mechanism so that files can be automatically published after being changed or updated.

10. A system for identifying and updating key content of electricity marketing policy documents, characterized in that, It includes a document feature recognition module, a business element quantification module, a semantic association modeling module, and an automatic update and iteration module; Document Feature Recognition Module: Used to perform dependency parsing on power marketing policy documents, parse the document's syntactic structure using a business feature dictionary, extract power marketing business features from key business terms, generate semantic vectors with business feature annotations, and assign part-of-speech tags and business feature tags. Business Element Quantification Module: Applying the Word2Vec word vector model and combining it with the characteristics of electricity marketing business, it generates preliminary word vectors of business elements after scenario adaptation; through BiLSTM and attention mechanism, it automatically extracts key business elements of policy documents, assigns business feature labels and attribute markers, constructs a business element system and standards, and generates a set of quantitative expressions of business elements in policy documents. Semantic association modeling module: It uses BERT pre-trained model and graph neural network to encode business elements semantically and form a semantic network graph; through cosine similarity algorithm, it establishes business element association between document versions, assigns version labels and change type labels, and generates a dynamic association data model of policy documents. Combined with an automatic update mechanism, it realizes the structured management of the content of different version documents. Automatic update and iteration module: It uses adaptive difference detection to monitor and record changes in the content of policy documents, combines multilayer perceptrons to perform dynamic semantic adjustments, assigns update tags and timestamps, generates updated and iterative versions of the documents, and automatically publishes them.

Images