Text mining-based ship fire risk factor identification method

Abstract

The invention discloses a ship fire risk factor identification method based on text mining. The ship fire risk factor identification method comprises the following steps: 1) obtaining ship fire risk text data and establishing a text database; 2) performing word segmentation on the ship fire risk text data to obtain a ship fire risk word segmentation feature item list; 3) obtaining a ship fire risk keyword list; 4) extracting related phrases, and constructing a keyword related phrase set; 5) clustering the keyword related phrase set, and performing semantic analysis on a clustering result to obtain a ship fire risk factor list; 6) calculating the occurrence probability of the risk factors, and obtaining a ship fire risk factor probability table; according to the ship fire hazard risk factor risk assessment method, the importance of the risk factors and the fire hazard scene is quantitatively assessed through the text mining technology and the semantic analysis model.

Description

TECHNICAL FIELD

[0001] The present application relates to natural language processing technology, and in particular to a ship fire risk factor identification method based on text mining. BACKGROUND

[0002] Ship fire accidents are one of the most important accidents that threaten maritime safety. The ship electrical system is complex, and the occurrence of fire has the characteristics of strong concealment and fast speed. Once a fire accident occurs, the complex cabin structure and equipment system are easy to cause the spread of the fire, and then seriously threaten the safety of life and property on the ship. In 2022, the fire ratio (excluding sunken ships) reached 44%, and has become the second largest accident causing ship loss. In 2017-2022, the insurance loss value caused by fire accounted for 26.5% of the total loss value, which has exceeded sunken ships and collisions, and has become the largest accident causing maritime insurance claims. Effectively identifying key risk factors in ship fire accidents and determining the accident scene have certain practical significance for preventing the occurrence of ship fire accidents.

[0003] At present, the ship fire risk identification mainly adopts the traditional safety analysis method, which highly depends on expert experience and has strong subjectivity and insufficient data support, resulting in that the identified risk factors are relatively rough and difficult to meet the evaluation needs of complex systems. Therefore, researchers use fuzzy set theory to improve the traditional method to realize the semi-quantitative identification of risk factors. However, this method is still limited by the quality of expert scoring and is difficult to ensure objectivity. Based on this, researchers began to use ship fire accident investigation reports to identify risk factors from real cases, which are more detailed and objective than expert experience. However, the existing analysis methods mostly rely on manual operation, which is low in efficiency and poor in consistency, and it is difficult to solve the problems of various report forms and high non-structured degree of text content.

[0004] With the development of natural language processing technology, researchers began to use these technologies to efficiently process accident reports and extract key information. Although existing research has summarized typical fire factors through word frequency analysis, it has not fully considered the semantic association of the overall expression of the text. At the same time, due to the different formats and standards of the report compilation, the uncertainty of the written expression is high, and the non-structured characteristics are significant, so the traditional text mining technology is difficult to effectively mine the key information therein, therefore, in order to improve the accuracy of the text mining technology to identify ship fire risks, professional field dictionaries and optimization algorithms need to be introduced to improve the text mining process through knowledge training. SUMMARY

[0005] The technical problem to be solved by the present application is to provide a ship fire risk factor identification method based on text mining in view of the defects in the prior art.

[0006] The technical scheme adopted by the present application to solve its technical problems is: a ship fire risk factor identification method based on text mining, comprising the following steps: 1) Obtain ship fire risk text data and establish a text database; 2) Tokenize the ship fire risk text data to obtain a ship fire risk tokenization feature item list; 3) Extract keywords based on the ship fire risk tokenization feature item list to obtain a ship fire risk keyword list; 4) Extract relevant word groups based on the ship fire risk keyword list to construct a keyword relevant word group set; 5) Cluster the keyword relevant word group set and perform semantic analysis on the clustering results to obtain a ship fire risk factor list; 6) Based on the ship fire risk factor list, calculate the occurrence probability of the risk factors to obtain a ship fire risk factor probability table; 7) Based on the ship fire risk factor probability table and the accident logic, construct a Boolean risk structure model; 8) Based on the ship fire risk factor probability table and the Boolean risk structure model, calculate the critical importance of the risk factors and the minimum risk set to form a ship fire risk factor and minimum risk set sorting list.

