Method and system for constructing power sensitive information theme recognition LDA model, medium and processor

Abstract

The invention discloses a method, a system, a medium and a processor for constructing a power sensitive information theme identification LDA model, and relates to the technical field of information intelligent identification. The method comprises the following steps: acquiring basic sensitive vocabularies in a power system; obtaining a sensitive vocabulary list according to similarity analysis of a preset target category sensitive word and each sensitive word in the sensitive vocabularies; performing theme division on the sensitive words in the sensitive word list according to a preset division rule to obtain a plurality of sensitive word sets with different themes; and inputting the sensitive word set of each topic as a subset into an LDA model for learning training, and generating a topic recognition LDA model with power sensitive information topic recognition output. According to the method, through deep mining and analysis of the text data of the power industry and combination of the topic recognition algorithm of the LDA model, the sensitive information topic is accurately recognized and extracted, the information security of the power industry is ensured, and the intelligence of the power industry is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent information identification, and in particular to a method, system, medium and processor for constructing an LDA model for power-sensitive information topic identification. Background Art

[0002] With the continuous advancement of information technology, the generation and circulation of information related to the power industry are becoming more and more frequent, including a large amount of sensitive information, such as power supply security, policies and regulations, market changes, etc. Once these sensitive information are leaked or used by criminals, they may have a serious impact on the normal operation of the power system, market order and public safety. Therefore, it is particularly important to effectively identify the subject of sensitive information in the power industry.

[0003] In recent years, deep learning-based methods have been widely used in the identification and protection of sensitive information in power systems. The performance of deep learning algorithms is highly dependent on the quantity and quality of training data. However, in the identification of sensitive information in power systems, it is difficult to obtain a large amount of data with accurate annotations. Therefore, the use of deep learning-based methods for the identification of sensitive information in power systems has limitations.

[0004] In view of this, it is necessary to provide a method, system, medium and processor for constructing an LDA model for power-sensitive information topic identification. Summary of the invention

[0005] In view of the difficulty in obtaining training data for the deep learning-based method used in the prior art to identify power sensitive information, the present invention provides a method, system, medium and processor for constructing an LDA model for identifying power sensitive information topics, which can accurately identify and extract sensitive information topics, ensure the information security of the power industry, and promote the intelligence of the power industry. The specific technical solution is as follows:

[0006] A method for constructing an LDA model for identifying power sensitive information topics, comprising:

[0007] Obtain basic sensitive words in the power system;

[0008] Obtaining a sensitive word list based on similarity analysis between a preset target category sensitive word and each sensitive word in the sensitive words;

[0009] The sensitive words in the sensitive word list are divided into topics according to a preset division rule to obtain a plurality of sensitive word sets of different topics;

[0010] The sensitive word set of each topic is input as a subset into the LDA model for learning and training, generating a topic identification LDA model with output of power sensitive information topic identification.

[0011] Preferably, the basic sensitive words in the power system are obtained including:

[0012] Obtain text information of power-related data in the power system and perform data cleaning;

[0013] Performing text segmentation processing on the cleaned text information;

[0014] The segmented text is processed according to the created stop word list to obtain basic sensitive words.

[0015] Preferably, the obtaining of the sensitive word list based on similarity analysis between the preset target category sensitive word and each sensitive word in the sensitive words comprises:

[0016] The similarity between the preset target category sensitive words and the word embedding of each sensitive word of the sensitive words is calculated, and a preset number of seed words are screened out according to the similarity score results, and a sensitive word list is obtained by sorting.

[0017] Preferably, the calculating of the similarity between the preset target category sensitive words and the word embedding of each sensitive word of the sensitive words, screening out a preset number of seed words according to the similarity score result, and arranging to obtain a sensitive word list includes:

[0018] The Word2vec word embedding model is selected to respectively train the target category sensitive words and each sensitive word of the sensitive words, so as to obtain the word embedding vector of the target category sensitive words and the embedding vector of each sensitive word of the sensitive words;

[0019] Calculating the cosine similarity between the word embedding vector of the target category sensitive word and the embedding vector of each sensitive word of the sensitive word;

[0020] According to the cosine similarity score, a preset number of new seed words corresponding to the target category sensitive words are screened out, and a sensitive word list is obtained.

