A Chemical Process Knowledge Extraction Method Integrating Auxiliary Annotation and Dependency Analysis

Abstract

This invention discloses a method for knowledge extraction from chemical processes that integrates assisted annotation and dependency analysis, belonging to the field of natural language processing technology. The method includes: collecting production and operation data of the chemical process as raw corpus; processing and annotating a portion of the data in the raw corpus, including dataset segmentation, keyword acquisition, keyword annotation, dataset annotation, and dataset format conversion to obtain a labeled dataset; inputting the labeled dataset and the raw data into a knowledge extraction model for preliminary knowledge extraction, including BiLSTM-CRF model training, using the trained model for named entity recognition, and using dependency parsing for relation extraction; fusing the initial results to complete the knowledge extraction task; storing the extracted knowledge in a graph database in triple format for easy querying and learning by staff, reducing reliance on experts for operation and maintenance work, improving operation and maintenance efficiency, and reducing personnel training costs.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, specifically to a knowledge extraction method for chemical production processes, and more particularly to a chemical process knowledge extraction method that integrates auxiliary annotation and dependency analysis. Background Technology

[0002] Chemical processes are an indispensable part of modern society, playing a crucial role in the production of chemicals, pharmaceuticals, food, and materials. Due to the complexity of the chemical reactions and physical processes involved, the large number of equipment, and the vast amount of related production and maintenance data, the operation and maintenance of chemical processes are highly challenging. The textual information in these numerous production and maintenance documents is often difficult for laypeople to understand, resulting in high learning costs for newly hired staff. The over-reliance on experienced personnel and experts in the operation and maintenance of chemical processes limits production efficiency and increases labor costs for enterprises.

[0003] To address these issues, knowledge extraction is needed to transform this information into structured data that is easy to understand and query. Knowledge extraction techniques can be broadly categorized into three types: rule-based and dictionary-based methods, machine learning-based methods, and deep learning-based methods. Rule-based and dictionary-based methods rely on manually defined rules or templates for knowledge extraction. While effective for specific domains, they require manual intervention and customization, and have poor portability. Machine learning-based methods treat tasks like named entity recognition as sequence labeling problems, using statistical models to identify entities. Compared to rule-based methods, they offer better generalization capabilities, but require a large amount of labeled data for training. Deep learning-based methods primarily rely on various neural networks to complete knowledge extraction tasks. These methods can automatically learn complex text features and have achieved significant results in knowledge extraction tasks in recent years. However, the amount of data required for training is even larger, and it is difficult to collect sufficient data in actual chemical production processes. Therefore, it is necessary to label the available raw text data.

[0004] Given the diverse and complex forms of Chinese text data, current text annotation primarily relies on manual annotation using software or platforms. These software programs are not only cumbersome to install and have demanding operating environment requirements, but also operate independently of the knowledge extraction process, resulting in poor overall methodological integration. Furthermore, when the same keyword appears in different sentences, manual annotation often needs to be repeated, significantly wasting time and reducing the overall efficiency of knowledge extraction. Therefore, developing efficient and holistic data annotation methods is crucial. Summary of the Invention

[0005] The purpose of this invention is to provide a chemical process knowledge extraction method that integrates auxiliary annotation and dependency analysis, overcoming the technical problems in the above-mentioned technologies, accurately completing the knowledge extraction task of production data and operation and maintenance data, and having the characteristics of good overall method and high efficiency.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for extracting knowledge from chemical processes that integrates assisted annotation and dependency analysis includes the following steps:

[0008] S1, Raw data collection; Production and maintenance data involved in the chemical production process are collected as raw data; The collected raw data is plain text data and does not require image information;

[0009] S2, Data Preprocessing and Auxiliary Labeling; Data preprocessing includes: performing two operations simultaneously on the original corpus; the first part is to segment the text according to the number of natural paragraphs at a specified ratio to obtain a segmented dataset; the second part is to perform word segmentation, stop word removal, duplicate word removal, and part-of-speech filtering to obtain a list of words to be labeled; data auxiliary labeling labels some words in the list of words to be labeled to obtain a labeled list of words; the labeled list of words is matched with the segmented dataset, and the corresponding labels of the matched words are labeled on the corresponding words in the segmented dataset to obtain a labeled dataset; the labeled dataset is converted to a format to generate a BMEOS labeled dataset for subsequent model training;