[0007] According to the above scheme, in the step 1), the sources of the ship fire risk text data include ship accident reports, industry standards, scientific research papers, shipping logs, and news reports.

[0008] According to the above scheme, in the step 1), it also includes the cleaning and preprocessing steps of the ship fire risk text data; specifically including removing irrelevant information, de-duplication, standardizing the text format and filtering stop words.

[0009] According to the above scheme, in the step 2), the tokenization feature item list is obtained, including: Based on the ship fire risk text database, the ship fire text data is tokenized by using the tokenization tool in natural language processing; In the tokenization process, a tokenization algorithm based on word frequency statistics or deep learning model is used to optimize the tokenization process; Based on the processed tokenization results, a ship fire risk tokenization feature item list is established.

[0010] According to the above scheme, in the step 3), based on the ship fire risk tokenization feature item list, the keywords are extracted, including: Using the TF-IDF extraction algorithm, calculate the TF-IDF value of the tokenization to obtain a tokenization TF-IDF list; Based on the TF-IDF value of the tokenization TF-IDF list, the keywords of the ship fire risk text are screened out; In combination with domain knowledge, irrelevant text keywords are excluded to obtain screened keywords.

[0011] According to the above scheme, in the step 4), relevant word groups are extracted based on the ship fire risk text keyword list, and the specific process is as follows: The semantic similarity between the ship fire risk text keywords and the related words is calculated by using co-occurrence analysis and word vector model, and the related combination list of the ship fire risk text keywords and the related words is obtained.

[0012] According to the above scheme, in the step 5), the relevant word group set is clustered, and the clustering result is analyzed to form a ship fire risk factor list, which includes: Based on the associated combination list of the ship fire risk text keywords and the related words, the relevant word groups are clustered by using a clustering algorithm, and a word group set with similar semantics is identified; Based on the clustering result, the word group set is analyzed in combination with the professional knowledge in the field of ship fire to obtain a ship fire risk factor list.

[0013] According to the above scheme, in the step 6), the ship fire risk factor probability is obtained, including: Based on the ship fire risk factor list, the occurrence frequency of each risk factor in the ship fire risk text is calculated to obtain the ship fire risk factor probability , and the calculation formula is as follows:

[0014] Where, r i is the i th risk factor, is the total number of texts, is the number of texts containing the risk factor .

[0015] According to the above scheme, in the step 7), a Boolean risk structure model is constructed, and the specific process is as follows: Based on the ship fire risk factor probability table, the logical relationship between each risk factor is determined in combination with the domain knowledge; Based on the logical relationship between each risk factor, the logical relationship between these risk factors is expressed in the form of Boolean logic expression to construct a Boolean risk structure model of ship fire, and the minimum risk set is obtained by Boolean logic calculation.

[0016] According to the above scheme, in the step 8), the critical importance of each risk factor is calculated, and the calculation formula is as follows:

[0017]

[0018] in, It is the probability of a fire accident occurring. It is the probability of the i-th risk factor occurring. It is the critical importance of the i-th risk factor. This represents the i-th minimal cut set where the fire accident occurred.

[0019] According to the above scheme, in step 8), based on the Boolean risk structure of ship fire, the minimum risk set MRS is obtained by Boolean logic calculation, which is the accident scenario. The critical importance of MRS is calculated using the following formula.

[0020] in, It is the critical importance of the minimum risk set. It is the probability of the i-th MRS occurring.

[0021] The present invention also provides an electronic device, comprising: One or more processors; and storage devices for storing one or more programs, Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method described in any of the above schemes.

[0022] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements any of the methods described above.

[0023] The beneficial effects of this invention are: This invention designs and optimizes a ship fire risk identification process based on text mining technology, including ship fire risk text data acquisition and database establishment, Chinese word segmentation and word segmentation feature extraction, keyword and related word extraction, keyword related phrase clustering and semantic analysis, risk factor list generation, risk factor probability calculation, Boolean risk structure model construction and critical importance calculation, forming a complete ship fire risk identification technology system.

