Entity information extraction method and device and electronic equipment

By extracting seed naming entities in the network alarm information corpus and identifying related terms, the naming entity representative words are expanded, and the time-consuming and labor-intensive problem of labeled data that relies on supervised learning in the existing technology is solved, and the accuracy and efficiency of entity extraction are improved.

CN119047469APending Publication Date: 2024-11-29CHINA MOBILE (SUZHOU) SOFTWARE TECH CO LTD +1
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410984567.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

The existing entity extraction technology relies on supervised learning, and the process of labeling data is time-consuming and labor-intensive. The lack of labeling data will lead to a degradation in the performance of the deep learning model and reduce the accuracy of entity extraction.

Method used

By obtaining the representative words of named entity for network alarm information and historical network alarm information to be marked, extracting seed named entities in the network alarm information corpus, identifying the related targets and quoting, expanding the representative words of named entity, and using these terms Perform entity information annotation processing, train entity information extraction models to reduce manual annotation costs and improve model performance.

Benefits of technology

It reduces the cost of manual labeling, improves the accuracy and efficiency of entity extraction tasks, enables the model to learn richer features from more data, and enhances generalization ability and robustness.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119047469A_ABST
    Figure CN119047469A_ABST
Patent Text Reader

Abstract

The invention relates to an entity information extraction method and device and electronic equipment, and relates to the technical field of cloud computing, and the method comprises the steps: obtaining named entity representative words and corresponding entity information of to-be-labeled network alarm information and historical network alarm information, and the entity information comprises coarse-grained entity information and fine-grained entity information; extracting seed named entities of the network alarm information corpus according to the coarse-grained entity information and the fine-grained entity information, and identifying target related lexical items to expand named entity representative words; performing entity labeling on the target related lexical item by utilizing the expanded named entity representative word to obtain entity information; and performing model training on the entity information extraction model by using the historical network alarm information and the entity information, and performing entity extraction on the network alarm information to be labeled by using the entity information extraction model to obtain the entity information corresponding to the network alarm information to be labeled. And the accuracy of tasks such as entity extraction is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of cloud computing technology, and in particular to an entity information extraction method, device and electronic device. Background Art

[0002] With the rapid development of the information age, intelligent operation and maintenance (AIOps) as an emerging technology attempts to combine artificial intelligence and operation and maintenance fields to process large and complex alarm information. Through knowledge extraction technologies such as entity extraction and relationship extraction, it helps operation and maintenance personnel to quickly identify key information, improve fault location speed and network management level.

[0003] At present, entity extraction is an important task in knowledge extraction, which aims to identify entities with specific meanings from text, such as names of people, places, and organizations. Entity extraction technology has evolved from rule-based methods to deep learning-based methods; compared with other methods, deep learning methods generally use an end-to-end approach, which can automatically discover hidden features and predict entity labels in texts, thus avoiding cumbersome rule formulation, improving extraction efficiency, and reducing labor costs.

[0004] However, most existing entity extraction technologies rely on supervised learning, which is to manually annotate a large amount of text data in advance to form a data set, and then perform entity extraction or knowledge extraction tasks. However, the process of annotating data takes a lot of time and money to complete, and if there is a lack of annotated data, it will make it difficult for deep learning models that rely on a large amount of annotated data to achieve excellent performance, thereby reducing the accuracy of tasks such as entity extraction. Summary of the invention

[0005] In view of this, the present application provides an entity information extraction method, device and electronic device, the main purpose of which is to solve the technical problem that most current entity extraction technologies rely on supervised learning, and the process of labeling data requires a lot of time and cost to complete. If there is a lack of labeled data, deep learning models that rely on a large amount of labeled data will find it difficult to achieve excellent performance, thereby reducing the accuracy of tasks such as entity extraction.

[0006] According to a first aspect of the present disclosure, a method for extracting entity information is provided, the method comprising:

[0007] Acquire multiple network alarm information to be annotated, acquire named entity representative words of multiple historical network alarm information, and acquire entity information annotated corresponding to the named entity representative words, wherein the entity information includes coarse-grained entity information and fine-grained entity information;

[0008] Extracting seed named entities in a network alarm information corpus according to the coarse-grained entity information and the fine-grained entity information, and identifying target-related terms in the network alarm information corpus that are related to the seed named entities, so as to expand the named entity representative words with the target-related terms;

[0009] Using the expanded named entity representative words to perform entity information tagging processing on the target related words to obtain the entity information corresponding to the expanded named entity representative words;

[0010] The entity information extraction model is trained using the historical network alarm information and the corresponding annotated entity information, and the trained entity information extraction model is used to extract entity information from the multiple network alarm information to be annotated to obtain entity information corresponding to the multiple network alarm information to be annotated.

[0011] According to a second aspect of the present disclosure, there is provided an entity information extraction device, the device comprising:

[0012] An acquisition module, used to acquire multiple network alarm information to be annotated, acquire named entity representative words of multiple historical network alarm information, and acquire entity information annotated corresponding to the named entity representative words, wherein the entity information includes coarse-grained entity information and fine-grained entity information;

[0013] An expansion module, configured to extract a seed named entity in a network alarm information corpus according to the coarse-grained entity information and the fine-grained entity information, and identify a target-related term related to the seed named entity in the network alarm information corpus, so as to expand the named entity representative word with the target-related term;

[0014] A tagging module, used to perform entity information tagging processing on the target related terms by using the expanded expanded named entity representative terms to obtain the entity information corresponding to the annotation of the expanded named entity representative terms;

[0015] The extraction module is used to use the historical network alarm information and the corresponding annotated entity information to train the entity information extraction model, and use the trained entity information extraction model to extract entity information from the multiple network alarm information to be annotated, so as to obtain the entity information corresponding to the multiple network alarm information to be annotated.

[0016] According to the third aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method of the aforementioned first aspect.

[0017] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the method of the aforementioned first aspect.

[0018] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein the computer program implements the method of the aforementioned first aspect when executed by a processor.

