Geological knowledge constrained text information extraction method and system

Abstract

The invention belongs to the technical field of geological text information processing, and discloses a geological knowledge constrained text information extraction method and system, and the method comprises the steps: defining a geological ontology model through obtaining and processing geological text data; annotating sentences according to the geological ontology model to obtain a geological entity relationship data set; constructing a geological named entity recognition model based on the span, and training the model by using the geological entity relationship data set; generating a geological entity pair by using the geological ontology model; and performing relation prediction by using the relation extraction model to realize geological information extraction of the input text. According to the method, redundant entity pairs are filtered through the geological ontology model, so that the unnecessary calculation amount is reduced, the processing speed is increased, and the accuracy of relation recognition is remarkably improved. Potential relation information is introduced into model input, the entity relation extraction precision is optimized, the value of knowledge in the geological field is fully played, and efficient automatic information extraction is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of geological text information processing, and in particular relates to a method and system for extracting text information constrained by geological knowledge. Background Art

[0002] With the continuous deepening of geological survey and exploration work, a vast amount of geological text materials have been accumulated in the geological field. These materials contain rich and valuable geological environment and ore deposit information, providing an important basis for geological exploration, ore prospecting prediction, and the management of geological resources. However, most current geological texts exist in unstructured data, and the processing methods often rely on manual labor. The traditional manual processing method not only has low efficiency and high cost, but is also easily affected by human factors, resulting in errors, which greatly limits the breadth and depth of its practical application.

[0003] To address this challenge, in recent years, geological text information extraction technology has emerged, bringing a revolutionary change to the automated processing of geological data. Existing geological information extraction technologies usually focus on extracting basic information such as stratigraphic distribution, lithological characteristics, and mineral resource distribution from geological texts. Although it has improved the information utilization efficiency to a certain extent and reduced the labor cost, there are still many deficiencies. Especially when dealing with complex geological knowledge, it still relies on simple keyword matching and rule-based extraction, lacking in-depth domain knowledge support.

[0004] Currently, most geological information extraction technologies mainly rely on general natural language processing methods, usually extracting knowledge based on text extraction models in the general field. Technologies in the general field often have broadness and relatively low professional depth, and cannot be deeply customized for specific geological disciplines. The research in the field of geology usually focuses on specific knowledge points such as specific ore deposit types, ore-forming processes, and geological structures, with unique domain characteristics. Therefore, more accurate text information extraction is required based on geological ontology models and professional knowledge.

[0005] However, the defect of the existing technology is that many geological information extraction methods fail to fully utilize the ontology models or domain knowledge unique to the geology field to constrain information extraction. Specifically, in the geological field, the relationship between an ore deposit and an ore body can be regarded as an inclusion relationship, while the relationship between an ore deposit and a stratum is usually an occurrence relationship. This kind of professional domain knowledge is crucial for accurately extracting geological information. For example, magmatic hydrothermal ore deposits are usually related to volcanic rocks or intrusive rocks, while stratiform hydrothermal ore deposits are more common in sedimentary rock formations. If these complex geological relationships cannot be fully incorporated into the information extraction model, it will directly affect the accuracy and effectiveness of information extraction.

[0006] Through the above analysis, the problems and defects existing in the prior art are as follows: Many geological information extraction methods fail to fully utilize the ontology models or domain knowledge unique to the geology field to constrain information extraction. Specifically, in the geological field, the relationship between ore deposits and ore bodies can be regarded as an inclusion relationship, while the relationship between ore deposits and strata is usually an occurrence relationship. Such professional domain knowledge is crucial for accurately extracting geological information. For example, magmatic hydrothermal deposits are usually related to volcanic rocks or intrusive rocks, while stratiform hydrothermal deposits are mostly found in sedimentary rock layers. If these complex geological relationships cannot be fully incorporated into the information extraction model, it will directly affect the accuracy and effectiveness of information extraction. Summary of the Invention

[0007] To overcome the problems existing in the related art, the disclosed embodiments of the present invention provide a method and system for text information extraction with geological knowledge constraints. The technical solutions are as follows:

[0008] The present invention is implemented as follows. The method for text information extraction with geological knowledge constraints includes the following steps:

[0009] S1. Preprocessing of geological text materials. Use a Python program to determine the data source for information extraction, delete irrelevant content in the text data, and obtain sentence-level geological text data after processing, thus realizing the acquisition and processing of geological text materials.