[0024] By using text mining technology and semantic analysis models, we have overcome the bottleneck of traditional methods in effectively processing unstructured and complex semantic data. We have quantitatively assessed the importance of risk factors and fire scenarios, providing a scientific basis for intelligent early warning and emergency prevention and control of ship fires. Attached Figure Description

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0027] like Figure 1 As shown, a method for identifying ship fire risk factors based on text mining includes the following steps: 1) Construction of a textual database of ship fire risk data Multi-source data was collected using automated web crawling, OCR recognition, manual data entry, and other data acquisition methods, including ship accident reports, industry standards, research papers, shipping logs, and news reports. Simultaneously, large language models (such as GPT) were used to assist in identifying key information points in the text. The text was then cleaned and preprocessed, including removing irrelevant content, deduplication, formatting, and filtering stop words, generating multilingual (Chinese and English) formatted text, and constructing a structured ship fire risk database.

[0028] 2) Word segmentation processing of ship fire risk text data For Chinese text, Jieba word segmentation tool combined with a professional thesaurus was used for word segmentation; for English text, NLTK tool was used for word segmentation and lemmatization. A large language model was used to optimize the contextual semantic analysis process. Furthermore, Wordcloud2 tool was used to convert the processed text data into a risk word cloud for visualization, and a list of feature terms for ship fire risk word segmentation was obtained.

[0029] 3) Keyword extraction for ship fire risk Based on the list of feature terms for ship fire risk word segmentation, the TF-IDF algorithm is used to extract highly representative keywords from the text (TF-IDF value > 10 or TF-IDF value ranking in the top 20%). Optionally, the vector representation of the keywords is calculated using Word2Vec embedding technology to further optimize the selection of keywords and obtain a list of keywords related to ship fire risk.

[0030] 4) Construction of a set of keyword phrases related to ship fire risk Based on a list of keywords related to ship fire risks, the cosine similarity between keywords and related words is calculated to extract related phrases and construct a set of related phrases for each keyword.

[0031] 5) Clustering and semantic analysis of related phrases related to ship fire risk keywords Clustering algorithms such as K-means, hierarchical clustering, and deep clustering can be used to classify the set of related keyword phrases into different categories. The SpaCy tool and a large language model are then used to perform semantic analysis on the clustering results to obtain risk factor category labels, resulting in a list of ship fire risk factors. 6) Probability calculation of ship fire risk factors Based on the frequency of occurrence of each risk factor in the database, the probability of its occurrence is calculated using statistical analysis algorithms. 7) Boolean logic calculation of ship fire risk By analyzing the logical relationships between risk scenarios using a large language model, the relationships between risk factors are transformed into logical operation expressions based on these relationships, and a Boolean risk structure for ship fires is constructed. Based on this Boolean risk structure, the minimum risk set within the structure is obtained through Boolean logic calculations, which represents the accident scenario of a ship fire. 8) Calculation of the critical importance of ship fire risk Based on the Boolean risk structure of ship fires, the critical importance of risk factors and accident scenarios is obtained using the critical importance calculation formula, forming a ranked list of ship fire risk factors and accident scenarios. In this embodiment, the specific process for constructing the ship fire risk text data database is as follows: 1. Obtain raw textual data on ship fire risks. Data sources include: Official websites and industry platforms: Domestic and international maritime management agencies: such as the China Maritime Safety Administration, the International Maritime Organization (IMO), the Marine Accident Investigation Branch (MAIB) of the UK, the Transport Safety Bureau of Australia (ATSB), and the Japan Transport Safety Board (JTSB), etc., provide accident investigation reports and regulatory documents; Industry associations and professional organizations: such as the Chamber of International Shipping (ICS), Lloyd's Register, and Det Norske Veritas (DNV), which publish industry technical standards and risk management reports; Ship operating companies and manufacturing companies: such as COSCO Shipping, Maersk, and Samsung Heavy Industries, which publish internal fire reports and safety manuals. Academic databases and research platforms: Domestic resources: such as CNKI, Wanfang Data, and VIP database, to obtain Chinese academic papers, conference proceedings, and survey data related to ship fires; International resources: such as Scopus, Web of Science, and IEEE Xplore, to collect research results on ship fire risk analysis, accident modeling, and prevention and control technologies.