[0019] The entity information extraction method, device and electronic device provided by the present invention, compared with the prior art, obtain multiple network alarm information to be annotated, obtain named entity representative words of multiple historical network alarm information, and obtain entity information annotated corresponding to the named entity representative words, wherein the entity information includes coarse-grained entity information and fine-grained entity information; extract seed named entities in the network alarm information corpus according to the coarse-grained entity information and the fine-grained entity information, and identify target-related terms related to the seed named entities in the network alarm information corpus, so as to expand the named entity representative words with the target-related terms; perform entity information annotating processing on the target-related terms with the expanded expanded named entity representative words, so as to obtain entity information annotated corresponding to the expanded named entity representative words; perform model training on the entity information extraction model with the historical network alarm information and the corresponding annotated entity information, and use the trained entity information extraction model to extract entity information from multiple network alarm information to be annotated, so as to obtain entity information corresponding to the multiple network alarm information to be annotated. For the disclosed scheme, by extracting seed named entities from the network alarm information corpus, and using the seed named entities, target-related terms related to the seed named entities are identified from the network alarm information corpus, thereby reducing the cost of manual labeling; because the named entity representative words are expanded by using target-related terms, the entity knowledge extraction model can learn richer features from more data, thereby improving the accuracy of tasks such as entity extraction.

[0020] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0023] Figure 1 A flowchart of the entity information extraction method provided by the embodiment of the present disclosure;

[0024] Figure 2 A schematic diagram of the overall process of an entity information extraction method provided by an embodiment of the present disclosure;

[0025] Figure 3 A schematic diagram of a flow chart of an entity definition and annotation module provided in an embodiment of the present disclosure;

[0026] Figure 4 A schematic diagram of a semi-supervised data annotation process provided by an embodiment of the present disclosure;

[0027] Figure 5 A flowchart of another entity information extraction method provided by an embodiment of the present disclosure;

[0028] Figure 6 A flowchart of expanding a named entity representative word provided by an embodiment of the present disclosure;

[0029] Figure 7 A multi-task learning architecture diagram provided by an embodiment of the present disclosure;

[0030] Figure 8 A structural diagram of an entity knowledge extraction model provided by an embodiment of the present disclosure;

[0031] Fig. 9 A schematic diagram of the structure of an entity information extraction device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0032] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description. It should be noted that, in the absence of conflict, the embodiments of the present disclosure and the features in the embodiments may be combined with each other.

[0033] The entity information extraction method, device and electronic device of the embodiments of the present disclosure are described below with reference to the accompanying drawings.

[0034] The present disclosure provides an entity information extraction method, device and electronic device, which reduces the cost of manual annotation by identifying target-related terms related to seed named entities from multiple network alarm information corpora; and expands the named entity representative words by using target-related terms, so that the entity knowledge extraction model can learn richer features from more data, thereby improving the accuracy of tasks such as entity extraction.

[0035] like Figure 1 As shown, an embodiment of the present disclosure provides an entity information extraction method, the method comprising:

[0036] Step 101: obtain multiple network alarm information to be annotated, obtain named entity representative words of multiple historical network alarm information, and obtain entity categories annotated corresponding to the named entity representative words, wherein the entity categories include coarse-grained entity categories and fine-grained entity categories.

[0037] The network alarm information to be marked may be alarm information of entity categories that have not yet been marked;

[0038] Historical network alarm information can be network alarm data collected over a period of time in the past;

[0039] Network alarm information can be event information generated by failures that trigger predetermined rules during the operation of various network devices. Through network alarm information, you can get abnormal notifications. Only after specific analysis can you get the cause of the abnormality and further handle the abnormality. Therefore, simple network alarm information cannot get the source of the alarm and the failure. The complicated and obscure professional terms in the network field also increase the difficulty of analysis. It must be analyzed by professional operation and maintenance personnel, and compared with other historical data for comprehensive analysis to locate the root cause of the alarm.

[0040] Generally speaking, network alarm information is automatically generated by computers during system operation. The more complex the network structure, the greater the possibility of problems. The more complex the problems that arise, the more complex they are, and they may even be caused by more than one reason.

[0041] The named entity representative words may be words representing specific named entity categories that can be extracted from historical network alarm information by annotation methods such as BIO annotation and / or BIOES annotation in the named entity recognition task;

[0042] Among them, B (Begin) in the BIO notation method can represent the first character in an entity, I (Inside) can represent other characters in the entity except the first character, and O (Outside) can represent non-entity characters.

[0043] BIOES notation may be an extension of BIO notation, where E may stand for End and S may stand for Single.

[0044] For the embodiments of the present disclosure, the marking method is taken as the BIO marking method as an example to illustrate the technical solution in the present disclosure, but it does not constitute a specific limitation on the technical solution in the present disclosure.

[0045] In actual operation, after defining the entity categories and annotation specifications, 300 sample information can be randomly selected from the historical network alarm information actually collected, and all types of alarm information in the sample information are covered to ensure the comprehensiveness and representativeness of the annotation. Then, the sample information can be annotated with BIO named entity recognition according to the pre-defined annotation specifications to construct a set of named entity representative words; after the annotation is completed, the annotation results can be verified. If the annotation results are inconsistent, the inconsistent results can be processed to ensure the quality of the annotation results.

[0046] Coarse-grained entity categories can be used to classify entities at a higher level, and can be divided into five types: alarm name information, alarm category, alarm network information, alarm details, and time;

[0047] Fine-grained entity categories can be further subdivided into coarse-grained entity categories, which can be divided into 21 types, including company / organization name, alarm ID, resource pool, alarm name, alarm type, alarm level, monitoring object, attack type, alarm IP, domain name, port number, network protocol, monitoring system / tool, process, application, network service, alarm source, impact scope, processing status, start time, and last occurrence time.

[0048] The corresponding relationship and examples are shown in Table 1 below:

[0049]

[0050] Table 1 Entity definitions and examples

[0051] This application specifically studies the text of network operation and maintenance alarm information, and defines two groups of named entities applicable to network alarm information according to different division granularities, adapting to the particularity and complexity of the network alarm field. Compared with the existing entity extraction methods in general fields or other specific fields, this application can better solve the problem that the entities contained in the network alarm field are very different from the entities contained in other fields, and provide more accurate and comprehensive recognition and extraction capabilities for various types and levels of entities in the network alarm field.