[0010] S3, Model Training and Preliminary Knowledge Extraction: The model is obtained by combining a bidirectional long short-term memory neural network with a stochastic conditional field. This model is used to perform named entity recognition of chemical texts. First, the model is trained using the BMEOS labeled dataset generated in S2. The trained model is used to complete named entity recognition of chemical texts, obtaining a set of chemical text entities and their corresponding tag sets. Then, the entities in the chemical text entity set are paired with their corresponding tags to obtain multiple entity-entity tag binary sequence structures. All entity-entity tag binary sequence structures are combined to obtain a chemical text entity quadruplet sequence structure of head entity-head entity tag-tail entity-tail entity tag. Dependency parsing is used to extract relations, resulting in multiple initialization process knowledge triplet sequence structures of head entity-relation-tail entity.

[0011] S4, Initial Knowledge Extraction Result Fusion: The similarity between the head and tail entities of the chemical text entity quadruple sequence structure and the initial process knowledge triple sequence structure is calculated. For chemical text entity quadruple sequence structures with similarity reaching a set value, the corresponding data is fused with the initial process knowledge triple sequence structure. The "relationships" in the initial process knowledge triple sequence structure are completed in the chemical text entity quadruple sequence structure, and the head and tail entity labels are removed, forming a new head entity-relationship-tail entity chemical process knowledge triple sequence, thus completing the knowledge fusion.

[0012] S5, knowledge storage and knowledge retrieval; stores the chemical process knowledge triple sequence data of all new head entity-relationship-tail entity obtained by knowledge fusion as a chemical process knowledge graph.

[0013] Furthermore, the chemical process includes both large and small-scale chemical process data capable of collecting textual data such as chemical process technical specifications, technical manuals, and fault handling logs, and is not limited to a single production process. The production data and operation and maintenance data include chemical process technical specifications for chemical production lines, technical manuals for chemical production equipment, and fault handling logs for production processes.

[0014] Furthermore, step S2 includes the following steps;

[0015] S21, Dataset Segmentation; The original corpus is segmented according to a specified ratio based on the number of lines to obtain a training set, a validation set, and a test set to be labeled; The specified ratio is 6:2:2;

[0016] S22, Keyword Acquisition: First, each sentence in the original corpus is segmented into words. The word segmentation list is processed by removing stop words and deduplication to obtain a new word list. Then, the new word list is subjected to part-of-speech analysis to select words with the part of speech of noun (n), personal name (nr), and place name (ns), forming the final output word list to be labeled.

[0017] S23, Keyword Tagging; Select each category of words from the word list to be tagged obtained in S22, add a label to each selected category of words, and assign the label of each category of words to each word in that category to obtain a tagged word list.

[0018] S24, Dataset Labeling; Match the unlabeled training set, validation set, and test set in S21 with the labeled word list in S23, add the labels corresponding to the matched words to the training set, validation set, and test set, and output the labeled training set, validation set, and test set, saving them in JSON format.

[0019] S25, Dataset Format Conversion: Convert the JSON-formatted training, validation, and test sets from S24 into data files labeled using the BMEOS annotation method, and save them as text data.

[0020] Furthermore, the JSON format in S24 includes: a sentence text "text" key, a sentence number "id" key, and a label key in the sentence; wherein, the "label" key includes the word first character position "start", word last character position "end", word name "text", and word label "labels" for each label.

[0021] Further, the text data in S25 includes: splitting words by character, with each character of the word labeled according to BMEOS format; if a word has no label, all Chinese characters are labeled as O; if the word's label is T, all Chinese characters are labeled as T, single-character words are labeled as ST, and the label for each character of a multi-character word is represented as:

[0022]

[0023] in, For the first A label with one character. The position of the first Chinese character. A indicates the position of the last Chinese character; B indicates the position of the marked Chinese character at the beginning of the word; M indicates the position of the marked Chinese character in the middle of the word; E indicates the position of the marked Chinese character at the end of the word.