[0032] News media and shipping logs: such as Xinhua News Agency, BBC and CNN, which include the latest news reports and related public opinion analysis on major ship fire accidents.

[0033] Expert Interviews and Field Research: Interviewees primarily included maritime investigation experts, fire engineers, ship designers, operations and maintenance personnel, and legal and insurance experts. The interviews were designed around topics such as accident causes, risk identification and prevention, supplementary accident information, and data and experience accumulation. Specific content covered fire causation analysis, accident investigation methods, fire prevention measures and improvement suggestions, with a focus on obtaining practical cases and implicit information. Interviews were conducted face-to-face or via video conference, with interview outlines sent in advance and the order of questions flexibly adjusted to ensure in-depth information gathering.

[0034] 2. Technical methods for obtaining textual data on ship fire risks: For electronic text data on ship fire risks: An automated web crawling program based on Python was developed, utilizing Selenium and BeautifulSoup tools to extract data from dynamic and static web pages. Ship fire-related keywords (such as "ship fire") were embedded in the crawling rules to filter target information and automatically remove advertising interference.

[0035] For paper-based text data related to ship fire risks: OCR text recognition technology, using Tesseract OCR and Google OCR API, is used to extract text from non-text format data (such as PDFs, images, and scanned documents).

[0036] Large language model information extraction: Optional, such as a large language model based on GPT, to automatically extract semantics from the acquired text data, and annotate key content (such as the cause of the fire, the impact of the accident and the response measures) to improve the efficiency and accuracy of information extraction. Data cleaning and standardization: The Python language is used to perform data cleaning operations, unify data formats, and transform multilingual data (Chinese and English) into structured text; stop word filtering and spell checking are performed to optimize text quality.

[0037] A textual database of ship fire risks was constructed based on the aforementioned original data sources and related technical methods.

[0038] In this embodiment, the specific process for word segmentation processing of ship fire risk text data is as follows: 1) For Chinese text, use the Jieba word segmentation tool for processing. Combine it with a professional thesaurus for word segmentation.

[0039] For example, for the sentence "A fire broke out on a ship due to an electrical fault, resulting in casualties," the segmentation result obtained using Jieba is as follows: ["ship", "due to", "electrical fault", "fire", ",", "resulting in", "personnel", "casualties"] Among them, "electrical fault" was identified as a technical term, and "fire" was also correctly segmented.

[0040] 2) For English text, use the NLTK (Natural Language Toolkit) tool for word segmentation.

[0041] For example, for the English sentence: "The ship caught fire due to electrical malfunction, causing significant damage and injuries," the result after NLTK word segmentation and lemmatization is as follows: ["The", "ship", "catch", "fire", "due", "to", "electrical", "malfunction", "cause", "significant", "damage", "and", "injury"] Among them, "catch" is the word form of "caught" after word reduction, and "damage" and "injury" were extracted as important risk words.

[0042] To optimize word segmentation results, during the segmentation of Chinese and English text data, on the one hand, the self-attention mechanism of large language models (such as GPT-4) and the Transforemer architecture are combined to perform semantic analysis of the context, identify polysemous words, eliminate ambiguous words, and derive the actual meaning of words. On the other hand, NER technology is used to identify specific entity words (ship type, fire type, accident time, accident location, accident cause, etc.) from the segmented words.

[0043] For example, in the sentence "In January 2024, the cargo ship Thomas suffered a fire during its voyage due to an electrical fault in the engine room," the large language model can identify "October 2023" as the date entity, "Thomas" as the ship name, "cargo ship" as the ship type, "during its voyage" as the location of the accident, "engine room" as the compartment where the fire started, "engine" as the faulty equipment, and "electrical fault" as the cause of the accident. The resulting word segmentation is as follows: [In January 2024, the cargo ship Thomas experienced a fire during its voyage due to an electrical malfunction in its engine room.] Risk term visualization involves using the Wordcloud2 tool to calculate the frequency of word segmentation and then visualizing the results. After word frequency analysis, a list of segmented terms related to ship fire risks can be obtained.

