Global labor employment risk early warning system and method based on AI drive

Abstract

The invention belongs to the technical field of AI, and discloses a global labor employment risk early warning system and method based on AI driving. Comprising the following steps: carrying out unstructured data acquisition; converting the unstructured data into text data; aiming at the text data, designing a rule knowledge base for identifying G risk types from the text data, the risk types including keywords, phrases and sentence patterns related to the risk types; calculating the key dynamic similarity in the keyword list by using a flexible cosine similarity method; extracting structured data from sentence patterns of the risk types through an information extraction technology, predicting a risk value of each risk type by using an SVR method, and calculating global employment risk data; setting a global employment risk threshold, and judging the risk severity; based on the risk severity, different early warnings are carried out; the recognition accuracy and the early warning response speed of the global labor employment risk are improved, and the stability of the global employment environment is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of AI technology, and specifically to an AI-driven global labor employment risk warning system and method. Background Art

[0002] Existing AI-driven global labor employment risk warning systems have many problems. First, existing global labor employment risk warning systems only collect structured data and text data, without considering image data and audio data; current enterprises' cross-border employment faces complex legal risks, such as: compliance problems caused by differences in labor laws in different countries; the opacity of adjudication standards in labor dispute cases; the low efficiency and limited coverage of traditional manual analysis of case laws.

[0003] Second, existing global labor employment risk warning systems often rely on centralized data storage and collection methods, and are easily affected by single-point failures, data tampering or loss;

[0004] In addition, existing global labor employment risk warning systems rely on static rules or keywords, such as pre-set keywords or using cosine similarity for keyword analysis, resulting in inaccurate recognition or omission when dealing with complex scenarios;

[0005] Finally, existing global labor employment risk warning systems are too conservative in dealing with low-risk situations, resulting in failure to respond in a timely manner when the risk changes greatly.

[0006] In view of this, the present invention proposes an AI-driven global labor employment risk warning system and method to solve the above problems. Summary of the Invention

[0007] In order to overcome the above defects of the prior art and to achieve the above object, the present invention provides the following technical solution, an AI-driven global labor employment risk warning system, including:

[0008] Unstructured data acquisition module: By deploying K_L nodes globally, using blockchain smart contracts to verify node identities, collecting unstructured data and encrypting it, generating redundant data blocks using erasure codes, storing them in IPFS and identifying them with CIDs for distributed storage, backup and reorganization;

[0009] Unstructured data conversion module: For image data and audio data in unstructured data, using image text intelligent extraction method to convert image data into text data; using audio text intelligent extraction method to convert audio data into text data;

[0010] Rule self - definition module: For text data, a rule knowledge base is designed to identify G risk types from the text data. The risk types include K_G keywords, phrases, and sentence patterns related to this risk type. Among them, the rule knowledge base includes a keyword list, a phrase list, and a sentence pattern list. The flexible cosine similarity method is used to calculate the dynamic similarity of key words in the keyword list.

[0011] Structured data prediction module: Through information extraction technology, extract the structured data within the F_D time period from the sentence patterns of each risk type, and after normalization, use the SVR method to predict the risk value of each risk type. Calculate the employment risk level data of each risk type in M countries through the weighted average method, and finally calculate the global employment risk data based on the employment risk level data of each risk type in M countries.

[0012] Global employment risk warning module: According to the historical global employment risk data, set the global employment risk threshold. Compare the global employment risk data with the global employment risk threshold to judge the severity of the risk. And based on the severity of the risk, conduct different warnings.

[0013] Low - risk warning module: When the global employment risk is at a low risk, the global employment risk warning module mainly archives the data, monitors abnormal fluctuations through the dynamic volatility smoothing index, and issues high - risk warnings or does not issue warnings.

[0014] Furthermore, the specific method of deploying K_L nodes globally, using blockchain smart contracts to verify node identities, collecting unstructured data and encrypting it, generating redundant data blocks using erasure codes, storing them in IPFS and identifying them with C ID for distributed storage, backup, and reorganization includes:

[0015] Step B1, Node setting: Deploy K_L nodes within the global scope.

[0016] Step B2, Node registration and authentication:

[0017] Apply the device or server to become a data collection node, and submit geographical location information, hardware configuration information, and public keys.

[0018] Verify the node identity through the blockchain smart contract.

[0019] The verified nodes are recorded on the blockchain, obtain a unique node ID, and become official unstructured data collection nodes for employment risks.

[0020] Step B3, Distributed Storage Optimization: Unstructured data is encrypted using a symmetric encryption algorithm, and redundant data blocks are generated using erasure codes; the redundant data blocks are stored in IPFS, uniquely identified using a C ID, and distributed stored and backed up; the data blocks are quickly retrieved via the C ID, decoded, and the unstructured data is reconstructed;

[0021] Unstructured data includes text data, image data, and audio data.

[0022] Furthermore, for the image data and audio data in the unstructured data, the specific methods of using image intelligent extraction to convert image data into text data and using audio intelligent extraction to convert audio data into text data include:

[0023] For image data, the median filtering method is used to denoise the image data to obtain denoised image data; the binarization algorithm is used to convert the denoised image data into a binary image; OCR technology is used to convert the content in the image data into preliminary text data; the NLP model is used to automatically correct the words misrecognized by OCR to obtain text data;

[0024] For audio data, the Wiener filtering method is used to eliminate noise, and then speech recognition technology is used to convert the audio content into text data.

[0025] Furthermore, for the text data, the specific method of designing a rule knowledge base for identifying G risk types from the text data includes:

[0026] Step A1, Design a rule knowledge base: Set the rules to include 3 lists, namely the keyword list, the phrase list, and the sentence pattern list;

[0027] The keyword list includes words directly related to labor employment, the phrase list includes phrase patterns directly related to labor employment, and the sentence pattern list includes sentences directly related to labor employment;

[0028] Among them, the keywords in the keyword list are obtained from the text data through the TF-IDF technology, the phrases in the phrase list are obtained through the n-gram model, and the sentences in the sentence list are obtained through the sentence segmentation technology;

[0029] Step A2, Automatically mark risk signals: Based on the keyword list, use the flexible cosine similarity method to identify the similarity between the keywords MQ and MP in the keyword list. If the similarity is higher than 80%, then set the keywords MQ and MP as the same type of keyword MQMP;

[0030] Based on the keyword MQMP, use string matching technology to query for phrases and sentence patterns containing the keywords MQ and MP in the phrase list and the sentence pattern list;

[0031] Phrases and sentence patterns containing the keywords MQ and MP, as well as the keyword MQMP, are denoted as the MQMP risk type;

[0032] Step A3: Repeat Step A2 to obtain G risk types.