[0024] Furthermore, in S3,

[0025] S31, Input the chemical text data from the labeled BMEOS labeled dataset into the model for training; The model includes an input layer, a long short-term memory network layer, a stochastic conditional field layer, and an output layer;

[0026] Furthermore, the rule matching method of S33 includes using dependency parsing to perform dependency relation analysis on the words in the sentence, extracting the relations between the obtained words and dependency relations according to the matching rules, and obtaining the initialization process knowledge triple sequence structure of head entity-relation-tail entity.

[0027] Furthermore, the matching rules include, with the verb as the core relation, the "relation" combination rule in the initialization process knowledge triple sequence structure of the head entity-relation-tail entity is "the adverbial relation of the core relation word + the core relation word / attributive relation word + the verb complement structure of the core relation word + the right appended relation of the core relation word";

[0028] Entity matching rules include the following:

[0029] Entity matching rule one: the subject-predicate relation of the core relation word is used as the head entity, and the verb-object relation of the core relation word is used as the tail entity;

[0030] Entity matching rule two: the subject-predicate relation of the core relation word is used as the head entity, and the prepositional relation of the verb-complement structure of the core relation word is used as the tail entity;

[0031] Entity matching rule three: the prepositional phrase in the adverbial-verb structure of the core relation word is used as the head entity, and the pre-object of the core relation word is used as the tail entity;

[0032] Entity matching rule four: core relation words are used as head entities, and verb-object relationships of core relation words are used as tail entities.

[0033] Furthermore, S4 includes the following steps;

[0034] S41, use word2vec to obtain the vector representation of the entity words of the two chemical knowledge tuple sequences in S3;

[0035] S42, using the entity word vector representation of the two chemical knowledge tuple sequences in S41, calculate the similarity of the head entity and the tail entity respectively to obtain the head similarity and the tail similarity.

[0036] S43, if both the head similarity and the tail similarity are greater than the set threshold, then the relationship of the initialization process knowledge triplet sequence structure of head entity-relation-tail entity is added to the chemical text entity quadruple sequence structure of head entity-head entity label-tail entity-tail entity label, forming: head entity-head entity label-relation-tail entity-tail entity label structure. At the same time, the head entity label and tail entity label are removed, forming: head entity-relation-tail entity, thus completing knowledge extraction; otherwise, the quadruple is removed, and knowledge fusion of the next quadruple in the chemical text entity quadruple sequence is performed.

[0037] Furthermore, the formula for calculating similarity in S42 is as follows:

[0038]

[0039] in, The vector representing the entity in the chemical engineering text entity quadruple. A vector representing the triplet of knowledge in the initialization process. and Represent the first of the two vectors respectively. Individual chemical entity feature vector data, This represents the length of the feature vector of a chemical entity.

[0040] If both the head similarity and tail similarity are greater than the set threshold of 0.8, the relation of the corresponding triple is added to the quadruple with which it is used to calculate the similarity, forming a quintuple with the structure "head entity, head entity label, relation, tail entity, tail entity label". At the same time, the label in the tuple is removed, forming a triple with the structure "head entity, relation, tail entity", thus completing knowledge extraction. Otherwise, the quadruple is removed, and knowledge fusion is performed on the next quadruple in the chemical engineering text entity quadruple sequence.

[0041] The advantages of this invention are:

[0042] 1. The knowledge extraction method for chemical processes proposed in this invention does not require a large amount of standardized labeled data as a training basis, does not involve tedious data processing, and does not require a large amount of manual labeling work. The knowledge extraction task can be completed by collecting Chinese text data of production and operation and maintenance data (examples: process technology regulations of chemical production lines, operation manuals of chemical production equipment, and fault handling logs) in chemical plants, which has strong versatility and feasibility.