[0044] In this embodiment, the specific process for extracting keywords related to ship fire risks is as follows: Calculation and ranking of TF-IDF values ​​for ship fire risk segmentation TF (Term Frequency) represents term frequency, and IDF (Inverse Document Frequency) represents inverse document frequency. The TF-IDF value of each word segment is calculated using a Python-based TF-IDF algorithm, as shown in the following formula:

[0045]

[0046]

[0047] In the formula, Representing text data word segmentation in Number of times it appears; Representing text data The total number of word segments in the text; It is the total number of all text data; It includes word segmentation The number of text data.

[0048] Based on the TF-IDF calculation results, a list of TF-IDF values ​​for ship fire risk words is obtained. Words with TF-IDF values ​​> 10 or those in the top 20% of the TF-IDF value ranking are selected as ship fire risk keywords.

[0049] Optimize keyword selection based on Word2Vec embedding technology; Word2Vec is a word vectorization technique that captures semantic relationships between words by mapping them to points in a low-dimensional vector space. Based on Skip-Gram or CBOW (Continuous Bag of Words) models, it converts words into fixed-length vectors and calculates the similarity between words based on context. By inputting segmented text data into the Word2Vec model, a vector representation of each word can be obtained. Then, the similarity between these words is calculated, and the words most relevant to the extracted keywords are selected for optimization.

[0050] Finally, based on the optimization results, keywords with large TF-IDF values ​​but no real meaning, such as "ship", "fire", "accident", and "ignition", were removed to obtain a list of keywords related to ship fire risk.

[0051] In this embodiment, the specific process for constructing the set of keyword-related phrases for ship fire risk is as follows: Cosine similarity is used to measure the semantic similarity between keywords and related words. Let the two n-dimensional feature word vectors be... , The formula for calculating cosine similarity is:

[0052] By setting a similarity threshold of 0.1, and using the cosine similarity between the keywords obtained above and other words, we can filter out word segments that are highly semantically related to the keywords, and then construct a set of related word groups for the keywords of ship fire risk.

[0053] In this embodiment, the specific process of clustering and semantic analysis of keyword-related phrases related to ship fire risk is as follows: Cluster analysis of keyword-related phrases for ship fire risk Clustering algorithms, such as K-means clustering, hierarchical clustering, or deep clustering, are used to perform clustering analysis on the above-obtained set of ship fire-related terms to obtain clustering results.

[0054] Identification of Ship Fire Risk Factors Based on Semantic Analysis Based on the clustering results described above, the SpaCy tool and a large language model are used to train the relevant word vectors through dependency parsing / deep learning models, thereby judging the accuracy of the clustering results and the actual semantics of the relevant word groups. For example, through semantic analysis of the words "electrical high temperature" and "circuit overheating," these two words belong to the same semantic category and are both part of the major risk factor category of "electrical fault."

[0055] Based on the results of the above clustering and semantic analysis, these related phrases are categorized into specific risk factor categories, with category labels such as "electrical failure," "mechanical failure," "fuel leak," and "operational error." Specific ship fire risk factor category labels are as follows: Electrical faults include "short circuit", "damaged electrical equipment", and "poor electrical contact". Mechanical failures include "engine overheating", "part damage", and "component aging". Improper operation category: including "operational error", "negligence", "smoking", etc. Fuel leaks include "pipeline rupture" and "fuel tank leak".

[0056] Based on the above tags, a list of ship fire risk factors is constructed.

[0057] In this embodiment, the specific process for calculating the probability of ship fire risk factors is as follows: Based on the constructed textual database of ship fire risks and the list of risk factors, and using Pandas tools, the probability of occurrence of risk factors is calculated using the following formula:

[0058] Where r i It is the i-th risk factor. This is the total amount of text data. It includes risk factors The amount of text data.