[0033] Furthermore, the specific manner of using the flexible cosine similarity method to identify the similarity between the keywords MQ and MP in the keyword list includes:

[0034] Step C1: Based on the keywords Aa and Ab, use the Word2Vec method to convert them into keyword vectors;

[0035] Step C2: For the keyword vectors, use the cosine similarity to calculate the static similarity between the keywords;

[0036] Step C3: Obtain the frequencies of the keywords Aa and Ab in the rule knowledge base and normalize the frequencies;

[0037] Step C4: Introduce an adjustment coefficient to control the influence of the word frequency on the static similarity;

[0038] Step C5: Finally, use the static similarity, the normalized word frequency, and the adjustment coefficient to obtain the dynamic similarity between the keywords Aa and Ab.

[0039] Furthermore, the method for obtaining the global employment risk data includes:

[0040] For each risk type, use information extraction technology to extract the structured data within the F_D time period from the sentence patterns in the risk type, and after normalization processing, obtain the predicted value of each risk type through the SVR method;

[0041] According to the predicted value of each risk type, use the weighted average method to calculate the employment risk level data of each risk type in M countries;

[0042] Calculate the global employment risk data based on the employment risk level data of each risk type in M countries.

[0043] Furthermore, the specific manner of setting the global employment risk threshold based on the historical global employment risk data and comparing the global employment risk data with the global employment risk threshold to judge the severity of the risk includes:

[0044] Calculate the mean and standard deviation of global employment risk data over the past D time periods; set the global employment risk threshold based on the mean and standard deviation; the threshold is [μ - σ, μ + σ]. If the global employment risk data is greater than μ + σ, it is a high risk; if the global employment risk data is greater than or equal to μ - σ and less than or equal to μ + σ, it is a medium risk; if the global employment risk data is less than μ - σ, it is a low risk.

[0045] Further, the specific method of issuing different warnings based on the severity of the risk includes:

[0046] When it is a high risk, the global employment risk warning module will immediately send an emergency alert to international organizations;

[0047] When it is a medium risk, the global employment risk warning module will immediately send an emergency alert to government agencies.

[0048] Further, when the global employment risk is at a low risk, the specific method for the global employment risk warning module to mainly perform data archiving and monitor abnormal fluctuations through the dynamic volatility smoothing index for high-risk warning or no warning includes:

[0049] When it is a low risk, the global employment risk warning module mainly archives the global employment risk data, and at the same time detects abnormal fluctuations through the dynamic volatility smoothing index. If the volatility of the current global employment risk data is greater than the dynamic volatility smoothing index, a high-risk warning is triggered; if the volatility of the current global employment risk data is less than or equal to the dynamic volatility smoothing index, no warning is issued;

[0050] Among them, the steps of the dynamic volatility smoothing index are: first, obtain the volatility data, that is, the difference between two adjacent global employment risk data;

[0051] Secondly, set the smoothing coefficient, with a range of 0 to 1;

[0052] In addition, use random initialization for the initial volatility smoothing index, and calculate the dynamic volatility smoothing index through the dynamic volatility smoothing method.

[0053] The AI-driven global labor employment risk warning method, which is implemented by applying to the AI-driven global labor employment risk warning system, includes:

[0054] Step SS1: By deploying K_L nodes globally, using blockchain smart contracts to verify node identities, collecting unstructured data, encrypting it, generating redundant data blocks using erasure codes, storing them in IPFS and identifying them with C ID for distributed storage, backup, and reorganization;

[0055] Step SS2: For the image data and audio data in the unstructured data, use the intelligent image text extraction method to convert the image data into text data; use the intelligent audio text extraction method to convert the audio data into text data;

[0056] Step SS3: For the text data, design a rule knowledge base to identify G risk types from the text data. The risk types include K_G keywords, phrases, and sentence patterns related to this risk type. Among them, the rule knowledge base includes a keyword list, a phrase list, and a sentence pattern list; use the flexible cosine similarity method to calculate the dynamic similarity of key points in the keyword list;

[0057] Step SS4: Extract the structured data within the F_D time period from the sentence patterns of each risk type through information extraction technology, and after normalization, use the SVR method to predict the risk value of each risk type; calculate the employment risk level data of each risk type in M countries through the weighted average method, and finally calculate the global employment risk data based on the employment risk level data of each risk type in M countries;

[0058] Step SS5: Set the global employment risk threshold according to the historical global employment risk data; compare the global employment risk data with the global employment risk threshold to judge the severity of the risk; and based on the severity of the risk, conduct different warnings;

[0059] Step SS6: When the global employment risk is at a low level, the global employment risk warning module mainly archives the data, monitors abnormal fluctuations through the dynamic volatility smoothing index, and issues high-risk warnings or does not issue warnings.