[0010] S2. Pre-define entity types and relationship types, and use the sentence-level geological text data to restrict the entity types and relationships in geological sentences; traverse all pairs of entity types to realize the construction of a geological ontology model.

[0011] S3. Annotate the sentence-level geological text data according to the geological ontology model to obtain a geological entity relationship data set, which is used for the training of subsequent named entity recognition models and relationship extraction models.

[0012] S4. Construct a span-based geological named entity recognition model, use the geological entity relationship data set to train the model, establish a geological named entity recognition model, and realize the recognition of geological entities in the input geological text.

[0013] S5. Generate geological entity pairs using the geological ontology model.

[0014] S6. Construct sentences containing geological knowledge based on the geological ontology model, use the geological relationship extraction model to perform relationship prediction, and realize the extraction of geological information from the input text.

[0015] In step S1, the acquisition and processing of geological text materials include:

[0016] Use a Python program to extract and separate text content, graphics, and tables from geological text materials, and select the text content as the data source for information extraction;

[0017] Apply a set of pattern matching rules to identify and extract chapters, and delete the irrelevant content in the text data;

[0018] By identifying the sentence-ending characters, the remaining paragraphs are segmented into sentence-level geological text data.

[0019] Furthermore, the chapters include titles, abstracts, chart and table titles, acknowledgments, and references;

[0020] The irrelevant content in the text data includes titles, author information, titles, acknowledgments, and references.

[0021] In step S2, the construction of the geological ontology model includes:

[0022] Pre-define entity types and relationship types, and extract ore-forming conditions and ore-forming anomaly information from the sentence-level geological text data; the entity types include tectonic units, metallogenic units, faults, tectonic deformations, ore deposits, strata, alteration, metamorphic rocks, sedimentary rocks, magmatic rocks, ore bodies, minerals, geophysical characteristics, chemical element characteristics, remote sensing markers, spatial positions, time, genetic types, grades, scales, ore types, and mineral zoning;

[0023] The relationship types include tectonic actions, ore-forming actions, sedimentary actions, metamorphic actions, magmatic actions, wall-rock alteration actions, control actions, limiting actions, indicating actions, ore-forming carriers or material sources, occurrence relationships, intrusion relationships, inclusion relationships, attribute relationships, spatial relationships, and time relationships;

[0024] Restrict the entity types and relationships in the sentence-level geological text data; traverse all pairs of entity types to determine whether the specified relationship types exist in each sentence, and obtain the geological ontology model; the geological ontology model consists of knowledge rules in the form of <head entity type, relationship type, tail entity type>, and this geological ontology model represents geological knowledge and is used to guide geological text annotation and information extraction.

[0025] In step S3, annotate the sentence-level geological text data according to the geological ontology model, including:

[0026] The constructed geological ontology model is used to automatically check the marked entity relationships, and the entity relationships that do not conform to the rules are automatically marked and modified again; through two rounds of iterative work of annotation, inspection, and revision, sentences with inclusion relationships are obtained and form a geological entity relationship dataset.

[0027] In step S4, a span-based geological named entity recognition model is constructed, and the model is trained using a geological entity relationship dataset, including:

[0028] The span-based geological named entity recognition model is used to recognize defined geological entities from the input sentence. The model mainly consists of three modules: text vector encoding, span generation and encoding, and span classification;

[0029] Text vector encoding module, when the input sentence is X = [x1, x2, …, x n , where x1 represents the 1st character, x2 represents the 2nd character, x n represents the nth character, [·] represents a list, the longest entity length is L characters, and the vector encoding e i of the ith character x i in the input sentence is generated using a pre-trained language model such as RobERTa;

[0030] Span generation and encoding module, which traverses all characters of the input sentence X to generate all potential spans S = {s 1,1 , s 1,2 , …, s 1,L , s 2,2 …, s n,n}, where s 1,1 = [x1], s 1,2 = [x1, x2], s 1,L = [x1, x2, …, x L , s 2,2 = [x2], s n,n = [x n , {·} represents a set; for the given ith span s i , the vector expression is h e (s i ) = [e STRAT(i) ; e END(i) ; w(s i )], where e STRAT(i) represents the vector encoding of the starting character of the span s i , e END(i) represents the vector encoding of the ending character of the span s i , and w(s i ) represents the length encoding of the span s i ;

[0031] Span classification module, which inputs h e (s i ) into a feedforward neural network to calculate the probability that the span s i belongs to a non-entity or a certain entity type, thereby obtaining the entity prediction result.