[0059] Using the above formula, the probabilities of the following risk factors can be calculated: Electrical fault: Occurred 150 times, total text quantity 500, probability:

[0060] Operational error: Occurred 100 times, total text quantity 500, probability:

[0061] Fuel leak: Occurred 50 times, total text count 500, probability:

[0062] The probability calculation for other risk factors can be performed by repeating the above process.

[0063] Based on the above calculation results, a probability table of ship fire risk factors is obtained.

[0064] The specific process for Boolean logic calculation of ship fire risk is as follows: 1) Analysis of the logical relationships between risk factors By using large language models (such as GPT-4, BERT, etc.), semantic analysis is performed on textual data on ship fire risk to extract causal relationships, interactions, and correlation patterns among different risk factors.

[0065] Here are some typical logical relations: Logical AND relationship: A fire may occur when two or more risk factors occur simultaneously. For example, an electrical fault and an operational error occurring simultaneously may cause a fire.

[0066] Logical OR relationship: Any one of certain risk factors can cause a fire. For example, either a fuel leak or an electrical fault could lead to a fire.

[0067] Conditional "if-then" relationship: Some risk factors may occur conditionally. For example, if there is aging electrical equipment on board a ship, the probability of its electrical failure will increase significantly, thereby exacerbating the risk of fire.

[0068] Establish the Boolean logic expression for the ship fire risk structure For each risk factor, we represent it with a Boolean value (0 or 1), where 1 indicates that the risk factor exists (i.e., occurs), and 0 indicates that the risk factor does not exist. The logical relationships between different risk factors are represented by Boolean operators (AND, OR, NOT).

[0069] For example, risk factors include: electrical fault (E), operational error (O), and fuel leak (F).

[0070] Based on the results of text analysis, the following Boolean expression can be obtained:

[0071] This means that a combination of electrical faults and operational errors, or a fuel leak, can lead to a ship fire.

[0072] For other risk factors, the internal logical relationships can be analyzed using the same method as above, and the corresponding Boolean logic expressions can be derived.

[0073] Based on the above Boolean logic expression, a Boolean risk structure for ship fires is constructed, as follows: Root node: Indicates whether a ship fire has occurred.

[0074] Intermediate nodes: represent various combinations of conditions (such as electrical faults and operational errors occurring simultaneously, or fuel leaks occurring).

[0075] Leaf nodes: Represent specific risk factors (such as electrical faults, operational errors, fuel leaks, etc.).

[0076] Based on the constructed Boolean risk structure for ship fires, Boolean operations are used to extract the minimum combination of events from the Boolean risk structure using logical solution methods (such as Karnaugh maps or conjunctive normal form (CNF) simplification techniques). By combining and simplifying the Boolean logic expressions, a set of risk factors is obtained. The occurrence of this set of risk factors constitutes the minimum condition leading to a fire, which is the accident scenario for ship fires.

[0077] The specific process for calculating the critical importance of ship fire risk is as follows: Based on the constructed Boolean logic structure for ship fire risk, the critical importance of risk factors and accident scenarios is calculated using Pandas tools. The calculation formula is as follows:

[0078]

[0079] in, It is the probability of a fire accident occurring. It is the probability of the i-th risk factor occurring. It is the critical importance of the i-th risk factor.

[0080] Based on the aforementioned Boolean risk structure for ship fires, the minimum risk set (MRS) is obtained through Boolean logic calculation, which is the accident scenario. The critical importance of the MRS is then calculated using the following formula.

[0081] in, It is the critical importance of the minimum risk set. It is the probability of the i-th MRS occurring.

[0082] Based on the calculation results of critical importance, the critical importance ranking of ship fire risk factors and accident scenarios can be obtained, thus identifying the key risk factors and accident scenarios in ship fire accidents.