[0021] Preferably, the step of screening out a preset number of new seed words corresponding to the target category sensitive words according to the cosine similarity score, and arranging the sensitive word list includes:

[0022] Sorting the cosine similarity score results in descending order;

[0023] According to a preset cosine similarity threshold, a preset number of seed words that are greater than the cosine similarity threshold and are ranked at the top are selected;

[0024] The seed words are sorted to obtain a sensitive word list.

[0025] Preferably, the sensitive vocabulary list is divided into topics according to a preset division rule to obtain multiple groups of sensitive word sets of different topic categories:

[0026] When identifying words with obvious markings in the sensitive vocabulary, classify them according to the obvious markings to obtain a set of sensitive words corresponding to the marked subject category;

[0027] When it is identified that there are no obvious identifying words in the sensitive vocabulary, classification is performed according to a coarse-grained classification method to obtain a plurality of sensitive word sets of different subject categories.

[0028] A system for constructing an LDA model for identifying power-sensitive information topics, applied to the aforementioned method for constructing an LDA model for identifying power-sensitive information topics, comprises:

[0029] A vocabulary acquisition unit, used to acquire basic sensitive words in the power system;

[0030] A vocabulary list integration unit, used for obtaining a sensitive vocabulary list according to similarity analysis between a preset target category sensitive word and each sensitive word in the sensitive vocabulary;

[0031] A topic classification unit, used to divide the sensitive vocabulary into topics according to a preset division rule to obtain a plurality of sensitive word sets of different topics;

[0032] The topic identification model generation unit is used to input the sensitive word set of each topic as a subset into the LDA model for learning and training, and generate a topic identification LDA model with power sensitive information topic identification output.

[0033] A computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the aforementioned method for constructing an LDA model for power-sensitive information topic identification.

[0034] A processor is used to run a program, wherein the program executes the aforementioned method for constructing an LDA model for power-sensitive information topic identification when running.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] A method for constructing an LDA model for identifying electric power sensitive information topics of the present invention obtains basic sensitive words in the electric power system; obtains a sensitive word list based on similarity analysis between preset target category sensitive words and each sensitive word in the sensitive words; divides the sensitive words in the sensitive word list into topics according to preset division rules to obtain sensitive word sets of multiple different topics; inputs the sensitive word set of each topic as a subset into the LDA model for learning and training, and generates a topic identification LDA model with output of electric power sensitive information topic identification. The present invention can more accurately identify and extract sensitive information topics through deep mining and analysis of text data in the electric power industry, combined with the topic identification algorithm of the LDA model, to provide strong support for the safe development and information management of the electric power industry, ensure the information security of the electric power industry, and promote the intelligence of the electric power industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0038] Figure 1 This is a flow chart of a method for constructing an LDA model for identifying power-sensitive information topics according to the present invention.

[0039] Figure 2 This is a flow chart of another embodiment of a method for constructing an LDA model for identifying power-sensitive information topics according to the present invention.

[0040] Figure 3 This is a flow chart of another embodiment of a method for constructing an LDA model for identifying power-sensitive information topics according to the present invention.

[0041] Figure 4 For the present invention Figure 2 This is a schematic diagram of a system for constructing an LDA model for power-sensitive information topic identification according to the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0044] It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0045] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0046] See the following examples Figures 1 to 4 .

[0047] The present application embodiment provides a method for constructing an LDA model for identifying power-sensitive information topics, including:

[0048] Step S1, obtaining basic sensitive words in the power system;

[0049] By using customized web crawlers to collect text data from within the power company and text information from the power company’s website, including customer data, private data, and company data, and obtaining a power data corpus, basic sensitive vocabulary in the power industry is obtained based on the collected text document information.

[0050] Specifically, the basic sensitive words in the power system are obtained including:

[0051] Obtain text information of power-related data in the power system and perform data cleaning. Identify and delete duplicate text data;

[0052] Performing text segmentation processing on the cleaned text information;

[0053] In order to better handle the professional terms and sensitive words in power-related data, the NLPIR system is used to segment Chinese text. The NLPIR system is a powerful tool that can not only segment Chinese words, but also support functions such as part-of-speech tagging, named entity recognition, new word recognition, and keyword extraction.

[0054] The segmented text is processed according to the created stop word list to obtain basic sensitive words.

[0055] By creating a stop word list, which contains some meaningless words and punctuation marks (such as commonly used but meaningless words like "de" (的), "shi" (是), etc.). These words and characters usually do not have practical meanings and will affect the accuracy of word segmentation results. Therefore, they are removed from the segmented text to ensure that the word segmentation results are clearer and more reliable.