[0060] Technical effects and advantages of the AI-driven global labor employment risk warning system and method of the present invention:

[0061] The present invention adopts a number of advanced technologies, improving the accuracy and efficiency of global employment risk monitoring and warning; by arranging multiple nodes globally, using blockchain to verify node identities and encrypt and store unstructured data to ensure the authenticity and security of the data;

[0062] At the same time, through the intelligent image extraction method and the intelligent audio extraction method, convert image and audio data into analyzable text, improving the processing ability of complex data;

[0063] Also design a flexible rule knowledge base, use the flexible cosine similarity method of keywords, phrases, and sentence patterns to identify risk types, predict risk values based on the SVR method, and finally calculate the global employment risk data through the weighted average method;

[0064] Through historical data analysis, the system sets dynamic thresholds for global employment risks and adopts multi-level early warning responses in combination with different risk levels; in case of high risks, it will send emergency alerts to international organizations, and in case of medium risks, it will notify government agencies;

[0065] In the early warning mechanism, the system sets up monitoring based on the dynamic volatility smoothing index, which can detect abnormal fluctuations in real time in a low-risk state and trigger high-risk early warnings;

[0066] In addition, in the low-risk early warning mechanism, the system sets up monitoring based on the dynamic volatility smoothing index, which can detect abnormal fluctuations in real time in a low-risk state and trigger high-risk early warnings; this flexible and refined early warning mechanism greatly enhances the ability to respond to sudden labor employment risks and ensures that risk emergency measures can be started quickly and effectively globally. Brief Description of the Drawings

[0067] Figure 1 It is a schematic diagram of the AI-driven global labor employment risk early warning system of the present invention;

[0068] Figure 2 It is a schematic diagram of the structure of the unstructured data acquisition module of the present invention;

[0069] Figure 3 It is a schematic diagram of the AI-driven global labor employment risk early warning method of the present invention. Detailed Embodiments

[0070] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0071] Embodiment 1

[0072] Please refer to Figure 1 and Figure 2 As shown, the AI-driven global labor employment risk early warning system in this embodiment includes:

[0073] Unstructured data acquisition module: By deploying K_L nodes globally, using blockchain smart contracts to verify node identities, collecting unstructured data and encrypting it, generating redundant data blocks using erasure codes, storing them in IPFS and identifying them with CID for distributed storage, backup, and reorganization;

[0074] Unstructured Data Conversion Module: For image data and audio data in unstructured data, use the intelligent image text extraction method to convert image data into text data; use the intelligent audio text extraction method to convert audio data into text data;

[0075] Rule Self - definition Module: For text data, design a rule knowledge base to identify G risk types from the text data; the risk types include K_G keywords, phrases, and sentence patterns related to this risk type; among them, the rule knowledge base includes a keyword list, a phrase list, and a sentence pattern list; use the flexible cosine similarity method to calculate the dynamic similarity of key points in the keyword list;

[0076] Structured Data Prediction Module: Extract structured data within the F_D time period from the sentence patterns of each risk type through information extraction technology, and after normalization, use the SVR method to predict the risk value of each risk type; calculate the employment risk level data of each risk type in M countries through the weighted average method, and finally calculate the global employment risk data based on the employment risk level data of each risk type in M countries;

[0077] Global Employment Risk Warning Module: Set the global employment risk threshold according to historical global employment risk data; compare the global employment risk data with the global employment risk threshold to judge the severity of the risk; and based on the severity of the risk, conduct different warnings;

[0078] Low - risk Warning Module: When the global employment risk data is at low risk, archive the data, and monitor abnormal fluctuations through the dynamic volatility smoothing index to issue a high - risk warning or not issue a warning.

[0079] Existing global labor employment risk warning systems usually rely on specific structured data in the data collection module, such as labor market statistical data, unemployment rate, employment numbers, wage levels, etc. released by the government; however, only collecting structured data has obvious drawbacks because it ignores the collection and analysis of image data and audio data; and there is also a large amount of valuable information in image and audio data, which can reveal changes in the labor market, social emotions, or potential risks of emergencies; existing global labor employment risk warning systems cannot capture these unstructured image and audio data, so there are lags or blind spots in identifying certain potential risks, and they cannot comprehensively and timely reflect the dynamic changes of the global labor market, resulting in the warning system lacking sensitivity and accuracy when facing emerging and complex risk situations;

[0080] The advantage of the present invention in collecting unstructured data lies in its ability to comprehensively capture and analyze diverse data sources such as images and audio, which contain a large amount of important information about social emotions, emergencies, and labor market dynamics. Different from the existing global labor employment risk warning systems that only rely on structured data (such as unemployment rate, number of employed people, etc.), the present invention can identify changes in social media, news reports, and public places in real time, and timely reflect potential labor employment risks, especially in complex and emerging risk scenarios, improving the sensitivity and accuracy of the warning system.

[0081] Existing global labor employment risk warning systems often rely on centralized data storage and collection methods, which means that all data is centrally stored in a single server or database. The main drawback of this method is its susceptibility to single-point failures. For example, if the server storing the data fails, is attacked by the network, or encounters technical problems, it may lead to data loss or inaccessibility, thereby affecting the warning function of the entire system. In addition, centralized storage is also vulnerable to the risk of data tampering or forgery. Especially when the security of the system is not effectively guaranteed, the integrity and accuracy of the data may be threatened, resulting in a reduction in the security of the unstructured data collection module in the AI-driven global labor employment risk warning system and rendering the stored data unauthoritative.

[0082] The specific method of distributed storage, backup, and reorganization by deploying K_L nodes globally, using blockchain smart contracts to verify node identities, collecting unstructured data and encrypting it, generating redundant data blocks using erasure codes, and storing them in IPFS and identifying them through CID includes:

[0083] Step B1, Node setting: Deploy K_L nodes globally. The nodes are registered through devices or servers and are used to collect unstructured data related to labor employment risks.

[0084] Step B2, Node registration and authentication: Each node needs to provide geographical location information, hardware configuration information (CPU, memory, storage, bandwidth, etc.), and public keys when registering; and use blockchain smart contracts to verify the identity of the nodes. The verified nodes will be recorded on the blockchain and obtain a unique node ID, officially becoming unstructured data collection nodes; these nodes will participate in data collection and storage.

[0085] The nodes obtain unstructured data from various data sources, including websites, APIs, sensors, and logs.

[0086] Step B3, Distributed storage: Store the collected unstructured data in IPFS.