[0043] 2. The knowledge extraction method for chemical processes proposed in this invention is a chemical entity annotation method that integrates highly automated annotation techniques, thereby improving the overall effectiveness of the method. Simultaneously, it significantly reduces the number of manual annotations of the corpus, improving the efficiency of Chinese text annotation and thus enhancing the overall efficiency of chemical knowledge extraction. It demonstrates advanced technology and high efficiency in this task.

[0044] 3. The chemical knowledge extraction method proposed in this invention combines rule matching, which ensures extraction accuracy while reducing the requirements for training data, reducing data labeling pressure, and further improving the computational speed and efficiency of chemical knowledge extraction.

[0045] 4. The knowledge extraction method for chemical processes proposed in this invention does not require additional processing of Chinese text data in chemical processes. It only requires inputting the text-formatted data into the model to complete the extraction of knowledge from production data and operation and maintenance data (e.g., process technical specifications, equipment operation manuals, and fault handling logs).

[0046] 5. The knowledge extraction method for chemical processes proposed in this invention is a universal and applicable means of knowledge extraction for chemical processes, not targeted at any specific production process or production unit. It is applicable to chemical processes with production data and operation and maintenance data (e.g., process technology specifications for chemical production lines, operation manuals for chemical production equipment, and fault handling logs), and has excellent versatility. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the knowledge extraction method for chemical processes described in this invention.

[0048] Figure 2 This is a block diagram illustrating the principle of the highly automated chemical entity labeling method described in this invention.

[0049] Figure 3 This is a flowchart of step S23 of the present invention.

[0050] Figure 4 This is a block diagram illustrating the principle of the chemical knowledge extraction method described in this invention. Eh*, R*, and Et* represent the "head entity," "relation," and "tail entity" in the triplet results of relation extraction, respectively; Eh, Lh, Et, and Lt represent the "head entity," "head entity label," "tail entity," and "tail entity label" in the quadruple results of named entity recognition, respectively; and Sh and St represent the head entity similarity and tail entity similarity, respectively. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] This invention addresses the difficulties in learning and retrieving production and maintenance data during chemical production processes. It proposes a chemical process knowledge extraction method that integrates assisted annotation and dependency analysis. The main implementation flowchart of this method is attached. Figure 1 As shown, the following is based on Figure 1 The system describes the entire process, which is mainly divided into five steps. The overall process of the knowledge extraction method for chemical processes can be described as follows:

[0053] First, raw corpora are collected. Then, a portion of the data in the raw corpora is processed and labeled to obtain a labeled dataset. The labeled dataset and the raw data are input into a knowledge extraction model for preliminary knowledge extraction. Named entity recognition is performed using the trained model, and relation extraction is performed using dependency parsing. The initial results are then fused to complete the knowledge extraction task. Finally, the extracted knowledge is stored in a graph database in the format of triples.

[0054] The knowledge extraction method for chemical processes provided by this invention requires first collecting production and operation data of the chemical process as raw corpus. Then, partial data processing and annotation operations are performed on the raw corpus. The operation steps include dataset segmentation, keyword acquisition, keyword annotation, dataset annotation, and dataset format conversion to obtain a labeled dataset. The labeled dataset and the raw data are input into a knowledge extraction model for preliminary knowledge extraction, including BiLSTM-CRF model training, named entity recognition using the trained model, and relation extraction using dependency parsing. The initial results are then fused to complete the knowledge extraction task. The extracted knowledge is stored in a graph database in the format of triples for easy querying by staff, thereby reducing reliance on experts during operation and maintenance work.

[0055] Specifically, S1 refers to the collection of various production and operation data in the chemical process as raw data. This data includes, but is not limited to, the process technology specifications of the chemical production line, the technical manuals of chemical production equipment, and the troubleshooting logs. It mainly relies on the extent to which previous staff recorded the logs and the completeness of the chemical process design data.