[0056] Step S2: Obtain a sensitive word list based on the similarity analysis between the preset target category sensitive words and each sensitive word in the sensitive words.

[0057] The preset target category sensitive words can be summarized and classified according to the actual usage content and nature, including common categories such as technology and equipment sensitive words, policy and regulation sensitive words, environmental protection and safety sensitive words, and publicity and advertising sensitive words, etc. For example, technology and equipment sensitive words like "thermal power technology", "smart grid", "ultra-high voltage power transmission", "new energy grid connection standard", etc. These words involve the core technical standards and equipment requirements of the power industry.

[0058] The obtaining of the sensitive word list based on the similarity analysis between the preset target category sensitive words and each sensitive word in the sensitive words includes:

[0059] Calculate the similarity between the preset target category sensitive words and the word embeddings of each sensitive word in the sensitive words, and screen out a preset number of seed words according to the similarity score results, and organize to obtain the sensitive word list.

[0060] Word embeddings contain rich semantic information, and the semantic distance between words can be represented by calculating the similarity between word embeddings.

[0061] Step S3: Perform topic partitioning on the sensitive words in the sensitive word list according to the preset partitioning rules to obtain multiple sensitive word sets with different topics.

[0062] Convert the sensitive word vocabulary into word vectors and expand the sensitive words under different categories through similarity calculation. To improve the efficiency of sensitive information topic recognition and the robustness of the model, and improve the model effect.

[0063] Specifically, the performing of topic partitioning on the sensitive word list according to the preset partitioning rules to obtain multiple sensitive word sets with different topic categories:

[0064] When there are obvious identification words in the sensitive word list, classify according to the obvious identification to obtain the sensitive word set corresponding to the identification topic category;

[0065] When there are no obvious identification words in the sensitive word list, classify according to the coarse-grained classification method to obtain multiple sensitive word sets with different topic categories.

[0066] Among them, for data with obvious words, such as "customer", "company", "user", etc., they can be classified according to their obvious identification. For the remaining data without obvious identification, they are divided according to the coarse-grained classification method proposed in this article. Through coarse-grained division, the text data can be preprocessed first to reduce the amount of data that the LDA model needs to process, thereby reducing the computational complexity and increasing the training speed. By coarse-grained division of the data, the data can be divided into multiple subsets, each of which may have stronger correlations. Doing so helps reduce the computational complexity of the LDA model and avoids the problem of excessive computation caused by directly inputting all the data into the model. When running the LDA model on each subset, the model will converge faster and find potential topics more accurately due to the reduced amount of data and stronger correlation.

[0067] Step S4: Input the sensitive word set of each topic as a subset into the LDA model for learning and training, and generate a topic identification LDA model with power sensitive information topic identification output.

[0068] The topic identification LDA model with power-sensitive information topic identification output is generated and deployed in practical applications, such as power-sensitive information monitoring system, to identify topics for new documents and take corresponding security measures based on the identification results.

[0069] In this embodiment, the coarse-grained division can also be customized according to the characteristics of the data and application requirements. For example, in the power data, it can be divided according to the data security level, and personal data, company data, national data, etc. can be processed separately. This customized division helps to improve the flexibility and accuracy of the model, so that each subset can better capture the subject information in a specific field. In this way, we can more effectively identify sensitive information topics, thereby better protecting data security and privacy.

[0070] The LDA model is a probabilistic generation model. It treats documents as bags of words, that is, documents are treated as a group of words, ignoring any grammatical or lexical order relationships. The LDA model is a three-layer hierarchical Bayesian model including document layer, topic layer and word layer, which can map high-dimensional text sets to low-dimensional latent semantic space. The distribution of documents to topics and the distribution of topics to words obey multinomial distribution. Implicit topics are regarded as soft clusters of word features. The LDA model achieves the generalization of text information at a more abstract level. The topic generation process is as follows:

[0071] p(θ m )=Dir(θ m |α)m=1,…,M

[0072] p(z m，n |θm )=Mult(z m，n |θ m )n=1,…,N m

[0073] p(w m，n |z m，n ,β)=Mult(w m，n |βz m，n )n=1,…,N m

[0074] Where Dir(·|·) represents the Dirichlet distribution function; Mult(·|·) represents the multinomial distribution function; M is the total number of documents. Nm is the total number of words in the mth document. β is the Dirichlet prior parameter of the multinomial distribution of words in each topic, and α is the Dirichlet prior parameter of the multinomial distribution of topics under each document. m，n W represents the topic of the nth word in the mth document. m，n is the nth word in the mth document, θ m Represents the topic distribution under the mth document.