[0087] Among them, the symmetric encryption algorithm is used to encrypt unstructured data to ensure that it will not be tampered with during the storage process;

[0088] The encrypted unstructured data is split into multiple data blocks, and these data blocks will be processed by erasure coding to generate redundant data blocks, ensuring that even if some data blocks are lost, the data can still be restored through the redundant data blocks;

[0089] The redundant data blocks are uploaded and stored in the IPFS network, and each redundant data block will obtain a unique CID when uploaded. This is the unique identifier for quickly locating and accessing the data. The data block can be quickly found through the CID; when data needs to be restored, the complete unstructured data is reconstructed by decoding the redundant data blocks;

[0090] Unstructured data includes text data, image data, and audio data;

[0091] Among them, IPFS is a distributed file storage and sharing protocol, aiming to store and transmit files in a decentralized manner. Different from traditional server-based storage systems, IPFS cuts files into multiple small pieces and distributes them for storage on multiple nodes in the network;

[0092] CID is the hash value used to uniquely identify a file or data block in IPFS. Each file or data block will generate a corresponding CID when uploaded to IPFS. This CID is an encrypted hash value based on the file content, ensuring the uniqueness and non-tamperability of each file or data block;

[0093] The symmetric encryption algorithm is an encryption method in which the same key is used for both data encryption and decryption; the encryption process converts plaintext data into ciphertext, and the decryption process uses the same key to restore the ciphertext to plaintext;

[0094] Erasure coding is a fault-tolerant technology that divides data into multiple data blocks and generates additional redundant data blocks, so as to ensure that the original data can still be restored through the remaining redundant blocks in the case of loss or damage of some data blocks;

[0095] By deploying multiple collection nodes globally and using a decentralized storage method (IPFS), the single point of failure problem can be effectively avoided. Even if some nodes fail or data is lost, due to the existence of redundant data blocks, the data can still be restored through other nodes, which greatly improves the fault tolerance of the system and the reliability of the data;

[0096] The symmetric encryption algorithm is adopted to encrypt the data, and the blockchain smart contract is used to verify the node identity, ensuring that the data source of each node is credible and cannot be tampered with; through the immutable feature, blockchain technology helps to ensure the integrity of data and prevent malicious tampering or forgery of data;

[0097] The distributed network enables the system to flexibly increase or decrease nodes according to needs, supporting real-time data collection and storage on a global scale; in addition, the IPFS network supports large-scale data storage and fast access, and data synchronization between different regions is more efficient, enhancing the scalability and flexibility of the overall system.

[0098] By using C ID (Content Identifier) to identify each redundant data block, fast positioning and access can be achieved; this ensures the fast retrieval and efficient reorganization of data after storage, improving the system response speed.

[0099] For image data and audio data in unstructured data, the image intelligent extraction method is used to convert image data into text data; the specific ways to use the audio intelligent extraction method to convert audio data into text data include:

[0100] Using the image intelligent extraction method to extract text data: for image data, the median filtering method is used to denoise the image data to obtain the denoised image data; the binarization algorithm is used to convert the denoised image data into a binary image; OCR technology is used to convert the content in the image data into preliminary text data; the NLP model is used to automatically correct the words misrecognized by OCR to obtain the text data;

[0101] Using the audio intelligent extraction method to extract text data: for audio data, the Wiener filtering method is used to eliminate noise, and then the speech recognition technology (ASR tool) is used to convert the audio content into text data;

[0102] Binarization algorithm: The binarization algorithm is an image processing technology used to convert a grayscale image into a black-and-white image; it sets a threshold, converting the part with pixel values higher than the threshold to white and the part lower than the threshold to black; the purpose of binarization is to simplify the image, making the text area more distinguishable from the background, which is beneficial for subsequent text extraction operations such as OCR;

[0103] OCR technology (Optical Character Recognition): OCR technology is a technology that can recognize characters in an image by computer and convert them into editable text; when the image contains text, OCR will recognize the shape, font, and arrangement of the text and convert it into digital text; it is commonly used for extracting text information from scanned documents, handwritten words, and images;

[0104] NLP Model (Natural Language Processing): An NLP model refers to artificial intelligence technology used to understand, process, and generate human language; during the processing of text data, the NLP model can identify and correct misidentifications that may occur in OCR technology, such as spelling mistakes and incorrect grammar, further improving the accuracy and readability of the text;

[0105] Wiener Filtering Method: The Wiener filtering method is an image processing and signal processing technology used to remove noise and smooth signals; in audio data processing, it estimates the statistical characteristics of the signal and noise, suppresses the noise according to the spectral characteristics of the signal, improves the audio quality, reduces the interference of background noise, and enhances the accuracy of speech recognition;

[0106] Speech Recognition Technology (ASR Tool): Speech recognition technology (ASR) is the technology of converting audio signals into text; it analyzes the acoustic patterns in speech and matches them with known vocabulary and speech models to identify words and sentences in the audio; ASR is widely used in fields such as voice assistants, transcription services, and real-time speech translation.

[0107] The defect of the existing global labor employment risk warning system is that it usually relies on static rules or preset keyword lists, which means that if a keyword with a similar meaning is not included in the rule base, the system cannot recognize this information, resulting in inaccurate recognition or omission when dealing with complex and dynamic scenarios. This method lacks adaptability to emerging risks or a rapidly changing social environment and cannot timely capture potential labor market risks;

[0108] For text data, the specific ways to design a rule knowledge base for identifying G risk types from text data include:

[0109] Step A1, Design a rule knowledge base: Set rules including 3 lists, namely a keyword list, a phrase list, and a sentence pattern list;

[0110] The keyword list includes words directly related to labor employment, the phrase list includes phrase patterns directly related to labor employment, and the sentence pattern list includes sentences directly related to labor employment;

[0111] Among them, the keywords in the keyword list are obtained from text data through the TF-IDF technology, the phrases in the phrase list are obtained through the n-gram model, and the sentences in the sentence list are obtained through the sentence segmentation technology;

[0112] Example: The keyword list includes words directly related to labor employment, such as: layoffs, wage freezes, labor disputes, strikes, protests, and lawsuits, etc.;

[0113] The phrase list includes phrase patterns directly related to labor employment, such as phrases like "mass layoffs", "class - action lawsuit by employees", "violation of labor laws", and "unpaid wages";

[0114] The sentence pattern list includes sentences directly related to labor employment, such as sentences like "a certain company is sued by a certain party due to a certain issue", "a large - scale strike occurs in a certain area", and "a certain party raises the minimum wage standard", etc.;