[0056] S2 is a data preprocessing and auxiliary annotation tool, and its main implementation framework is shown in the attached figure. Figure 2 As shown, it mainly includes dataset segmentation, keyword extraction, keyword annotation, dataset annotation, and dataset format conversion.

[0057] S21 is a dataset segmentation method that detects the number of natural segments in the original corpus. The length of the segmented training set is The length of the validation set is The length of the test set is The ratio of the two parts is 6:2:2. , and The calculation formula is expressed as:

[0058]

[0059]

[0060]

[0061] S22 is keyword acquisition. First, the original corpus... Data preprocessing involves the model segmenting the input raw corpus into sentences, and then performing word segmentation, stop word removal, and deduplication on each sentence to obtain a preliminary vocabulary list. , represented as:

[0062]

[0063]

[0064]

[0065] in, Representing the original corpus, This represents the first in the chemical engineering corpus. The first sentence One word, This represents the total number of sentences in the chemical engineering corpus. This indicates a deduplication operation. Indicates the first List of chemical engineering terms after removing duplicates from sentences. This represents a list of stop words in chemical engineering texts. A list of terms indicating the initial processing.

[0066] Then, the obtained vocabulary list Perform part-of-speech analysis to filter out words with the part of speech of noun (n), personal name (nr), and place name (ns), forming the final output list of words to be labeled.

[0067] S23 is keyword tagging; the process is shown in the attached diagram. Figure 3 As shown in the diagram, the process first reads the list of words to be labeled obtained in S22, selects words with the same label that are to be labeled, inserts and saves the label. If further labeling is needed, the already labeled words are removed and the word list is updated; if further labeling is not needed, the labeled word list is output.

[0068] S24 is the dataset annotation process. It matches the unannotated training, validation, and test sets from S21 with the labeled word list from S23, adds the corresponding labels to the matching words, and outputs the annotated training, validation, and test sets in JSON format. The saved JSON format includes the sentence text "text", the sentence number "id", and the labels within the sentence "label". The "label" key includes the first word position "start", the last word position "end", the word name "text", and the word label "labels" for each label.

[0069] S25 is the dataset format conversion, which converts the labeled JSON training, validation, and test sets from S24 into text data. During annotation, words are first split into characters, and then each character is labeled according to the BMEOS format. If a word has no label, all Chinese characters are labeled as O; if a word's label is T, all Chinese characters are labeled as T, single-character words are labeled as ST, and the label for each character in a multi-character word is as follows:

[0070]

[0071] in, For the first A label with one character. The position of the first Chinese character. This indicates the position of the last Chinese character.

[0072] The principle flowchart of the knowledge extraction method is attached. Figure 4 As shown, it includes two parts: preliminary knowledge extraction and the fusion of preliminary knowledge extraction results.

[0073] S3 is the preliminary knowledge extraction, which mainly includes three parts: model training, named entity recognition, and relation extraction.

[0074] S31 is model training. Labeled data is input into a model combining a Bidirectional Long Short-Term Memory (BiLSTM) neural network and a Random Conditional Field (CRF) for training. The model includes an input layer, a BiLSTM network layer, a CRF layer, and an output layer. The BiLSTM captures contextual information from the text. The output of the BiLSTM generates a score for each label through a fully connected layer. The CRF layer receives the scores from the BiLSTM and considers the relationships between labels, finding the sequence with the highest probability among all possible label sequences. The BiLSTM-CRF neural network is fed in batches of 32 samples each, with parameters and hyperparameters such as the learning rate, spiking neuron threshold, and number of graph convolutional layers set, and the model is trained.

[0075] S32 is Named Entity Recognition. It uses a trained model for named entity recognition. The model takes sentence-level text as input and outputs multiple quadruple sequences with the structure "head entity, head entity label, tail entity, tail entity label".

[0076] S33 is a relation extraction method that uses rule matching to extract relations. The input is sentence-level text information, and the output is multiple triples with the structure "head entity, relation, tail entity". This method usually uses verbs as core relation words. The "relation" combination rule in the triple is "adverbial-head relation word of the core relation word + core relation word / attributive-head relation word + verb-complement structure of the core relation word + right-addressed relation word of the core relation word".