[0052] Step 102: extract seed named entities in the network alarm information corpus according to the coarse-grained entity categories and the fine-grained entity categories, and identify target-related terms related to the seed named entities in the network alarm information corpus to expand the named entity representative words using the target-related terms.

[0053] Among them, the network alarm information corpus can be a database for collecting network alarm information; the seed named entity can be an entity instance that is highly related to a specific named entity category, which can be used to expand the set of named entity representative words, thereby improving the performance of the entity extraction model.

[0054] In a specific application scenario, for example, Figure 2 As shown in the figure, coarse-grained entity categories and fine-grained entity categories can be defined according to the characteristics of historical network alarm information. According to the 5 coarse-grained entity categories and 21 fine-grained entity categories, 150 representative words closely related to each entity category can be selected from the historical network alarm information, and the selected representative words can be combined to form seed named entities for expanding the named entity representative words, while ensuring that the coarse-grained and fine-grained selected representative words are not repeated. The selection of seed named entities needs to take into account the range of network alarm information types it covers. Therefore, it is specially selected from the processed historical alarm information to reduce manual workload and include most alarm types as much as possible to ensure its coverage.

[0055] For the embodiments of the present disclosure, Figure 2 and Figure 3 As shown in the figure, as a possible implementation method, a network alarm information corpus can be collected, and candidate words can be extracted from the collected network alarm information corpus, wherein the candidate words can be words that are semantically similar to the seed named entity or frequently co-occur in the text; the TF-IDF algorithm can be used to screen the candidate words to obtain target-related terms related to the seed named entity; the K-means algorithm can be used to cluster data points based on the seed named entity and the target-related terms to expand the representative words of the named entity, thereby reducing the workload of manual annotation and improving the performance of entity extraction.

[0056] This application proposes a semi-supervised learning method, which expands seed named entities and then automatically annotates them, thereby reducing manual annotation costs and improving data utilization and model performance. Compared with existing supervised learning-based methods, this application can effectively use unlabeled data or weakly labeled data to enhance the generalization ability and robustness of the model, and reduce the model's dependence on the quality and quantity of labeled data.

[0057] Step 103: Use the expanded named entity representative words to perform entity information tagging processing on the target related terms to obtain the entity information corresponding to the tag of the expanded named entity representative words.

[0058] For the embodiments of the present disclosure, Figure 4 As shown, the expanded named entity representative words can be used to perform entity information annotation processing on the target related terms. As a possible implementation method, the network alarm information to be annotated in the network alarm information corpus can be preprocessed, which may include operations such as word segmentation and removal of stop words to obtain a clean text representation.

[0059] A word vector model (such as Word2Vec, GloVe or BERT, etc.) can be used to convert the preprocessed text into a first word vector, and the expanded named entity representative word can be converted into a second word vector. The similarity between the first word vector and the second word vector can be calculated by calculating cosine similarity or other methods, and based on the similarity calculation result, each target-related term of the network alarm information to be annotated is annotated as the entity information corresponding to the most similar named entity representative word.

[0060] If the target term has high similarity with multiple named entity representative words, the target term can be annotated as multiple entity information.

[0061] Step 104: Use historical network alarm information and corresponding annotated entity information to train an entity information extraction model, and use the trained entity information extraction model to extract entity information from multiple network alarm information to be annotated, to obtain entity information corresponding to multiple network alarm information to be annotated.

[0062] For the disclosed embodiments, the annotated historical network alarm information and the corresponding annotated entity information can be used to train the entity information extraction model through methods such as conditional random field (CRF) and named entity recognition (NER), and the trained entity information extraction model can be used to extract entity information from the network alarm information to be annotated, so as to obtain entity information corresponding to multiple network alarm information to be annotated, wherein the entity information may include named entities, entity categories, etc.

[0063] Through the above steps, the named entities and their related attributes in the network alarm information to be annotated can be automatically extracted, thereby improving the efficiency and accuracy of tasks such as entity extraction.

[0064] In summary, compared with the prior art, the entity information extraction method disclosed in the present invention obtains multiple network alarm information to be labeled, obtains named entity representative words of multiple historical network alarm information, and obtains entity information labeled corresponding to the named entity representative words, and the entity information includes coarse-grained entity information and fine-grained entity information; extracts seed named entities in the network alarm information corpus according to the coarse-grained entity information and the fine-grained entity information, and identifies target-related terms related to the seed named entities in the network alarm information corpus, so as to expand the named entity representative words with the target-related terms; performs entity information labeling processing on the target-related terms with the expanded expanded named entity representative words, so as to obtain the entity information labeled corresponding to the expanded named entity representative words; uses the historical network alarm information and the corresponding labeled entity information to train the entity information extraction model, and uses the trained entity information extraction model to extract entity information from multiple network alarm information to be labeled, so as to obtain the entity information corresponding to the multiple network alarm information to be labeled. For the disclosed scheme, by extracting seed named entities from the network alarm information corpus, and using the seed named entities, target-related terms related to the seed named entities are identified from the network alarm information corpus, thereby reducing the cost of manual labeling; because the named entity representative words are expanded by using target-related terms, the entity knowledge extraction model can learn richer features from more data, thereby improving the accuracy of tasks such as entity extraction.

[0065] Further, as a refinement and extension of the above embodiment, in order to fully illustrate the specific implementation process of the method of this embodiment, this embodiment provides the following Figure 5 The specific method shown comprises the following steps:

[0066] Step 201: Identify target-related terms related to a seed named entity in a network alarm information corpus, so as to expand the named entity representative words by using the target-related terms.