[0115] Step A2, automatically mark risk signals: Based on the keyword list, use the flexible cosine similarity method to identify the similarity between keywords MQ and MP in the keyword list. If the similarity is higher than 80%, then set keywords MQ and MP as the same type of keyword MQMP;

[0116] Based on keyword MQMP, use string - matching technology to query phrases and sentence patterns containing keywords MQ and MP in the phrase list and the sentence pattern list;

[0117] And record the phrases and sentence patterns containing keywords MQ and MP, as well as keyword MQMP, as the MQMP risk type;

[0118] Step A3, repeat Step A2 to obtain G risk types;

[0119] Example: Keyword list: ["layoffs", "staff reduction", "strike", "labor dispute"]; Phrase list: ["mass layoffs", "staff reduction plan", "collective strike", "labor dispute"]; Sentence pattern template list: ["..., a certain company is sued due to layoffs,...", "..., a certain company implements a staff reduction plan,...", "..., a strike occurs in a certain area,..."];

[0120] Identify keyword similarity. Assume the keywords to be compared are "layoffs" and "strike"; Use the flexible cosine similarity method to calculate the similarity. The result shows that the similarity is lower than 80%, so they are not classified as the same type of keyword;

[0121] Assume the keywords to be compared are "layoffs" and "staff reduction"; Use the flexible cosine similarity method to calculate the similarity. The result shows that the similarity is higher than 80%, so they are classified as the same type of keyword, and the keyword is re - recorded as "layoffs / staff reduction";

[0122] Step 2: Query phrases and sentence patterns in the phrase list and the sentence pattern list that contain the keyword "layoffs / staff reduction"; Phrase list: ["mass layoffs", "staff reduction plan", "collective strike", "labor dispute"]. Use string - matching technology to find phrases containing "layoffs" or "staff reduction"; Matching results: ["mass layoffs", "staff reduction plan"];

[0123] List of sentence pattern templates: ["..., a company was sued for layoffs,...", "..., a company implemented a staff reduction plan,...", "..., a strike occurred in a certain region,..."], use string matching technology to find sentence patterns containing "layoffs" or "staff reduction"; Matching results: ["..., a company was sued for layoffs,...", "..., a company implemented a staff reduction plan,..."];

[0124] Step 3: Uniformly record the classified keyword "layoffs / staff reduction", the matched phrases ["mass layoffs", "employee reduction plan"] and the sentence patterns ["..., a company was sued for layoffs,...", "..., a company implemented a staff reduction plan,..."] as the layoffs / staff reduction risk type;

[0125] TF-IDF technology: TF-IDF is a statistical method used to measure the importance of words in a text; it consists of two parts: term frequency (TF) represents the frequency of a certain word appearing in a document, and inverse document frequency (IDF) measures the rarity of the word appearing in the entire corpus; by combining these two, TF-IDF technology can highlight keywords that frequently appear in a specific document but are relatively rare in the entire corpus, thus helping to extract keywords with high information content and distinctiveness;

[0126] n-gram model: The n-gram model is a language model used to process and analyze text sequences. It represents the characteristics of the text by splitting the text into consecutive subwords of length n (n-grams); in this model, n can be any positive integer, and common ones are 1-gram, 2-gram, and 3-gram, etc.; the n-gram model can capture the local relationships and context information between words and is widely used in natural language processing tasks such as text generation, machine translation, and sentiment analysis;

[0127] Sentence segmentation technology is a basic task in natural language processing, aiming to split a continuous text into independent sentences; this technology usually relies on punctuation marks (such as full stops, question marks, and exclamation marks) to judge sentence boundaries for sentence division;

[0128] The idea of the present invention is to manage keywords, phrases, and sentence patterns separately by designing a flexible rule knowledge base, and combine TF-IDF technology, n-gram model, and sentence segmentation technology to extract relevant information, enabling the system to capture more comprehensive multi-level information related to labor employment; at the same time, using the flexible cosine similarity method to calculate the similarity between keywords can more accurately identify words and phrases with similar semantics but different expressions; in this way, the present system has higher sensitivity, accuracy, and adaptability in dealing with the ever-changing labor employment market.

[0129] The existing cosine similarity only calculates the static similarity based on the angle of word vectors, ignoring the word frequency information and context dynamics. As a result, the similarity of high-frequency words may be overestimated, while the similarity of low-frequency but semantically related words may be underestimated, leading to inaccurate calculated similarity, and thus causing inaccurate recognition and omission;

[0130] The specific ways to use the flexible cosine similarity method to identify the similarity between keywords MQ and MP in the keyword list include:

[0131] Step C1: Based on keywords Aa and Ab, use the Word2Vec method to convert them into keyword vectors;

[0132] Step C2: For the keyword vectors, use cosine similarity to calculate the static similarity between keywords;

[0133] Step C3: Obtain the frequencies of keywords Aa and Ab in the rule knowledge base and normalize the frequencies;

[0134] Step C4: Introduce an adjustment coefficient to control the influence of word frequency on the static similarity;

[0135] Step C5: Finally, use the static similarity, normalized word frequency, and adjustment coefficient to obtain the dynamic similarity between keywords Aa and Ab;

[0136] Among them, when calculating the static similarity, the formula is: Where, V Aa and V Ab are the keyword vectors of keywords Aa and Ab respectively, and sim cos (Aa, Ab) is the static similarity between keywords Aa and Ab;

[0137] The formula for calculating the dynamic similarity is: Based on the static similarity, introduce the normalized word frequency and adjustment coefficient to obtain the dynamic similarity. The formula is: Where, sim ada (Aa, Ab) is the dynamic similarity between keywords Aa and Ab, λ is the adjustment coefficient, and its value range is from 0 to 1, which is obtained by the random method. f Aa and f Ab are the normalized word frequencies of keywords Aa and Ab respectively;

[0138] Word2Vec is a technology that converts words into vector representations. It maps each word to a high-dimensional vector space through a neural network model, where words with similar meanings are closer in vector distance in this space;

[0139] The present invention uses a flexible cosine similarity method to identify the similarity of keywords, which has multiple advantages: First, by combining Word2Vec to convert into keyword vectors, the semantic information of keywords can be captured, improving the accuracy of similarity calculation; Second, using cosine similarity to calculate static similarity can effectively measure the relative position and semantic similarity between keywords; After introducing term frequency normalization and adjustment coefficients, it can not only eliminate the interference of high-frequency words on similarity calculation, but also dynamically adjust the similarity according to the actual term frequency, improving the adaptability and flexibility of the method, thereby improving the accuracy of calculating similarity and solving the problem of omission.