[0075] In the LDA model, the following variables are included: k is the number of topics; M is the total number of documents; N is the total number of documents. d is the number of words in document d; α is the prior parameter of topic distribution; β is the prior parameter of word distribution; Z di The i-th word in document d; W di The i-th word in document d; θ d is the topic distribution of document d; φ k is the word distribution of topic k.

[0076] For each word w di Theme di , its conditional probability is:

[0077]

[0078] in, Indicates that in document d, except for w di In addition, the number of words assigned to topic k; Indicates that in document d, except for W di The total number of words other than represents the number of times word v appears in topic k, excluding Wdi; Indicates that in topic k, except for w di The total number of words other than .

[0079] Based on the above explanation of the LDA model, the improved sensitive word weighted LDA model is adopted in this embodiment. The sensitive topic word distribution can make up for the shortcomings of LDA in identifying the co-occurrence relationship of low-frequency words and increase the distribution weight of words with low co-occurrence frequency but obvious sensitive characteristics. In this way, more words with low-frequency implicit relationships can be discovered.

[0080] The sensitive word weighted LDA model adds a constraint variable δ based on the standard LDA model. The calculation of model parameters mainly includes the calculation of document-topic distribution θ and topic-word distribution φ. The model is derived using the folding and Gibbs sampling method based on the Monte Carlo Markov chain method. The folding Gibbs sampling method is a Markov chain Monte Carlo method for latent variable models (such as latent Dirichlet allocation LDA). The folding Gibbs sampling formula is as follows:

[0081]

[0082] Among them, p(w i |z i ,δ) is given a topic z i and parameter δ, the i-th word w i The probability of represents the number of words belonging to topic k except the i-th word; represents the number of words except the i-th word in document d; V is the size of the vocabulary; k is the total number of topics.

[0083] In this embodiment, the process of generating documents by using the sensitive word weighted LDA (Latent Dirichlet Allocation) model is as follows:

[0084] For each document: Select a topic distribution θ that follows a Dirichlet distribution (α). For each word w in the document j (j ranges from 1 to N), if w j belongs to the sensitive word set Ssen, select the topic distribution φs of the sensitive word, which follows the η-fold Dirichlet distribution (β) (η is a weight factor); otherwise select the topic distribution φ, which follows the Dirichlet distribution (β). For each word w in document d i Location, select theme z n , which follows the topic distribution θ. If w j Belongs to the sensitive word set Ssen, select w from the topic distribution φs of sensitive words j Otherwise, select w from the topic distribution φ i .

[0085] For the topic model, the parameters are set as follows: a = 50 / K, where K is the number of topics; β = 0.01, the number of iterations is 100; the value of n-top is 15, that is, the top 15 words with the highest probability values ​​for each topic are regarded as topic words.

[0086] Perplexity is used as an evaluation criterion for topic models. Perplexity is a metric in information theory, which is often used to evaluate the pros and cons of probabilistic models and their generalization ability. The perplexity of a probabilistic model is defined as the energy based on the entropy of the probabilistic model. The smaller the perplexity, the stronger the generalization ability of the model, and the generated topics will have higher performance. In order to find the optimal number of topics, the model is trained with the number of topics ranging from 10 to 140. The span of the value is 10. Optimal Perplexity Number of Topics in LDA Model, LDA is selected as the baseline, proving that the W-LDA model has better generalization ability.

[0087] A method for constructing an LDA model for identifying electric power sensitive information topics of the present invention obtains basic sensitive words in the electric power system; obtains a sensitive word list based on similarity analysis between preset target category sensitive words and each sensitive word in the sensitive words; divides the sensitive words in the sensitive word list into topics according to preset division rules to obtain sensitive word sets of multiple different topics; inputs the sensitive word set of each topic as a subset into the LDA model for learning and training, and generates a topic identification LDA model with output of electric power sensitive information topic identification. The present invention can more accurately identify and extract sensitive information topics through deep mining and analysis of text data in the electric power industry, combined with the topic identification algorithm of the LDA model, to provide strong support for the safe development and information management of the electric power industry, ensure the information security of the electric power industry, and promote the intelligence of the electric power industry.