[0067] For the embodiments of the present disclosure, the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm can be used to expand the representative words of named entities, where TF-IDF is an algorithm for extracting important words from text. The importance of a word can be measured by calculating its TF-IDF value in the network alarm information corpus, capturing keywords related to specific named entities in the network alarm information corpus, and using this to expand the named entity vocabulary set, thereby providing more vocabulary features for the entity extraction model. The flowchart is shown in FIG. Figure 6 shown.

[0068] Specifically, a network alarm information corpus can be collected and data preprocessing is performed on the network alarm information corpus. The preprocessing may include using the Jieba word segmentation tool to split the to-be-annotated network alarm information in the network alarm information corpus into individual words or phrases (i.e., terms), and operations such as removing meaningless stop words in the terms to obtain target terms.

[0069] Among them, stop words can be words that frequently appear in the text but contribute little to the text content, such as "of", "and", "is", etc.

[0070] Calculate the TF-IDF value (i.e., the metric value) of the target terms of each to-be-annotated network alarm information in the network alarm information corpus. Among them, the TF-IDF value consists of two parts: term frequency (TF) and inverse document frequency (IDF); the term frequency can be the frequency of a certain target term appearing in the network alarm information corpus, and the inverse document frequency can be the rarity of a certain target term in the network alarm information corpus.

[0071] The higher the calculated TF-IDF value of the target term, the more important the target term is in the to-be-annotated network alarm information, and at the same time, the lower the frequency of its appearance in the entire network alarm information corpus.

[0072] A preset metric threshold (e.g., 0.5) can be set. If the TF-IDF value of a target term is greater than or equal to the preset metric threshold, then the target term is determined to be a target-related term related to the seed named entity. Then, all eligible target-related terms are added to the named entity representative word set to expand the named entity representative words.

[0073] Correspondingly, identifying the target-related terms related to the seed named entity in the network alarm information corpus to expand the named entity representative words using the target-related terms may specifically include:

[0074] Perform word segmentation on the network alarm information in the network alarm information corpus, and剔除 the stop words in the terms after the word segmentation to obtain target terms;

[0075] Calculate the metric value of each target term based on the term frequency value and the inverse document frequency value of the target term in the network alarm information;

[0076] Determine the target terms with metric values greater than the preset metric threshold as the target-related terms related to the seed named entity to expand the named entity representative words using the target-related terms.

[0077] Step 202: Expand the named entity representative words using the target-related terms.

[0078] Specifically, as Figure 6 shown, the named entity representative words and the seed named entities can be used as data points, and the target-related terms and the seed named entities can be used as the initial clustering centers. When allocating, each data point can be assigned to the cluster represented by the nearest clustering center according to the distance metric (such as Euclidean distance or cosine similarity, etc.).

[0079] After all data points are assigned to the clusters, the average value of the data points within each cluster can be calculated, and these average values can be used as the new clustering centers. Repeat the allocation and update of the clustering centers until the clustering centers no longer change significantly or reach a predetermined number of iterations.

[0080] During the clustering process, the seed words (i.e., the seed named entities), as the known categories, constrain all data points, so that the data points similar to the seed words are grouped into the same cluster, making the clustering result not only based on the similarity between data points, but also guided by the seed words, ensuring the semantic relevance of the clustering result. Obtain the seed-based clustering result and associate the named entity representative words with the seed named entities, so that the named entity representative words are assigned to the clusters represented by the seed named entities that are most similar to them.

[0081] Correspondingly, using the target-related terms and the seed named entities as the initial clustering centers, calculate the distance metric values between each data point and the initial clustering centers, and assign each data point to the cluster where the initial clustering center corresponding to the minimum distance metric value is located;

[0082] According to the average value of the data points within each cluster, iteratively update the clustering centers until the moving position of the clustering centers relative to the preset clustering centers is less than the preset moving position, to obtain the iteratively expanded named entity representative words, where the expanded named entity representative words include the seed named entities, the target-related terms, and the named entity representative words.

[0083] Step 203: Calculate the first word vector of the target-related terms and the second word vector of the expanded named entity representative words, and calculate the similarity degree value between the first word vector and the second word vector; label the target-related terms with similarity values greater than the preset similarity value as the corresponding entity information, so as to obtain the entity information corresponding to the labeled expanded named entity representative words.

[0084] For the embodiments of the present disclosure, as a possible implementation method, word segmentation processing and stop word removal processing can be performed on the to-be-labeled network alarm information in the network alarm information corpus. Among them, the word segmentation processing can be to split the to-be-labeled network alarm information into individual words or phrases (i.e., the to-be-labeled word items), and the stop word removal processing can be to remove the meaningless words in the to-be-labeled word items, such as "of", "and", "is", etc., to obtain the target-related terms.

[0085] The word vector model (such as Word2Vec) can be trained to input each target-related term into the trained word vector model to obtain the first word vector of each target-related term, and to input each expanded named entity representative word into the trained word vector model to obtain the second word vector of each expanded named entity. The Word2Vec algorithm can convert text content into vector operations in a K-dimensional space, using spatial distance to represent semantic similarity.

[0086] The expanded named entity representative words include a set of seed named entities, target-related terms, and named entity representative words.

[0087] The similarity value between the first word vector and the second word vector can be calculated by calculating cosine similarity or other methods, and based on the similarity calculation result, each target-related term of the network alarm information to be labeled is labeled as the entity category corresponding to the most similar named entity representative word.

[0088] As a possible implementation method, this application uses the CBOW model in the Word2Vec algorithm, and counts the number of times each word appears in the entire data set after word segmentation, and sets a minimum word frequency threshold (for example, 2) to filter out some noise and occasional words, while retaining those entities that appear frequently in the data to ensure that important entities are not missed, thereby improving the accuracy of entity extraction. Among them, the CBOW model can learn word vectors by predicting the central word, and can use context information to improve the quality of word vectors.

[0089] In this way, the Word2Vec algorithm can help the model identify and classify entities in the text, thereby improving the performance of entity extraction. This method combines a small amount of manually annotated seed named entities with a large amount of unlabeled data to achieve a balance between cost and performance.