[0140] The ways to obtain global employment risk data include:

[0141] For each risk type, use information extraction technology to extract structured data within the F_D time period from the sentence patterns in the risk type, and after normalization processing, obtain the predicted value of each risk type through the SVR method;

[0142] According to the predicted value of each risk type, use the weighted average method to calculate the employment risk level data of each risk type in M countries;

[0143] Calculate the global employment risk data based on the employment risk level data of each risk type in M countries;

[0144] For example, there are a, b, c, d, and e risk types; in the a risk type, there are keywords and phrases and sentence patterns related to the keywords; use information extraction technology to extract structured data from the sentence patterns, normalize these data, and use them as the input of the SVR model to obtain the predicted value of the a risk type; among the M countries, there are also a, b, c, d, and e risk types, but the sentence patterns in the risk types are different, so the predicted values of the a risk type obtained using the SVR model are also different; now, it is necessary to calculate the global employment risk level data of the a risk type based on the predicted values of the a risk type in M countries; repeat this operation to calculate the global employment risk level data of the b, c, d, and e risk types; then calculate the global employment risk data based on the global employment risk level data of the a, b, c, d, and e risk types;

[0145] Among them, information extraction technology is a technology that automatically extracts structured information from unstructured or semi-structured data.

[0146] According to historical global employment risk data, set the global employment risk threshold; compare the global employment risk data with the global employment risk threshold, and the specific ways to judge the severity of the risk include:

[0147] Calculate the mean and standard deviation of the global employment risk data over the historical D time periods; set the global employment risk threshold based on the mean and standard deviation; the threshold is [μ - σ, μ + σ]. If the global employment risk data is greater than μ + σ, it is a high risk; if the global employment risk data is greater than or equal to μ - σ and less than or equal to μ + σ, it is a medium risk; if the global employment risk data is less than μ - σ, it is a low risk.

[0148] Specific ways to issue different warnings based on the severity of the risk include:

[0149] When it is a high risk, the global employment risk warning module will immediately send an emergency alert to international organizations;

[0150] When it is a medium risk, the global employment risk warning module will immediately send an emergency alert to government agencies.

[0151] Existing employment risk warning methods usually regard the "low risk" state as a safe state, and no warning will be triggered in this state. If the market, industry or policy environment suddenly changes drastically (such as the introduction of new regulations, large-scale enterprise layoffs), but it is still within the low risk threshold, there is actually a risk, but the existing methods will not give any warning about this, resulting in the failure to respond to sudden events in a timely manner;

[0152] When the global employment risk is at a low risk, the global employment risk warning module mainly archives data, and monitors abnormal fluctuations through the dynamic volatility smoothing index. Specific ways to issue a high risk warning or not include:

[0153] When it is a low risk, the global employment risk warning module mainly archives the global employment risk data, and at the same time detects abnormal fluctuations through the dynamic volatility smoothing index. If the volatility of the current global employment risk data is greater than the dynamic volatility smoothing index, the high risk warning strategy is triggered. If the volatility of the current global employment risk data is less than or equal to the dynamic volatility smoothing index, no warning is issued;

[0154] Among them, the steps of the dynamic volatility smoothing index are: first, obtain the volatility data, that is, the difference between two adjacent global employment risk data;

[0155] Secondly, set the smoothing coefficient, and the range is from 0 to 1;

[0156] In addition, use random initialization for the initial volatility smoothing index, and calculate the dynamic volatility smoothing index through the dynamic volatility smoothing method;

[0157] The calculation formula of the dynamic volatility smoothing method is: Among them, VSI t is the volatility smoothing index at time t, α is the smoothing coefficient, δ t―jis the volatility data at the t-j moment. The volatility data is the difference between two adjacent global employment risk data. j is the volatility index, H_L is the total number of volatilities, and VSI t―1 is the volatility smoothing index at the t-1 moment;

[0158] By introducing a dynamic volatility smoothing index, abnormal fluctuations can still be monitored in a low-risk state, thus avoiding the risk brought by emergencies ignored by existing methods. This method can dynamically adjust the risk warning threshold, adapt to changes, improve the sensitivity and accuracy of early warning, reduce false alarms and ensure timely detection of potential risks; at the same time, through data archiving and historical volatility analysis, the tracking of long-term trends and decision support are enhanced, making the overall risk warning system more intelligent, stable and efficient.

[0159] In this embodiment, a number of advanced technologies are adopted to improve the accuracy and efficiency of global employment risk monitoring and early warning; by deploying multiple nodes globally, using blockchain to verify node identities and encrypt and store unstructured data to ensure the authenticity and security of data;

[0160] At the same time, through the image intelligent extraction method and the audio intelligent extraction method, image and audio data are converted into analyzable text, improving the processing ability of complex data;

[0161] A flexible rule knowledge base is also designed. The flexible cosine similarity method of keywords, phrases and sentence patterns is used for risk type identification, and the risk value is predicted based on the SVR method. Finally, the global employment risk data is calculated by the weighted average method;

[0162] Through historical data analysis, the system will set dynamic thresholds for global employment risks and adopt multi-level early warning responses in combination with different risk levels; in case of high risks, an emergency alarm will be sent to international organizations, and in case of medium risks, government agencies will be notified;

[0163] In the early warning mechanism, the system sets up monitoring based on the dynamic volatility smoothing index, which can detect abnormal fluctuations in real time in a low-risk state and trigger a high-risk early warning;

[0164] In addition, in the low-risk early warning mechanism, the system sets up monitoring based on the dynamic volatility smoothing index, which can detect abnormal fluctuations in real time in a low-risk state and trigger a high-risk early warning; this flexible and refined early warning mechanism greatly enhances the ability to respond to sudden labor employment risks and ensures that risk emergency measures can be quickly and effectively launched globally.