[0077] The entity matching rules in triples are as follows:

[0078] Entity matching rule 1: The subject-predicate relation of the core relation word is the head entity, and the verb-object relation of the core relation word is the tail entity; Entity matching rule 2: The subject-predicate relation of the core relation word is the head entity, and the prepositional relation of the verb-complement structure of the core relation word is the tail entity; Entity matching rule 3: The prepositional relation of the adverbial-head structure of the core relation word is the head entity, and the pre-object of the core relation word is the tail entity; Entity matching rule 4: The core relation word is the head entity, and the verb-object relation of the core relation word is the tail entity.

[0079] S4 is the fusion of preliminary knowledge extraction results, which includes three parts: entity word embedding, similarity calculation, and knowledge fusion.

[0080] S41 is entity word embedding, which uses word2vec to obtain the vector representation of entity words in the two chemical knowledge tuple sequences in S3.

[0081] S42 is the similarity calculation. Using the entity word vector representations of the two chemical knowledge tuple sequences in S41, the similarity is calculated for the head and tail entities respectively, yielding the head similarity and tail similarity. The similarity calculation formula is as follows:

[0082]

[0083] in, The vector representing the entity in the chemical engineering text entity quadruple. A vector representing the triplet of knowledge in the initialization process. and Represent the first of the two vectors respectively. Individual chemical entity feature vector data, This represents the length of the feature vector of a chemical entity.

[0084] S43 is knowledge fusion. If both the head similarity and the tail similarity are greater than the set similarity threshold of 0.8, the relation of the triple is added to the quadruple to form a quintuple with the structure "head entity, head entity label, relation, tail entity, tail entity label". At the same time, the label in the tuple is removed to form a triple with the structure "head entity, relation, tail entity", thus completing knowledge extraction. Otherwise, the quadruple is removed and knowledge fusion is performed on the next quadruple in the chemical text entity quadruple sequence.

[0085] S5 is a knowledge storage system. Knowledge extracted from production and operation and maintenance data (e.g., process technology regulations for chemical production lines, operation manuals for chemical production equipment, and fault handling logs) is stored in a graph database, facilitating staff access and learning, improving operation and maintenance efficiency, and reducing personnel training costs.

[0086] This invention proposes a knowledge extraction method for chemical processes. It establishes a knowledge extraction model with a highly automated chemical entity annotation method. The method annotates the collected raw corpus through an annotation module, and uses BiLSTM-CRF and rule matching for knowledge extraction, obtaining knowledge in triplet format. This achieves efficient knowledge extraction, and the obtained knowledge is stored in a graph database for easy access and learning by staff, improving operational efficiency and reducing personnel training costs. The knowledge extraction method proposed in this invention is particularly advantageous for data such as the numerous production and operational data in chemical processes. Traditional deep learning models require manual annotation and processing of text data beforehand, resulting in poor overall consistency. When the same keyword appears in different sentences, manual annotation often needs to be repeated, which not only wastes a lot of time but also reduces the overall efficiency of the knowledge extraction task. The model proposed in this invention obtains a tagged vocabulary list by performing operations such as word segmentation, part-of-speech filtering, and category labeling on the original corpus. It then skims the original data, automatically labels the matched words, and finally performs data format conversion. The output data can be directly used for deep learning model training, which greatly improves the efficiency of knowledge extraction tasks and the overall integrity of knowledge extraction models.

[0087] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example.

[0088] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

[0089] Finally, it should be noted that the basic principles, main features, and advantages of the present invention have been described above. The above embodiments are merely illustrative of the technical solutions of this application and are not intended to limit it. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the scope of the technical solutions of the embodiments of this application. Various changes and modifications can be made to the present invention without departing from the spirit and scope of the invention, and all such changes and modifications fall within the scope of the present invention as claimed.