[0032] In step S5, a geological entity pair is generated using the geological ontology model, including:

[0033] After obtaining the entities in the input sentence, the entities are transformed into entity pairs to predict the semantic relationship between the two entities in the sentence, and the expression form is <head entity, entity type 1, tail entity, entity type 2, sentence>; when the entity list identified in the geological sentence X is E = [(e1, T1), (e2, T2), …, (e n , T n )], where e1 and T1 respectively represent the first entity in the sentence and the entity type corresponding to this entity, e2 and T2 respectively represent the second entity in the sentence and the entity type corresponding to this entity, e n , T n respectively represent the nth entity in the sentence and the entity type corresponding to this entity, and n represents the total number of entities.

[0034] Furthermore, the generation of geological entity pairs using the geological ontology model specifically includes:

[0035] Traverse the entities in the entity list E to generate all entity pairs [e h1 , T h1 , e t1 , T t1 ), (e h2 , T h2 , e t2 , T t2 ), …, (e hs , T hs , e ts , T ts )], where (e h1 , T h1 , e t1 , T t1 ) represents the first group of entity pairs, e h1 , T h1 respectively represent the head entity and the entity type corresponding to this entity, e t1 , T t1 respectively represent the tail entity and the entity type corresponding to this entity, (e h2 , T h2 , e t2 , T t2 ) represents the second group of entity pairs, e h2 , T h2 respectively represent the head entity and the entity type corresponding to this entity, e t2 , T t2 respectively represent the tail entity and the entity type corresponding to this entity, (e hs , T hs , e ts , T ts) represents the entity pair of the s-th group, e hs , T hs respectively represent the head entity and the type corresponding to this entity, e ts , T ts respectively represent the tail entity and the type corresponding to this entity, where the value of s is n×(n - 1);

[0036] For each group of entity pairs, make a judgment. For the i-th group of entity pairs (e hi , T hi , e ti , T ti ), for the head entity type T hi and the tail entity type T t , if this combination of entity types exists in the defined geological ontology model, then generate a relationship prediction task (e hi , T hi , e ti , T th , X), otherwise, delete this entity pair;

[0037] Obtain the combination of entity pairs of a series of relationships to be predicted.

[0038] In step S6, construct sentences containing geological knowledge based on the geological ontology model, use the geological relationship extraction model for relationship prediction, and implement the extraction of geological information from the input text, including:

[0039] The geological relationship extraction model is used to predict the semantic relationship of the input entity pair in the sentence. This model is mainly composed of four modules: potential relationship type retrieval, relationship type and entity type fusion, entity pair vector expression, and relationship classification;

[0040] The potential relationship type retrieval module, for a given input (e h , T h , e t , T t , X), judge the relationship between the head entity e h and the tail entity e t in the sentence X. According to the head entity type T h and the tail entity type T t , retrieve the relationship types {pr1, pr2,..., pr m} that meet the conditions from the geological ontology model, where pr1 represents the first relationship type that meets the conditions, pr2 represents the second relationship type that meets the conditions, and pr m represents the m-th relationship type that meets the conditions;

[0041] Relationship type and entity type fusion module, potential relationship types and entity types are merged into sentence X. The entity types are separated by "#" and concatenated in front of the input sentence, in the form of "#pr1##pr2#…#pr m #", and the entity type information is fused near the entity, in the form of "<S:T h >e h < / S:T h >", and "<O:T t >e t < / O:T t >", to obtain the processed sentence where S represents the label of the head entity and O represents the label of the tail entity;

[0042] Entity pair vector representation module, using a pre-trained language model such as RobERTa to obtain the vector encoding e of the i-th character in the sentence i ; create the vector representation of the entity pair as where represents the vector encoding of <S:T h >, represents the vector encoding of <O:T t >;

[0043] Relationship classification module, input h r (e h ,e t ) into a feed-forward neural network, and calculate the probability that there is no relationship or a certain relationship type between the head entity e h and the tail entity e t , so as to obtain the relationship prediction result;

[0044] According to the relationship extraction model, the relationship type prediction r of the input (e h ,T h ,e t ,T t ,X) is obtained, and a set of structured text information (e h ,T h ,r,e t ,T t ) is obtained, and entity and relationship extraction are performed on all input sentences to complete the geological information extraction of the input text.