[0165] Embodiment 2

[0166] Please refer to Figure 3As shown in the figure, for the parts not described in detail in this embodiment, refer to the description in Embodiment 1. A global labor employment risk warning method based on AI driving is provided, including:

[0167] Step SS1: By deploying K_L nodes globally, using blockchain smart contracts to verify node identities, collecting unstructured data, encrypting it, generating redundant data blocks using erasure codes, storing them in IPFS and identifying them with C ID for distributed storage, backup, and reorganization;

[0168] Step SS2: For the image data and audio data in the unstructured data, use the image text intelligent extraction method to convert the image data into text data; use the audio text intelligent extraction method to convert the audio data into text data;

[0169] Step SS3: For the text data, design a rule knowledge base for identifying G risk types from the text data. The risk types include K_G keywords, phrases, and sentence patterns related to this risk type; among them, the rule knowledge base includes a keyword list, a phrase list, and a sentence pattern list; use the flexible cosine similarity method to calculate the dynamic similarity of key points in the keyword list;

[0170] Step SS4: Extract the structured data within the F_D time period from the sentence patterns of each risk type through information extraction technology, perform normalization processing, and then use the SVR method to predict the risk value of each risk type; calculate the employment risk level data of each risk type in M countries through the weighted average method, and finally calculate the global employment risk data based on the employment risk level data of each risk type in M countries;

[0171] Step SS5: Set the global employment risk threshold according to the historical global employment risk data; compare the global employment risk data with the global employment risk threshold to judge the severity of the risk; and based on the severity of the risk, conduct different warnings;

[0172] Step SS6: When the global employment risk is at a low risk, the global employment risk warning module mainly archives the data, monitors abnormal fluctuations through the dynamic volatility smoothing index, and issues high-risk warnings or does not issue warnings.

[0173] Embodiment 3

[0174] This embodiment publicly provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it realizes the operation mode of the above-provided global labor employment risk warning system and method based on AI driving.

[0175] Since the electronic device introduced in this embodiment is the electronic device used to implement the AI-driven global labor employment risk warning system and method in the embodiments of the present application, based on the AI-driven global labor employment risk warning system and method introduced in the embodiments of the present application, those skilled in the art can understand the specific implementation manners and various variations of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiments of the present application will not be described in detail here. As long as those skilled in the art implement the electronic device used in the AI-driven global labor employment risk warning system and method in the embodiments of the present application, it falls within the scope of protection of the present application.

[0176] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to obtain a formula that is closest to the actual situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0177] The above is only the preferred implementation manner of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for ordinary technical users in the technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. An AI-driven global labor employment risk early warning system, characterized in that, Including: Unstructured data collection module: By deploying \(K_L\) nodes globally, using blockchain smart contracts to verify node identities, collecting unstructured data and encrypting it, generating redundant data blocks using erasure codes, storing them in IPFS and identifying them with CIDs for distributed storage, backup, and reorganization; Unstructured data conversion module: For image data and audio data in unstructured data, using image intelligent extraction methods to convert image data into text data; using audio intelligent extraction methods to convert audio data into text data; Rule self - definition module: For text data, designing a rule knowledge base to identify \(G\) risk types from the text data, where the risk types include keywords, phrases, and sentence patterns related to this risk type; Among them, the rule knowledge base includes a keyword list, a phrase list, and a sentence pattern list; Using the flexible cosine similarity method to calculate the dynamic similarity of key words in the keyword list; Structured data prediction module: Extracting structured data within the \(F_D\) time period from the sentence patterns of each risk type through information extraction technology, and after normalization, using the SVR method to predict the risk value of each risk type; Calculating the employment risk level data of each risk type in \(M\) countries through the weighted average method, and finally calculating the global employment risk data based on the employment risk level data of each risk type in \(M\) countries; Global employment risk warning module: Setting the global employment risk threshold according to historical global employment risk data; Comparing the global employment risk data with the global employment risk threshold to judge the severity of the risk; And based on the severity of the risk, conducting different warnings; Low - risk warning module: When the global employment risk is at a low risk, the global employment risk warning module mainly archives data and monitors abnormal fluctuations through the dynamic volatility smoothing index to issue high - risk warnings or not issue warnings.

2. The AI-driven global labor employment risk warning system according to claim 1, wherein The specific method of deploying \(K_L\) nodes globally, using blockchain smart contracts to verify node identities, collecting unstructured data and encrypting it, generating redundant data blocks using erasure codes, storing them in IPFS and identifying them with CIDs for distributed storage, backup, and reorganization includes: Step B1, Node setting: Deploy \(K_L\) nodes within the global scope; Step B2, Node registration and authentication: Applying a device or server to become a data collection node and submitting geographical location information, hardware configuration information, and public keys; Verifying the node identity through blockchain smart contracts; The verified nodes are recorded on the blockchain, obtaining a unique node ID and becoming official unstructured data collection nodes for employment risks; Step B3, Distributed storage optimization: Encrypting unstructured data using the symmetric encryption algorithm and generating redundant data blocks using erasure codes; Storing the redundant data blocks in IPFS, using CIDs for unique identification for distributed storage and backup; Quickly finding data blocks through CIDs, decoding, and reorganizing unstructured data; Unstructured data includes text data, image data, and audio data.

3. The AI-driven global labor employment risk early warning system according to claim 2, characterized in that, For the image data and audio data in unstructured data, the specific method of using image intelligent extraction method to convert image data into text data and using audio intelligent extraction method to convert audio data into text data includes: For image data, use the median filtering method to denoise the image data to obtain denoised image data; use the binarization algorithm to convert the denoised image data into a binary image; use OCR technology to convert the content in the image data into preliminary text data; use the NLP model to automatically correct the words misrecognized by OCR to obtain text data; For audio data, use the Wiener filtering method to eliminate noise, and then use speech recognition technology to convert the audio content into text data.