[0088] Preferably, in a preferred implementation of the present application, the calculating of the similarity between the preset target category sensitive words and the word embedding of each sensitive word of the sensitive words, screening out a preset number of seed words according to the similarity score result, and sorting out the sensitive word list includes:

[0089] Step S21: training the target category sensitive words and each sensitive word of the sensitive words respectively by selecting a Word2vec word embedding model to obtain a word embedding vector of the target category sensitive words and an embedding vector of each sensitive word of the sensitive words;

[0090] Use Google's word2vec model for word embedding training. The main parameters of word2vec include Size (word embedding dimension) specifies the dimension size of the word embedding vector. Each word will be represented as a vector with Size dimension. A larger Size value can capture more semantic information, but it will also increase computational complexity and memory consumption. Window (context window size) specifies the maximum distance between the current word and the context word. That is, during training, words within the window size range will be considered as context words. A larger Window value can capture a wider range of context information, but may introduce more noise. A smaller Window value focuses more on local context information. , Min_count (minimum threshold for word frequency) Negative (number of negative samples) specifies the number of negative samples corresponding to each positive sample in negative sampling. Negative sampling is a method to speed up the training process by reducing the number of updated word pairs. A moderate Negative value can improve model performance and training efficiency. Iter (number of iterations) specifies the number of iterations of model training, that is, how many times the model trains the entire corpus. More iterations can improve the accuracy of the model, but it will also increase the training time.

[0091] In this embodiment, Window (context window size) is set to 3, Min_count (minimum threshold of word frequency) is set to 4, rounds is set to 100, and Negative (number of negative samples) is set to 6. The Skip-gram model combined with negative sampling is used to train word embedding. The text is trained by Word2vec, and all words are represented by R-dimensional vectors. Then, the content of each text is mapped to the R-dimensional vector. Therefore, the semantic similarity of the text can be represented by the similarity of the space vector. After training with word2vec, word embedding is obtained.

[0092] Step S22, calculating the cosine similarity between the word embedding vector of the target category sensitive word and the embedding vector of each sensitive word of the sensitive words;

[0093] By calculating the cosine similarity between the word embedding vector of the target category sensitive word and the embedding vector of each sensitive word of the sensitive word, the smaller the similarity, the greater the distance, and the greater the similarity, the smaller the distance.

[0094] Step S23: According to the cosine similarity score, a preset number of new seed words corresponding to the target category sensitive words are screened out, and a sensitive word list is obtained.

[0095] According to the cosine similarity score, a preset number of seed words with the closest semantic distance to each sensitive word are filtered out. After the similarity calculation, some sensitive words are effectively expanded. After obtaining the word embedding, the 10 closest words to each sensitive word are calculated based on the basic sensitive words, but some non-sensitive words are also found. Therefore, it is necessary to further manually filter the calculation results and sort out the final sensitive word list to ensure the reliability of the expanded sensitive words.

[0096] Compared with neural language models, the main advantage of word2vec is that it discards the hidden layer that requires tedious calculations and supports adding additional information to the language model. Based on the advantages of word embeddings in semantic computing, the method of calculating the distance between word embeddings is used to expand the sensitive vocabulary.

[0097] In other embodiments, the word embedding model may use the GloVe word embedding model, the FastText word embedding model, and the ELMO (Embeddings from Language Model) word embedding model in addition to the Word2Vec model. Each model has its own unique characteristics and advantages. In practical applications, a suitable word embedding model can be selected according to the requirements of a specific task.

[0098] Preferably, in a preferred implementation of the present application, the method of screening out a preset number of new seed words corresponding to the target category sensitive words according to the cosine similarity score, and arranging the sensitive word list includes:

[0099] Step S201, sorting the cosine similarity score results in descending order;

[0100] By sorting in descending order, the words with higher cosine similarity scores are placed first. This ensures that in subsequent steps, the words that are most similar to a given criterion or query are considered first.

[0101] Improved filtering efficiency: The sorted list makes the filtering process more efficient because you only need to focus on the high-similarity words at the top of the list, without indiscriminately checking the entire vocabulary set.