[0045] Another object of the present invention is to provide a text information extraction system constrained by geological knowledge, which is used to regulate the text information extraction method constrained by geological knowledge. The system includes:

[0046] The geological text data acquisition and processing module is used to determine the data source for information extraction, identify and extract chapters, delete irrelevant content in the text data, and divide the remaining paragraphs into sentence-level geological text data;

[0047] The geological ontology model construction module is used to pre-define entity types and relationship types, traverse to determine whether a specified relationship type exists in geological sentences, and then obtain a geological ontology model;

[0048] The entity relationship annotation module is used to annotate sentence-level geological text data according to the geological ontology model, and through two rounds of iterative work of annotation, inspection and revision, a geological entity relationship data set is obtained;

[0049] The geological entity recognition module is used to construct a span-based named entity recognition model, and the span-based named entity recognition model is used to identify defined geological entities from input sentences;

[0050] The geological entity pair generation module is used to generate geological entity pairs by using the geological ontology model;

[0051] The relationship prediction module is used to construct sentences containing geological knowledge based on the geological ontology model, and use the relationship extraction model to predict the semantic relationship of the input entity pair in the sentence;

[0052] The geological information extraction module is used to obtain structured text information according to the semantic relationship of the input entity pair in the sentence; extract entities and relationships from the input sentence to complete the geological information extraction of the input text.

[0053] Combining all the above technical solutions, the beneficial effects of the present invention are as follows:

[0054] First, the present invention provides a text information extraction method based on geological knowledge constraints. Different from the prior art, this method realizes efficient and accurate extraction from geological text to geological entity relationship data set by integrating the geological ontology model. By introducing professional geological knowledge into the information extraction process and comprehensively using the geological ontology model in links such as named entity recognition and relationship extraction, the accuracy and reliability of information extraction are improved.

[0055] The present invention provides a method for extracting geological text entities and relationships with geological knowledge constraints (GKERE). By using a geological ontology model to filter out a large number of irrelevant entities and relationships in geological data, unnecessary computational efforts are reduced, the processing speed is enhanced, and the accuracy of relationship recognition is significantly improved. Additionally, according to the types of entity pairs, the method retrieves potential relationship types from the ontology model and incorporates this relationship information into the model input. This process not only optimizes the extraction accuracy of entity relationships but also fully exploits the value of geological domain knowledge, enabling the relationship extraction to more accurately identify complex relationships unique to the geological domain and achieving efficient automated information extraction.

[0056] Second, the precision, recall, and F1 score of the text information extraction model with geological knowledge constraints (GKERE) proposed by the present invention are 75.37%, 76.28%, and 75.82% respectively, which are superior to several other information extraction models. The experimental results show that compared with the baseline model, the model of the present invention has better performance in geological entity and relationship extraction.

[0057] The present invention introduces a geological ontology model as additional information to constrain the extraction of geological entities and relationships. To verify the effectiveness of the geological ontology model, the module involving the geological ontology model was removed from the proposed model, and the resulting model is called the backbone model. The precision of this model on the test set is 67.01%, the recall is 62.56%, and the F1 score is 64.71%, indicating that the introduction of the geological ontology model has brought a performance improvement of more than 10 percentage points, highlighting the effectiveness of the geological ontology model for geological information extraction.

[0058] The present invention also calculated the number of entity pairs generated before and after adding the geological ontology model and the calculation speed. The backbone model generated 60,678 pairs of entities by traversing all entities in the training set, and the training speed was 0.22 sentences per second. After adding the geological ontology model module, the number of entity pairs decreased to 38,195, and the training speed increased to 0.35 sentences per second. The geological ontology model module eliminated 37% of the redundant entity pairs during the training process, thus increasing the training speed by approximately 57%. On the test set, the backbone model generated 17,154 pairs of entity pairs, and the inference speed was 1.63 sentences per second. After adding the geological ontology model module, 10,895 pairs of entity pairs were generated, and the inference speed was 2.47 sentences per second. During the inference process, the geological ontology model module eliminated 36% of the redundant entity pairs, increasing the inference speed by approximately 51%. The experimental results show that the geological ontology model of the present invention effectively filters out redundant entity pairs, reduces the computational overhead, and thus improves the accuracy and efficiency of the model.