4. The AI-driven global labor employment risk warning system according to claim 3, characterized in that The specific method of designing a rule knowledge base for identifying G risk types from text data includes: Step A1, design a rule knowledge base: Set the rules to include 3 lists, namely the keyword list, the phrase list, and the sentence pattern list; The keyword list includes words directly related to labor employment, the phrase list includes phrase patterns directly related to labor employment, and the sentence pattern list includes sentences directly related to labor employment; Among them, the keywords in the keyword list are obtained from the text data through the TF-IDF technology, the phrases in the phrase list are obtained through the n-gram model, and the sentences in the sentence list are obtained through the sentence segmentation technology; Step A2, automatically mark risk signals: Based on the keyword list, use the flexible cosine similarity method to identify the similarity between the keywords MQ and MP in the keyword list. If the similarity is higher than 80%, then set the keywords MQ and MP as the same type of keyword MQMP; Based on the keyword MQMP, use the string matching technology to query the phrases and sentence patterns containing the keywords MQ and MP in the phrase list and the sentence pattern list; And record the phrases and sentence patterns containing the keywords MQ and MP and the keyword MQMP as the MQMP risk type; Step A3, repeat Step A2 to obtain G risk types.

5. The AI-driven global labor employment risk warning system according to claim 4, characterized in that, The specific method of using the flexible cosine similarity method to identify the similarity between the keywords MQ and MP in the keyword list includes: Step C1, based on the keywords Aa and Ab, use the Word2Vec method to convert them into keyword vectors; Step C2, for the keyword vectors, use the cosine similarity to calculate the static similarity between the keywords; Step C3, obtain the frequencies of the keywords Aa and Ab in the rule knowledge base and normalize the frequencies; Step C4, introduce an adjustment coefficient to control the influence of word frequency on the static similarity; Step C5, finally use the static similarity, the normalized word frequency, and the adjustment coefficient to obtain the dynamic similarity between the keywords Aa and Ab.

6. The AI-driven global labor employment risk warning system according to claim 5, characterized in that, The acquisition method of the global employment risk data includes: For each risk type, use information extraction technology to extract structured data within the F_D time period from the sentence patterns in the risk type, and after normalization processing, obtain the prediction value of each risk type through the SVR method; According to the prediction value of each risk type, use the weighted average method to calculate the employment risk level data of each risk type in M countries; Calculate the global employment risk data based on the employment risk level data of each risk type in M countries.

7. The AI-driven global labor employment risk warning system according to claim 6, wherein Set the global employment risk threshold according to the historical global employment risk data; The specific methods for comparing the global employment risk data with the global employment risk threshold to judge the severity of the risk include: Calculate the average value and standard deviation of the global employment risk data in the historical D time periods; set the global employment risk threshold based on the average value and standard deviation; the threshold is [μ - σ, μ + σ]. If the global employment risk data is greater than μ + σ, it is a high risk; if the global employment risk data is greater than or equal to μ - σ and less than or equal to μ + σ, it is a medium risk; if the global employment risk data is less than μ - σ, it is a low risk.

8. The AI-driven global labor employment risk early warning system according to claim 7, characterized in that The specific methods for giving different warnings based on the severity of the risk include: When it is a high risk, the global employment risk warning module will immediately send an emergency alert to the international organization; When it is a medium risk, the global employment risk warning module will immediately send an emergency alert to the government agency.

9. The AI-driven global labor employment risk early warning system according to claim 8, wherein When the global employment risk is at a low risk, the specific methods for the global employment risk warning module to mainly archive the data and monitor abnormal fluctuations through the dynamic volatility smoothing index, and give a high-risk warning or not give a warning include: When it is a low risk, the global employment risk warning module mainly archives the global employment risk data, and at the same time detects abnormal fluctuations through the dynamic volatility smoothing index. If the volatility of the current global employment risk data is greater than the dynamic volatility smoothing index, a high-risk warning is triggered. If the volatility of the current global employment risk data is less than or equal to the dynamic volatility smoothing index, no warning is given; Among them, the steps of the dynamic volatility smoothing index are: first, obtain the volatility data, that is, the difference between two adjacent global employment risk data; Secondly, set the smoothing coefficient, and the range is from 0 to 1; In addition, use random initialization for the initial volatility smoothing index, and calculate the dynamic volatility smoothing index through the dynamic volatility smoothing method.

10. An AI-driven global labor employment risk warning method, which is applied to the AI-driven global labor employment risk warning system according to any one of claims 1 to 9, and is characterized in that, Include: Step SS1: By deploying K_L nodes globally, use the blockchain smart contract to verify the node identities, collect unstructured data, encrypt it, generate redundant data blocks using erasure codes, store it in IPFS and identify it through CID for distributed storage, backup and reorganization; Step SS2: For the image data and audio data in the unstructured data, use the image text intelligent extraction method to convert the image data into text data; use the audio text intelligent extraction method to convert the audio data into text data; Step SS3: For the text data, design a rule knowledge base to identify G risk types from the text data. The risk types include K_G keywords, phrases and sentence patterns related to this risk type; Among them, the rule knowledge base includes a keyword list, a phrase list and a sentence pattern list; use the flexible cosine similarity method to calculate the critical dynamic similarity in the keyword list; Step SS4: Extract structured data within the F_D time period from the sentence patterns of each risk type through information extraction technology, and after normalization, use the SVR method to predict the risk values of each risk type; calculate the employment risk level data of each risk type in M countries through the weighted average method, and finally calculate the global employment risk data based on the employment risk level data of each risk type in M countries; Step SS5: Set the global employment risk threshold according to the historical global employment risk data; Compare the global employment risk data with the global employment risk threshold to judge the severity of the risk; and based on the severity of the risk, conduct different early warnings; Step SS6: When the global employment risk is at a low level, the global employment risk early warning module mainly archives data, monitors abnormal fluctuations through the dynamic volatility smoothing index, and issues high-risk early warnings or does not issue early warnings.

Images