[0083] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A ship fire risk factor identification method based on text mining, characterized in that, The method comprises the following steps: 1) obtaining ship fire risk text data and establishing a text database; 2) performing word segmentation on the ship fire risk text data to obtain a ship fire risk word segmentation feature item list; 3) extracting keywords based on the ship fire risk word segmentation feature item list to obtain a ship fire risk keyword list; 4) extracting relevant word groups based on the ship fire risk keyword list to construct a keyword relevant word group set; 5) clustering the keyword relevant word group set and performing semantic analysis on the clustering results to obtain a ship fire risk factor list; 6) calculating the occurrence probability of risk factors based on the ship fire risk factor list to obtain a ship fire risk factor probability table; 7) constructing a Boolean risk structure model based on the ship fire risk factor probability table and the accident logic; 8) calculating the critical importance of risk factors and the minimum risk set based on the ship fire risk factor probability table and the Boolean risk structure model to form a ship fire risk factor and minimum risk set sorting list.

2. The text mining-based ship fire risk factor identification method according to claim 1, characterized in that, In the step 1), the sources of the ship fire risk text data include ship accident reports, industry standards, scientific research papers, shipping logs and news reports. 3.The text mining based ship fire risk factor identification method according to claim 1, characterized in that, In the step 1), the ship fire risk text data also includes cleaning and preprocessing steps; specifically, irrelevant information is removed, duplicates are removed, the text format is standardized, and stop words are filtered.

4. The text mining-based ship fire risk factor identification method according to claim 1, characterized in that, In the step 2), the word segmentation feature item list is obtained, including: Based on the ship fire risk text database, the ship fire text data is processed by word segmentation tools in natural language processing; In the word segmentation process, a word segmentation algorithm based on word frequency statistics or deep learning model is used to optimize the word segmentation process; Based on the processed word segmentation results, a ship fire risk word segmentation feature item list is established. 5.The text mining based ship fire risk factor identification method according to claim 1, wherein, In the step 3), keywords are extracted based on the ship fire risk word segmentation feature item list, including: Using the TF-IDF extraction algorithm, the TF-IDF value of word segmentation is calculated to obtain a word segmentation TF-IDF list; Based on the TF-IDF value of the word segmentation TF-IDF list, the keywords of the ship fire risk text are screened out; Combined with domain knowledge, irrelevant text keywords are excluded to obtain the screened keywords.

6. The text mining-based ship fire risk factor identification method according to claim 1, characterized in that, In the step 4), relevant word groups are extracted based on the ship fire risk text keyword list, specifically as follows: Using co-occurrence analysis and word vector model to calculate the semantic similarity of ship fire risk text keywords and ship fire risk related words, and sorting, to obtain an associated combination list of ship fire risk text keywords and related words.

7. The text mining-based ship fire risk factor identification method according to claim 1, characterized in that, In the step 5), the relevant word group set is clustered, and the clustering results are semantically analyzed to form a ship fire risk factor list, including: Based on the associated combination list of ship fire risk text keywords and related words, a clustering algorithm is used to cluster the relevant word groups to identify a word group set with similar semantics; Based on the clustering results, the word group set is semantically analyzed based on the professional knowledge of ship fire field to obtain a ship fire risk factor list. 8.The text mining based ship fire risk factor identification method according to claim 1, wherein, In the step 6), the ship fire risk factor probability is obtained, including: Based on the ship fire risk factor list, the occurrence frequency of each risk factor in the ship fire risk text is calculated to obtain a ship fire risk factor probability The calculation formula is as follows: wherein r i is the ith risk factor, is the total number of texts, is the number of texts containing the risk factor . 9.The text mining based ship fire risk factor identification method according to claim 1, wherein, In the step 7), the Boolean risk structure model is constructed, specifically as follows: Based on the ship fire risk factor probability table, a logical relationship between the risk factors is determined in combination with domain knowledge; Based on the logical relationship between the risk factors, the logical relationship between the risk factors is expressed in the form of a Boolean logic expression, a Boolean risk structure model of ship fire is constructed, and a minimum risk set is obtained through Boolean logic calculation. 10.The text mining based ship fire risk factor identification method according to claim 1, wherein, In the step 8), a critical importance of each risk factor is calculated, and the calculation formula is as follows: wherein, is the probability of a fire accident occurring, is the probability of occurrence of the i-th risk factor, is the critical importance of the i-th risk factor, represents the i-th minimal cut set of a fire accident.

Images