[0059] Third, as a positive effect of the present invention, it is also reflected in the following important aspects:

[0060] Geological exploration and prospecting: This method can automatically extract important information from geological exploration reports, such as deposit type, scale, grade, ore-controlling structures, geophysics, geochemistry, remote sensing, etc., thereby improving the efficiency and accuracy of prospecting prediction and significantly reducing the cost of traditional manual analysis.

[0061] Data management of mining enterprises: Through this technology, mining companies can quickly build and update geological databases from geological document materials, improve data utilization rate, and provide data support for mineral resource assessment and investment decision-making.

[0062] Government supervision and mineral resource assessment: Government agencies can use this method to quickly analyze geological data, improve the supervision efficiency of mineral resource development, and conduct mineral resource assessment to promote the development of the mining industry.

[0063] Academic research and geological big data analysis: This technology can be widely used in the construction of knowledge graphs, intelligent question-answering systems and scientific research in the geological field, accelerating the collation and dissemination of geological knowledge.

[0064] Since this technology can be directly applied to enterprises, scientific research institutions and government departments, it has obvious market competitiveness. In the future, it can form industry standards in the field of geological information services and bring huge commercial value.

[0065] Fourth, the present invention breaks through the limitations of traditional machine learning models in modeling geological relationships. By combining geological knowledge ontology constraints and deep learning, it effectively solves the problems of insufficient accuracy in information extraction and extraction results not conforming to geological cognition for a long time, enabling geological text data to be structured and utilized more efficiently.

[0066] The prior art overly relies on general natural language processing methods and lacks the support of geological professional knowledge. Most existing information extraction models are based on natural language processing technologies in general fields, such as sequence labeling (BERT-BiLSTM-CRF), word frequency statistics, rule matching, etc. However, geological texts have strong domain characteristics, involving a large number of professional terms and complex geological entity relationships, making the effects of general models often poor. The present invention adopts a method combining geological knowledge constraints and deep learning, introducing the geological ontology model into the information extraction process, which can ensure that the extraction results conform to geological logic and overcome the limitations of traditional general natural language processing methods.

[0067] The information extraction model has insufficient understanding ability of geological relationships and cannot accurately identify complex geological relationships. Traditional information extraction methods often rely on statistical learning or simple pattern matching, making it difficult to capture complex geological entity relationships. For example, the occurrence relationship between ore deposits and strata, the symbiotic relationship between ore deposits and minerals, and the causal relationship between geological structures and mineralization. These relationships are often ignored or simplified to simple entity co-occurrence relationships in traditional methods, resulting in a decrease in the accuracy and reliability of information extraction. The present invention uses a relationship extraction model constrained by geological knowledge, introducing geological ontology knowledge constraints in the relationship prediction process to ensure that the model can understand and correctly identify professional relationships in the geological field.

[0068] The present invention effectively overcomes the technical biases in traditional geological information extraction techniques by introducing a geological knowledge constraint mechanism and combining deep learning and pre-trained language models. The main manifestations are as follows: It overcomes the dependence on general natural language processing methods and improves the professionalism of geological information extraction; it overcomes the problem of insufficient understanding ability of traditional information extraction methods for complex geological relationships. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The accompanying drawings herein are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure;

[0070] Figure 1 It is a flowchart of a text information extraction method with geological knowledge constraints provided by an embodiment of the present invention;

[0071] Figure 2 It is a structural diagram of a span-based named entity recognition model provided by an embodiment of the present invention;

[0072] Figure 3 It is a structural diagram of a relationship extraction model based on a geological ontology model provided by an embodiment of the present invention;

[0073] Figure 4 It is an example diagram integrating entity types and potential relationship information provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0074] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following detailed description of the specific embodiments of the present invention will be given in conjunction with the accompanying drawings. Many specific details are set forth in the following description to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0075] The innovation points of the present invention are as follows:

[0076] (1) Develop a text information extraction method based on geological knowledge constraints;

[0077] The technical solution of the present invention effectively constrains and filters entities and relationships in geological texts through a geological ontology model. Traditional geological information extraction methods often fail to fully consider the professional knowledge in the geological field. However, the present invention solves the following problems of traditional methods by introducing domain-specific knowledge into the information extraction process: the problem of redundant information extraction. The ontology model can help filter out redundant entity pairs that are irrelevant to the actual meaning of geological texts, thereby reducing unnecessary computational effort and improving the processing speed of the model; the improvement of relationship accuracy. Through the knowledge constraints of the geological ontology model, the extracted entity relationships are more in line with the actual background of the geological field, significantly improving the accuracy of relationship recognition.