[0090] Step 204: Use the historical network alarm information as the input feature of the entity information extraction model, use the entity information as the training label of the entity information extraction model, iteratively train the entity information extraction model until the total loss function value of the entity information extraction model is less than the preset loss function threshold, and determine that the entity information extraction model training is completed.

[0091] The entity information includes first entity information and second entity information.

[0092] For the embodiment of the present disclosure, historical network alarm information is used as an input feature of the entity information extraction model, entity information is used as a training label of the entity information extraction model, and the entity information extraction model is iteratively trained until the total loss function value of the entity information extraction model is less than a preset loss function threshold, and it is determined that the entity information extraction model training is completed, which may specifically include:

[0093] Perform word segmentation processing on the historical network alarm information, and remove stop words in the word items after the word segmentation processing to obtain a text sequence of the historical network alarm information;

[0094] The pre-trained language model in the entity information extraction model is used to convert the historical network warning information text sequence into an embedded vector form, and the bidirectional long short-term memory network in the entity information extraction model is used to extract features from the embedded vector to obtain long-term dependency features;

[0095] Calculate the word-level vector features of the corresponding entity information in the historical network alarm information to obtain coarse-grained word-level vector features and fine-grained word-level vector features;

[0096] The coarse-grained word-level vector features and the long-term dependency features are concatenated to obtain the coarse-grained embedding layer vector features, and the fine-grained word-level vector features and the long-term dependency features are concatenated to obtain the fine-grained embedding layer vector features;

[0097] The attention mechanism in the entity information extraction model is used to select key fine-grained vector features in the fine-grained word-level vector features, and the attention mechanism in the entity information extraction model is used to select key coarse-grained vector features in the coarse-grained embedding layer vector features;

[0098] Using the conditional random field model in the entity information extraction model to predict the entity information of the historical network alarm information based on the key fine-grained vector features to obtain the first predicted entity information, and using the conditional random field model to predict the entity information of the historical network alarm information based on the key coarse-grained vector features to obtain the second predicted entity information;

[0099] Perform a weighted sum operation of a loss function on a first loss function value between the first predicted entity information and the first entity information, and a second loss function value between the second predicted entity information and the second entity information, to obtain a total loss function value;

[0100] If the total loss function value is less than the preset loss function threshold, the entity information extraction model training is determined to be completed.

[0101] Wherein, the bidirectional long short-term memory network may include a forward long short-term memory layer and a reverse long short-term memory layer;

[0102] Accordingly, the bidirectional long short-term memory network in the entity information extraction model is used to extract features from the embedded vector to obtain long-term dependency features, which may include:

[0103] A forward long short-term memory layer is used to obtain long-term positive dependency features from the start time point of the embedding vector to a preset time point, and a reverse long short-term memory layer is used to obtain long-term reverse dependency features from the end time point of the embedding vector to a preset time point;

[0104] The long-term positive dependency features and the long-term reverse dependency features are fused to obtain the long-term dependency features of the embedding vector.

[0105] For the disclosed embodiments, compared with single-task learning which uses limited information, multi-task learning optimizes the learning process based on the correlation of different tasks, and can capture more effective information. In order to make full use of the auxiliary information of related tasks to alleviate the data sparsity problem, the accuracy and performance of each task can be effectively improved.

[0106] Specifically, Figure 7 As shown in the figure, this application proposes a multi-task network alarm information entity knowledge extraction model, which simultaneously handles coarse-grained entity extraction and fine-grained entity extraction tasks, realizes the mutual complementation and mutual promotion of the two entity extraction predictions, and thus improves the information extraction performance of network alarm information text. The model structure is shown in Figure 8 As shown, it is divided into three parts: shared encoder, private encoder and decoder.

[0107] Among them, the shared encoder consists of two main parts: a shared embedding layer and a shared encoding layer.

[0108] First, after the historical network alarm information text is preprocessed by word segmentation and stop word removal, the historical network alarm information text sequence S = {S1, S2, …, S n} is input into the large-scale pre-trained language model RoBERTa to obtain the word-level embedding representation of the input sequence X = {X1,X2,…,X n}, where Xn represents the embedding vector of the nth word and n represents the number of words in the input sequence. The shared embedding layer will provide a common underlying representation for all tasks.

[0109] On the basis of the shared embedding layer, X is used as the input of the feature extraction layer. The context information in the historical network alarm information is very important for the entity extraction task. Therefore, this application uses a bidirectional long short-term memory network (LSTM) to obtain the context information in the alarm data. Let the output of LSTM be h = {h1,h2,…,h n}, in the bidirectional LSTM module of the model, its output is the output sequence of the forward LSTM And the output sequence of the reverse LSTM By position splicing, we get

[0110] Among them, the bidirectional long short-term memory network (Bi-LSTM) can be a special recurrent neural network (RNN). The bidirectional long short-term memory network can extract long-term dependency features in network alarm information from two directions (forward and backward) at the same time, so as to more comprehensively understand the context information of the network alarm information.

[0111] Among them, the private encoder is a key component of multi-task learning. It is connected to the shared encoder and provides task-specific encoding representations for coarse-grained entity extraction tasks and fine-grained entity extraction tasks respectively. By introducing the private encoder, the interdependence and correlation between different tasks can be fully utilized to improve the generalization ability and learning effect of the model.

[0112] For coarse-grained and fine-grained entity extraction, private embedding layers, self-attention mechanisms, and conditional random fields are used as their private encoder networks.

[0113] Specifically, firstly, based on the two data corpora with coarse-grained annotation and fine-grained annotation, the word-level vector E of the coarse-grained annotation is calculated using the Word2vec algorithm. c ={E c1 ,E c2 ,…,E cn} and fine-grained annotated word-level vector E f ={E f1 ,E f2 ,…,E fn}, and then concatenate the output results h of the shared encoder to obtain the coarse-grained embedding layer vector X of the private encoder. c =concat{h,E c} and the fine-grained embedding layer vector X f =concat{h,E f In this way, the model can learn common semantic information on the underlying representation of the shared embedding layer and perform features at a specific granularity level on independent private embedding layer sub-layers.