[0102] Step S202: According to a preset cosine similarity threshold, a preset number of seed words that are greater than the cosine similarity threshold and are ranked at the top are selected;

[0103] The preset cosine similarity threshold is used to distinguish the degree of similarity between a word and a given standard or query. Only those words with a similarity higher than the threshold will be further considered, which helps to exclude irrelevant words and reduce noise. By presetting the number limit, it can ensure that the number of seed words finally screened out is within a controllable range, avoiding too many or too few words from affecting subsequent analysis. At the same time, this also makes the screening process more accurate and efficient.

[0104] Step S203: sorting the seed words to obtain a sensitive word list.

[0105] The selected seed words are summarized and sorted to form the final sensitive word list. The sorting process includes removing duplicate words, adjusting the word order, and other operations to ensure the accuracy and readability of the list. The sensitive word list is the basis for subsequent analysis and application. Through this step, it is possible to clearly show which words are considered sensitive, which is convenient for subsequent public opinion monitoring, information security management, and other operations.

[0106] The embodiment of the present application further provides a system for constructing an LDA model for identifying power-sensitive information topics, which is applied to the aforementioned method for constructing an LDA model for identifying power-sensitive information topics, including:

[0107] A vocabulary acquisition unit, used to acquire basic sensitive words in the power system;

[0108] A vocabulary list integration unit, used for obtaining a sensitive vocabulary list according to similarity analysis between a preset target category sensitive word and each sensitive word in the sensitive vocabulary;

[0109] A topic classification unit, used to divide the sensitive vocabulary into topics according to a preset division rule to obtain a plurality of sensitive word sets of different topics;

[0110] The topic identification model generation unit is used to input the sensitive word set of each topic as a subset into the LDA model for learning and training, and generate a topic identification LDA model with power sensitive information topic identification output.

[0111] In this embodiment, the explanation of the functions corresponding to each unit is the same as the explanation of the aforementioned method for constructing an LDA model for power-sensitive information topic identification, and will not be repeated here.

[0112] The present invention provides a system for constructing an LDA model for identifying electric power sensitive information topics. The system acquires basic sensitive words in the electric power system through a vocabulary acquisition unit. The vocabulary list integration unit obtains a sensitive word list based on similarity analysis between a preset target category sensitive word and each sensitive word in the sensitive word list. The topic classification unit divides the sensitive words in the sensitive word list into topics according to preset division rules to obtain sensitive word sets of multiple different topics. The topic identification model generation unit inputs the sensitive word set of each topic as a subset into the LDA model for learning and training to generate a topic identification LDA model with output of electric power sensitive information topic identification. The present invention can more accurately identify and extract sensitive information topics through deep mining and analysis of text data in the electric power industry, combined with the topic identification algorithm of the LDA model, to provide strong support for the safe development and information management of the electric power industry, ensure the information security of the electric power industry, and promote the intelligence of the electric power industry.

[0113] An embodiment of the present application also provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the aforementioned method for constructing an LDA model for power-sensitive information topic identification.

[0114] The technical effect of this embodiment is the same as the technical effect of a method for constructing an LDA model for power-sensitive information topic identification in an embodiment, and will not be repeated here.

[0115] The present invention can be used in many general or special computer system environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like.

[0116] An embodiment of the present application further provides a processor, which is used to run a program, wherein the program executes the aforementioned method for constructing an LDA model for identifying power-sensitive information topics when running.

[0117] The technical effect of this embodiment is the same as the technical effect of the method for constructing an LDA model for power-sensitive information topic identification in Embodiment 1, and will not be repeated here.

[0118] The processor in this embodiment may be a central processing unit (CPU), a controller, a microcontroller, or other data processing chips.

[0119] Compared with the prior art, the present invention has the following beneficial effects:

[0120] A method for constructing an LDA model for identifying electric power sensitive information topics of the present invention obtains basic sensitive words in the electric power system; obtains a sensitive word list based on similarity analysis between preset target category sensitive words and each sensitive word in the sensitive words; divides the sensitive words in the sensitive word list into topics according to preset division rules to obtain sensitive word sets of multiple different topics; inputs the sensitive word set of each topic as a subset into the LDA model for learning and training, and generates a topic identification LDA model with output of electric power sensitive information topic identification. The present invention can more accurately identify and extract sensitive information topics through deep mining and analysis of text data in the electric power industry, combined with the topic identification algorithm of the LDA model, to provide strong support for the safe development and information management of the electric power industry, ensure the information security of the electric power industry, and promote the intelligence of the electric power industry.