Claims

1. A method for extracting knowledge from chemical processes by integrating auxiliary annotation and dependency analysis, characterized in that, Includes the following steps: S1, Raw data collection; collect production and operation data involved in the chemical production process as raw data; the raw data is plain text data. S2, Data Preprocessing and Data-Auxiliary Labeling: The data preprocessing yields two parts: the first part is the segmented dataset, and the second part is the list of words to be labeled. A new list of words is obtained by segmenting each sentence in the original corpus, removing stop words and duplicates, and then performing part-of-speech tagging on this new list to form the final output list of words to be labeled. Data-aided labeling annotates a portion of the words in the list of words to be labeled, resulting in a tagged list of words. The tagged list of words and the segmented dataset are then labeled to obtain a dataset labeled in BMEOS format. S3, Model Training and Preliminary Knowledge Extraction: The model is obtained by combining a bidirectional long short-term memory neural network with a stochastic conditional field. This model is used to perform named entity recognition of chemical engineering texts. First, the model is trained using the BMEOS format-annotated dataset generated in S2. The trained model is used to complete named entity recognition of chemical engineering texts, obtaining a set of chemical engineering text entities and their corresponding tag sets. Then, the entities in the chemical engineering text entity set are combined with their corresponding tags to obtain a chemical engineering text entity quadruple sequence structure of head entity-head entity tag-tail entity-tail entity tag. Dependency parsing is used to extract relations, resulting in multiple initialization process knowledge triplet sequence structures of head entity-relation-tail entity. S4, preliminary knowledge extraction results fusion; The similarity between the head entity of the chemical text entity quadruple sequence structure and the head entity of the initialization process knowledge triple sequence structure is calculated. The similarity between the tail entity of the chemical text entity quadruple sequence structure and the tail entity of the initialization process knowledge triple sequence structure is also calculated. The data corresponding to the chemical text entity quadruple sequence structure and the initialization process knowledge triple sequence structure with the similarity reaching the set value are fused to complete the knowledge fusion. S5: Knowledge storage and knowledge retrieval; storing all new head entity-relationship-tail entity chemical process knowledge triple sequences obtained from knowledge fusion into a chemical process knowledge graph.

2. The chemical process knowledge extraction method integrating auxiliary annotation and dependency analysis according to claim 1, characterized in that, The chemical process includes large and small chemical process data, such as chemical process technology specifications, technical manuals, and fault handling log text data, which can be collected; the production data and operation and maintenance data include the process technology specifications of the chemical production line, the technical manuals of the chemical production equipment, and the production process fault handling logs.

3. The chemical process knowledge extraction method integrating auxiliary annotation and dependency analysis according to claim 1, characterized in that, S2 includes the following steps; S21, Dataset Segmentation; The original corpus is segmented according to a specified ratio based on the number of lines to obtain a training set, a validation set, and a test set to be labeled; The specified ratio is 6:2:2; S22, Keyword Acquisition: First, each sentence in the original corpus is segmented into words. The word segmentation list is processed by removing stop words and deduplication to obtain a new word list. Then, the new word list is subjected to part-of-speech analysis to select words with the part of speech of noun (n), personal name (nr), and place name (ns), forming the final output word list to be labeled. S23, Keyword Tagging; Select each category of words from the word list to be tagged obtained in S22, add a label to each selected category of words, and assign the label of each category of words to each word in that category to obtain a tagged word list. S24, Dataset Labeling; Match the unlabeled training set, validation set, and test set in S21 with the labeled word list in S23, add the labels corresponding to the matched words to the training set, validation set, and test set, and output the labeled training set, validation set, and test set, saving them in JSON format. S25, Dataset Format Conversion: Convert the JSON-formatted training, validation, and test sets from S24 into data files labeled using the BMEOS annotation method, and save them as text data.

4. The chemical process knowledge extraction method integrating auxiliary annotation and dependency analysis according to claim 3, characterized in that, The JSON format in S24 includes: sentence text (text key), sentence number (id key), and label key in the sentence; wherein, the label key includes the word start position, word end position, word name (text), and word label (labels) for each label.