[0114] Next, the self-attention mechanism can be used to input X for the coarse-grained entity extraction task c And the fine-grained entity extraction task input X f Weighted processing is performed to obtain task-specific feature representations. The self-attention mechanism can assign different weights according to the importance of the input, thereby focusing on key information.

[0115] The calculation formula of the self-attention mechanism is as follows:

[0116]

[0117] C=AV

[0118] Among them, Q, K, V are obtained from the input matrix X through linear transformation:

[0119] Q=XW Q

[0120] K=XW K

[0121] V=XW V

[0122] W Q ,W K , W V is a trainable parameter matrix.

[0123] C is the feature representation of the self-attention mechanism for a specific task.

[0124] In the decoder, the present application can input the feature representations C obtained from the two tasks into a conditional random field (CRF) for label prediction. The CRF model can model the label sequence and consider the dependencies between labels, thereby improving the accuracy of the prediction.

[0125] Finally, a weighted combination loss function method is used to achieve the integration of multi-task learning. The losses l1 and l2 of the two tasks are calculated separately, and then combined into a loss L = αl1 + βl2. The parameter loss weights α, β are used to control the importance between the two tasks. The final output result is a label sequence Y = {y1, y2, ..., y n}.

[0126] This application proposes a method for extracting entity knowledge from network alarm information based on multi-task learning. Through feature learning of shared encoders, task-specific information extraction of private encoders, and sequence labeling of CRF decoders, the two tasks of coarse-grained entity extraction and fine-grained entity extraction are optimized simultaneously during the training process, so that the model can make full use of the relevant features between the two tasks and improve the overall performance by promoting and complementing each other. The purpose of accurately identifying named entities in network alarm information text is achieved. This helps operation and maintenance personnel quickly obtain key fault information of alarms under massive alarms, meet the needs of quickly locating and solving faults under large-scale alarms, and improve work efficiency.

[0127] In summary, compared with the prior art, the entity information extraction method disclosed in the present invention obtains multiple network alarm information to be labeled, obtains named entity representative words of multiple historical network alarm information, and obtains entity information labeled corresponding to the named entity representative words, and the entity information includes coarse-grained entity information and fine-grained entity information; extracts seed named entities in the network alarm information corpus according to the coarse-grained entity information and the fine-grained entity information, and identifies target-related terms related to the seed named entities in the network alarm information corpus, so as to expand the named entity representative words with the target-related terms; performs entity information labeling processing on the target-related terms with the expanded expanded named entity representative words, so as to obtain the entity information labeled corresponding to the expanded named entity representative words; uses the historical network alarm information and the corresponding labeled entity information to train the entity information extraction model, and uses the trained entity information extraction model to extract entity information from multiple network alarm information to be labeled, so as to obtain the entity information corresponding to the multiple network alarm information to be labeled. For the disclosed scheme, by extracting seed named entities from the network alarm information corpus, and using the seed named entities, target-related terms related to the seed named entities are identified from the network alarm information corpus, thereby reducing the cost of manual labeling; because the named entity representative words are expanded by using target-related terms, the entity knowledge extraction model can learn richer features from more data, thereby improving the accuracy of tasks such as entity extraction.

[0128] Based on the above Figure 1 and Figure 5 The present disclosure provides a device for extracting entity information, such as Fig. 9 As shown, the device includes: an acquisition module 31, an expansion module 32, a labeling module 32, and an extraction module 33;

[0129] An acquisition module 31 is used to acquire multiple network alarm information to be annotated, acquire named entity representative words of multiple historical network alarm information, and acquire entity information annotated corresponding to the named entity representative words, wherein the entity information includes coarse-grained entity information and fine-grained entity information;

[0130] An expansion module 32, configured to extract a seed named entity in the network alarm information corpus according to the coarse-grained entity information and the fine-grained entity information, and identify a target-related term related to the seed named entity in the network alarm information corpus, so as to expand the named entity representative word with the target-related term;

[0131] The tagging module 33 is used to perform entity information tagging processing on the target related terms by using the expanded expanded named entity representative words to obtain the entity information corresponding to the annotation of the expanded named entity representative words;

[0132] The extraction module 34 is used to use the historical network alarm information and the corresponding annotated entity information to perform model training on the entity information extraction model, and use the trained entity information extraction model to extract entity information from the multiple network alarm information to be annotated, so as to obtain the entity information corresponding to the multiple network alarm information to be annotated.

[0133] In a specific application scenario, the expansion module 32 may be used to perform word segmentation processing on the network alarm information in the network alarm information corpus, and remove stop words from the word items after the word segmentation processing to obtain target word items;

[0134] Calculate a metric value of each target term according to the term frequency value and the inverse document frequency value of the target term in the network alarm information;

[0135] A target term whose metric value is greater than a preset metric threshold is determined as a target-related term related to the seed named entity, so as to expand the named entity representative word with the target-related term.

[0136] In a specific application scenario, the expansion module 32 may be used to use the target-related terms and the seed named entities as initial cluster centers, calculate the distance metric between each data point and the initial cluster center, and assign each data point to the cluster where the initial cluster center corresponding to the minimum distance metric is located;

[0137] According to the average value of the data points in each cluster, the cluster center is iteratively updated until the movement position of the cluster center and the preset cluster center is less than the preset movement position, so as to obtain the iterative expanded named entity representative words, wherein the expanded named entity representative words include the seed named entity, the target-related terms and the named entity representative words.

[0138] In a specific application scenario, the tagging module 33 may be used to calculate a first word vector of the target-related term and a second word vector of the expanded named entity representative word, and to calculate a similarity value between the first word vector and the second word vector;

[0139] The target-related terms with similarity values ​​greater than a preset similarity value are marked as corresponding entity information to obtain the entity information marked corresponding to the expanded named entity representative term.

[0140] In a specific application scenario, the extraction module 34 can be used to use the historical network alarm information as the input feature of the entity information extraction model, use the entity information as the training label of the entity information extraction model, iteratively train the entity information extraction model until the total loss function value of the entity information extraction model is less than the preset loss function threshold, and determine that the entity information extraction model training is completed.