[0121] Those of ordinary skill in the art will appreciate that the units of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0122] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc.

[0123] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0124] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nlyMemory), random access memory (RAM, RandomAccessMemory), mobile hard disk, magnetic disk or optical disk, etc., which can store program code.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A method for constructing an LDA model for identifying power sensitive information topics, characterized in that: include: Obtain basic sensitive words in the power system; Obtaining a sensitive word list based on similarity analysis between a preset target category sensitive word and each sensitive word in the sensitive words; The sensitive words in the sensitive word list are divided into topics according to a preset division rule to obtain a plurality of sensitive word sets of different topics; The sensitive word set of each topic is input as a subset into the LDA model for learning and training, generating a topic identification LDA model with output of power sensitive information topic identification.

2. According to claim 1, a method for constructing an LDA model for identifying power-sensitive information topics is characterized in that: The basic sensitive words in the power system include: Obtain text information of power-related data in the power system and perform data cleaning; Performing text segmentation processing on the cleaned text information; The segmented text is processed according to the created stop word list to obtain basic sensitive words.

3. The method for constructing an LDA model for identifying power-sensitive information topics according to claim 1, characterized in that: The sensitive word list obtained by similarity analysis between the preset target category sensitive word and each sensitive word in the sensitive words includes: The similarity between the preset target category sensitive words and the word embedding of each sensitive word of the sensitive words is calculated, and a preset number of seed words are screened out according to the similarity score results, and a sensitive word list is obtained by sorting.

4. The method for constructing an LDA model for identifying power-sensitive information topics according to claim 3, characterized in that: The calculating of the similarity between the preset target category sensitive words and the word embedding of each sensitive word of the sensitive words, screening out a preset number of seed words according to the similarity score result, and sorting out the sensitive word list includes: The target category sensitive words and each sensitive word of the sensitive words are trained respectively by selecting a Word2vec word embedding model to obtain a word embedding vector of the target category sensitive words and an embedding vector of each sensitive word of the sensitive words; and the cosine similarity between the word embedding vector of the target category sensitive words and the embedding vector of each sensitive word of the sensitive words is calculated; According to the cosine similarity score, a preset number of new seed words corresponding to the target category sensitive words are screened out, and a sensitive word list is obtained.

5. The method for constructing an LDA model for identifying power-sensitive information topics according to claim 4, characterized in that: The step of selecting a preset number of new seed words corresponding to the target category sensitive words according to the cosine similarity score and sorting out a sensitive word list includes: Sorting the cosine similarity score results in descending order; According to a preset cosine similarity threshold, a preset number of seed words that are greater than the cosine similarity threshold and are ranked at the top are selected; The seed words are sorted to obtain a sensitive word list.

6. The method for constructing an LDA model for identifying power-sensitive information topics according to claim 1, characterized in that: The sensitive vocabulary list is divided into topics according to a preset division rule to obtain multiple groups of sensitive word sets of different topic categories: when a word with obvious identification is identified in the sensitive vocabulary list, it is classified according to the obvious identification to obtain a sensitive word set of the corresponding identification topic category; When it is identified that there are no obvious identifying words in the sensitive vocabulary, classification is performed according to a coarse-grained classification method to obtain a plurality of sensitive word sets of different subject categories.

7. A system for constructing an LDA model for identifying power-sensitive information topics, characterized in that: A method for constructing an LDA model for identifying power-sensitive information topics as described in any one of claims 1 to 6, comprising: A vocabulary acquisition unit, used to acquire basic sensitive words in the power system; A vocabulary list integration unit, configured to obtain a sensitive vocabulary list based on similarity analysis between a preset target category sensitive word and each sensitive word in the sensitive vocabulary; A topic classification unit, used to divide the sensitive vocabulary into topics according to a preset division rule to obtain a plurality of sensitive word sets of different topics; The topic identification model generation unit is used to input the sensitive word set of each topic as a subset into the LDA model for learning and training, and generate a topic identification LDA model with power sensitive information topic identification output.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute a method for constructing an LDA model for power-sensitive information topic identification as described in any one of claims 1 to 6.

9. A processor, characterized in that: The processor is used to run a program, wherein the program, when running, executes a method for constructing an LDA model for power-sensitive information topic identification as described in any one of claims 1 to 6.

Images