5. The chemical process knowledge extraction method integrating auxiliary annotation and dependency analysis according to claim 3, characterized in that, The text data in S25 includes: splitting words by character, with each character of the word labeled according to BMEOS format; if the word has no label, all Chinese characters are labeled as O; if the word's label is T, all Chinese characters are labeled as T, single-character words are labeled as ST, and the label for each character of a multi-character word is represented as: ， in, For the first A label with one character. The position of the first Chinese character. A indicates the position of the last Chinese character; B indicates the position of the marked Chinese character at the beginning of the word; M indicates the position of the marked Chinese character in the middle of the word; E indicates the position of the marked Chinese character at the end of the word.

6. The chemical process knowledge extraction method integrating auxiliary annotation and dependency analysis according to claim 1, characterized in that, In S3, S31, Input the chemical text data from the labeled BMEOS format dataset into the model for training; The model includes an input layer, a long short-term memory network layer, a stochastic conditional field layer, and an output layer; S32 uses a trained model for named entity recognition. The input of the model is sentence-level text information, and the output consists of multiple quadruple sequences with the following structure: head entity, head entity label, tail entity, and tail entity label. S33 uses rule matching to extract relations. The input is sentence-level text information, and the output is multiple triplets with the structure of head entity, relation, and tail entity.

7. The chemical process knowledge extraction method integrating auxiliary annotation and dependency analysis according to claim 6, characterized in that, The rule matching method of S33 includes using dependency parsing to perform dependency relation analysis on the words in the sentence, extracting the relations between the obtained words and dependency relations according to the matching rules, and obtaining the initialization process knowledge triple sequence structure of head entity-relation-tail entity.

8. The chemical process knowledge extraction method integrating auxiliary annotation and dependency analysis according to claim 7, characterized in that, The matching rules include, with verbs as the core relation, the relation combination rules in the initialization process knowledge triple sequence structure of head entity-relation-tail entity are: the inverse relation of the core relation word + the core relation word / attributive relation word + the verb complement structure of the core relation word + the right appended relation of the core relation word; Entity matching rules include the following: Entity matching rule one: the subject-predicate relation of the core relation word is used as the head entity, and the verb-object relation of the core relation word is used as the tail entity; Entity matching rule two: the subject-predicate relation of the core relation word is used as the head entity, and the prepositional relation of the verb-complement structure of the core relation word is used as the tail entity; Entity matching rule three: the prepositional phrase in the adverbial-verb structure of the core relation word is used as the head entity, and the pre-object of the core relation word is used as the tail entity; Entity matching rule four: core relation words are used as head entities, and verb-object relationships of core relation words are used as tail entities.

9. The chemical process knowledge extraction method integrating auxiliary annotation and dependency analysis according to claim 6, characterized in that, S4 includes the following steps; S41, use word2vec to obtain the vector representation of the entity words of the two chemical knowledge tuple sequences in S3; S42, using the entity word vector representation of the two chemical knowledge tuple sequences in S41, calculate the similarity of the head entity and the tail entity respectively to obtain the head similarity and the tail similarity. S43, if both the head similarity and the tail similarity are greater than the set threshold, then the relationship of the initialization process knowledge triplet sequence structure of head entity-relation-tail entity is added to the chemical text entity quadruple sequence structure of head entity-head entity label-tail entity-tail entity label, forming: head entity-head entity label-relation-tail entity-tail entity label structure. The head entity label and tail entity label are removed to form: head entity-relation-tail entity, thus completing knowledge extraction; otherwise, the quadruple is removed, and knowledge fusion of the next quadruple in the chemical text entity quadruple sequence is performed.

10. The chemical process knowledge extraction method integrating auxiliary annotation and dependency analysis according to claim 9, characterized in that, The formula for calculating similarity in S42 is as follows: ， in, The vector representing the entity in the chemical engineering text entity quadruple. A vector representing the triplet of knowledge in the initialization process. and Represent the first of the two vectors respectively. Feature vector data of chemical entities This represents the length of the feature vector of a chemical entity.

Images