[0141] In a specific application scenario, the entity information includes first entity information and second entity information, and the extraction module 34 can be used to perform word segmentation processing on the historical network alarm information, and remove stop words in the word items after the word segmentation processing to obtain a text sequence of the historical network alarm information;

[0142] The historical network warning information text sequence is converted into an embedded vector form by using a pre-trained language model in the entity information extraction model, and the embedded vector is subjected to feature extraction by using a bidirectional long short-term memory network in the entity information extraction model to obtain a long-term dependency feature;

[0143] Calculate the word-level vector features of the corresponding entity information in the historical network alarm information to obtain coarse-grained word-level vector features and fine-grained word-level vector features;

[0144] Performing feature concatenation processing on the coarse-grained word-level vector features and the long-term dependency features to obtain coarse-grained embedding layer vector features, and performing feature concatenation processing on the fine-grained word-level vector features and the long-term dependency features to obtain fine-grained embedding layer vector features;

[0145] Using the attention mechanism in the entity information extraction model to select key fine-grained vector features in the fine-grained word-level vector features, and using the attention mechanism in the entity information extraction model to select key coarse-grained vector features in the coarse-grained embedding layer vector features;

[0146] Using the conditional random field model in the entity information extraction model to predict entity information of the historical network alarm information based on the key fine-grained vector features to obtain first predicted entity information, and using the conditional random field model to predict entity information of the historical network alarm information based on the key coarse-grained vector features to obtain second predicted entity information;

[0147] Perform a weighted sum operation of a loss function on a first loss function value between the first predicted entity information and the first entity information, and a second loss function value between the second predicted entity information and the second entity information to obtain a total loss function value;

[0148] If the total loss function value is less than a preset loss function threshold, it is determined that the entity information extraction model training is completed.

[0149] In a specific application scenario, the bidirectional long short-term memory network includes a forward long short-term memory layer and a reverse long short-term memory layer, and the extraction module 34 can be used to use the forward long short-term memory layer to obtain the long-term forward dependency feature from the start time point of the embedded vector to the preset time point, and use the reverse long short-term memory layer to obtain the long-term reverse dependency feature from the end time point of the embedded vector to the preset time point;

[0150] The long-term positive dependency feature and the long-term reverse dependency feature are feature fused to obtain the long-term dependency feature of the embedding vector.

[0151] It should be noted that for other corresponding descriptions of the functional units involved in the entity information extraction device provided in the present disclosure, please refer to Figure 1 and Figure 5 The corresponding description in will not be repeated here.

[0152] Based on the above Figure 1 and Figure 5 The method shown in the present disclosure, accordingly, also provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned Figure 1 and Figure 5 The method shown.

[0153] Based on this understanding, the technical solution of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of various implementation scenarios of the present disclosure.

[0154] Based on the above Figure 1 and Figure 5 The method shown, and Fig. 9 In order to achieve the above-mentioned purpose, the embodiment of the present disclosure also provides an electronic device that can be configured on the vehicle (such as an electric vehicle) side, and the device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-mentioned Figure 1 and Figure 5 The method shown.

[0155] Optionally, the above-mentioned physical device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), etc.

[0156] Those skilled in the art will appreciate that the above-mentioned physical device structure provided by the present disclosure does not constitute a limitation on the physical device, and may include more or fewer components, or a combination of certain components, or different arrangements of components.

[0157] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the above-mentioned physical device, and supports the operation of the information processing program and other software and / or programs. The network communication module is used to realize the communication between the components inside the storage medium, and the communication with other hardware and software in the information processing physical device.

[0158] Through the description of the above implementation modes, those skilled in the art can clearly understand that the present disclosure can be implemented by means of software plus a necessary general hardware platform, or can be implemented by hardware. According to the entity information extraction method, device and electronic device provided by the present invention, compared with the prior art, the present invention obtains multiple network alarm information to be annotated, obtains named entity representative words of multiple historical network alarm information, and obtains entity information annotated corresponding to the named entity representative words, wherein the entity information includes coarse-grained entity information and fine-grained entity information; extracts seed named entities in the network alarm information corpus according to the coarse-grained entity information and the fine-grained entity information, and identifies target-related terms related to the seed named entities in the network alarm information corpus, so as to expand the named entity representative words with the target-related terms; performs entity information annotating processing on the target-related terms with the expanded expanded named entity representative words, so as to obtain entity information annotated corresponding to the expanded named entity representative words; performs model training on an entity information extraction model with the historical network alarm information and the corresponding annotated entity information, and uses the trained entity information extraction model to extract entity information from multiple network alarm information to be annotated, so as to obtain entity information corresponding to multiple network alarm information to be annotated. For the disclosed scheme, by extracting seed named entities from the network alarm information corpus, and using the seed named entities, target-related terms related to the seed named entities are identified from the network alarm information corpus, thereby reducing the cost of manual labeling; because the named entity representative words are expanded by using target-related terms, the entity knowledge extraction model can learn richer features from more data, thereby improving the accuracy of tasks such as entity extraction.

[0159] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0160] The above are only specific embodiments of the present disclosure, so that those skilled in the art can understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to these embodiments herein, but will conform to the widest scope consistent with the principles and novel features applied herein.

Claims

1. A method for extracting entity information, characterized in that: The method comprises: Acquire multiple network alarm information to be annotated, acquire named entity representative words of multiple historical network alarm information, and acquire entity information annotated corresponding to the named entity representative words, wherein the entity information includes coarse-grained entity information and fine-grained entity information; Extracting seed named entities in a network alarm information corpus according to the coarse-grained entity information and the fine-grained entity information, and identifying target-related terms in the network alarm information corpus that are related to the seed named entities, so as to expand the named entity representative words with the target-related terms; Using the expanded named entity representative words to perform entity information tagging processing on the target related words to obtain the entity information corresponding to the expanded named entity representative words; The entity information extraction model is trained using the historical network alarm information and the corresponding annotated entity information, and the trained entity information extraction model is used to extract entity information from the multiple network alarm information to be annotated to obtain entity information corresponding to the multiple network alarm information to be annotated.

2. The method according to claim 1, characterized in that The step of identifying target-related terms related to the seed named entity in the network alarm information corpus to expand the named entity representative word by using the target-related terms includes: Performing word segmentation processing on the network alarm information in the network alarm information corpus, and removing stop words from the word items after the word segmentation processing to obtain target word items; Calculate the metric value of each target term according to the term frequency value and the inverse document frequency value of the target term in the network alarm information; A target term whose metric value is greater than a preset metric threshold is determined as a target-related term related to the seed named entity, so as to expand the named entity representative word with the target-related term.

3. The method according to claim 1, characterized in that The step of expanding the named entity representative word by using the target related words includes: Taking the target-related terms and the seed named entities as initial cluster centers, calculating the distance measurement value between each data point and the initial cluster center, and assigning each data point to the cluster where the initial cluster center corresponding to the minimum distance measurement value is located; According to the average value of the data points in each cluster, the cluster center is iteratively updated until the movement position of the cluster center and the preset cluster center is less than the preset movement position, so as to obtain the iterative expanded named entity representative words, wherein the expanded named entity representative words include the seed named entity, the target-related terms and the named entity representative words.

4. The method according to claim 1, characterized in that: The step of using the expanded named entity representative words to perform entity information tagging processing on the target related words to obtain the entity information tagged corresponding to the expanded named entity representative words includes: Calculating a first word vector of the target related term and calculating a second word vector of the expanded named entity representative word, and calculating a similarity value between the first word vector and the second word vector; The target-related terms with similarity values ​​greater than a preset similarity value are marked as corresponding entity information to obtain the entity information marked corresponding to the expanded named entity representative term.

5. The method according to claim 1, characterized in that The using the historical network alarm information and the corresponding annotated entity information to perform model training on the entity information extraction model includes: The historical network alarm information is used as the input feature of the entity information extraction model, and the entity information is used as the training label of the entity information extraction model. The entity information extraction model is iteratively trained until the total loss function value of the entity information extraction model is less than the preset loss function threshold, and it is determined that the training of the entity information extraction model is completed.

6. The method according to claim 5, characterized in that The entity information includes first entity information and second entity information; The using the historical network alarm information as the input feature of the entity information extraction model, using the entity information as the training label of the entity information extraction model, iteratively training the entity information extraction model until the total loss function value of the entity information extraction model is less than a preset loss function threshold, and determining that the entity information extraction model training is completed, includes: Performing word segmentation processing on the historical network alarm information, and removing stop words from the word items after the word segmentation processing, to obtain a text sequence of the historical network alarm information; The historical network warning information text sequence is converted into an embedded vector form by using a pre-trained language model in the entity information extraction model, and the embedded vector is subjected to feature extraction by using a bidirectional long short-term memory network in the entity information extraction model to obtain a long-term dependency feature; Calculate the word-level vector features of the corresponding entity information in the historical network alarm information to obtain coarse-grained word-level vector features and fine-grained word-level vector features; Performing feature concatenation processing on the coarse-grained word-level vector features and the long-term dependency features to obtain coarse-grained embedding layer vector features, and performing feature concatenation processing on the fine-grained word-level vector features and the long-term dependency features to obtain fine-grained embedding layer vector features; Using the attention mechanism in the entity information extraction model to select key fine-grained vector features in the fine-grained word-level vector features, and using the attention mechanism in the entity information extraction model to select key coarse-grained vector features in the coarse-grained embedding layer vector features; Using the conditional random field model in the entity information extraction model to predict entity information of the historical network alarm information based on the key fine-grained vector features to obtain first predicted entity information, and using the conditional random field model to predict entity information of the historical network alarm information based on the key coarse-grained vector features to obtain second predicted entity information; Perform a weighted sum operation of a loss function on a first loss function value between the first predicted entity information and the first entity information, and a second loss function value between the second predicted entity information and the second entity information to obtain a total loss function value; If the total loss function value is less than a preset loss function threshold, it is determined that the entity information extraction model training is completed.

7. The method according to claim 6, characterized in that The bidirectional long short-term memory network includes a forward long short-term memory layer and a reverse long short-term memory layer; The step of extracting features from the embedding vector using the bidirectional long short-term memory network in the entity information extraction model to obtain long-term dependency features includes: The forward long short-term memory layer is used to obtain long-term forward dependency features from the start time point of the embedding vector to a preset time point, and the reverse long short-term memory layer is used to obtain long-term reverse dependency features from the end time point of the embedding vector to the preset time point; The long-term positive dependency feature and the long-term reverse dependency feature are feature fused to obtain the long-term dependency feature of the embedding vector.

8. An entity information extraction device, the device comprising: An acquisition module, used to acquire multiple network alarm information to be annotated, acquire named entity representative words of multiple historical network alarm information, and acquire entity information annotated corresponding to the named entity representative words, wherein the entity information includes coarse-grained entity information and fine-grained entity information; An expansion module, configured to extract a seed named entity in a network alarm information corpus according to the coarse-grained entity information and the fine-grained entity information, and identify a target-related term related to the seed named entity in the network alarm information corpus, so as to expand the named entity representative word with the target-related term; A tagging module, used to perform entity information tagging processing on the target related terms by using the expanded expanded named entity representative terms to obtain the entity information corresponding to the annotation of the expanded named entity representative terms; The extraction module is used to use the historical network alarm information and the corresponding annotated entity information to train the entity information extraction model, and use the trained entity information extraction model to extract entity information from the multiple network alarm information to be annotated, so as to obtain the entity information corresponding to the multiple network alarm information to be annotated.

9. A non-transitory computer-readable storage medium having computer instructions stored thereon, characterized in that: The computer instructions are used to make a computer execute the method as claimed in claims 1-7.

10. A computer program product, characterized in that The invention comprises a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Citation Information

Cited By

  • Named entity recognition method and device, equipment and storage medium

    CN120373305A

  • Alarm information processing method, system and device and computer storage medium

    CN121193586A