[0078] (2) Propose a retrieval mechanism for relationship types based on entity pair types;

[0079] The present invention innovatively proposes a retrieval mechanism for relationship types based on entity pair types. This mechanism can automatically retrieve possible relationship types from the geological ontology model according to the types of entity pairs identified in the input text and introduce this relationship information into the input of the model. The innovation of this process lies in: accurately matching the unique relationships in the geological field. By introducing the geological professional knowledge in the ontology model, it ensures that the unique geological relationships such as ore deposits, structures, and metallogenic processes involved in the relationship extraction process can be correctly identified; improving the efficiency and accuracy of information extraction. Traditional methods often ignore the constraints and limitations of relationship types during relationship extraction, while the present invention greatly improves the information extraction accuracy and working efficiency of the model through in-depth analysis of entity pair types.

[0080] (3) Efficient entity relationship extraction and knowledge-driven automated information processing;

[0081] The solution of the present invention also realizes efficient automated information extraction, especially in the application in the geological field. By combining the geological ontology model with modern deep learning technologies, in a way that combines knowledge-driven and data-driven: adopting a span-based named entity recognition method to accurately extract geological entities, and through combination with the geological ontology model, generating entity pairs that conform to geological cognition; based on the knowledge information in the geological ontology model, performing relationship recognition on the extracted entity pairs, improving the ability to extract complex relationships; this method effectively improves the processing efficiency of geological texts through an automated process, especially in the processing of massive geological data, greatly reducing the need for manual intervention.

[0082] (4) A composite modeling method that combines a pre-trained language model and a geological ontology model;

[0083] The present invention combines a pre-trained language model (such as RobERTa) with a geological ontology model for composite modeling: leveraging the powerful context understanding ability of the pre-trained language model to help the model accurately understand the semantics in geological texts; by embedding the geological ontology model into the model, precise extraction based on geological professional knowledge is achieved, effectively improving the accuracy and effectiveness in the information extraction process.

[0084] The core innovation of the technical solution of the present invention lies in achieving efficient and accurate information extraction from geological texts by combining a geological ontology model with deep learning technology. Especially in the process of entity relationship extraction, a knowledge constraint mechanism is innovatively introduced, greatly improving the extraction accuracy and efficiency. This innovation not only provides an intelligent solution for geological information processing but also lays a foundation for the structuring of geological data and knowledge sharing, with broad application prospects and social value.

[0085] Example 1, as Figure 1 shown, the method for extracting text information with geological knowledge constraints provided by the embodiment of the present invention includes the following steps:

[0086] S1, preprocessing of geological text materials, using a Python program to determine the data source for information extraction, deleting irrelevant content in the text data, and obtaining sentence-level geological text data after processing, realizing the acquisition and processing of geological text materials;

[0087] The acquisition and processing of geological text materials include:

[0088] Using a Python program to extract and separate text content, graphics, and tables from geological text materials, and selecting the text content as the data source for information extraction;

[0089] Applying a set of pattern matching rules to identify and extract chapters, and deleting irrelevant content in the text data;

[0090] By identifying the end characters of sentences, the remaining paragraphs are segmented into sentence-level geological text data.

[0091] Among them, the chapters include titles, abstracts, chart and table titles, acknowledgments, and references;

[0092] The irrelevant content in the text data includes titles, author information, titles, acknowledgments, and references.

[0093] In this invention, 68 academic papers and 4 mineral exploration reports related to Xinjiang granite pegmatite-type lithium ore were collected. These collected documents are usually in PDF format and need to be processed. First, a Python program is used to extract and separate the text content, graphics, and tables from the PDF, and the text content is selected as the data source for information extraction. Subsequently, a set of pattern matching rules is applied to identify and extract sections such as titles, abstracts, chart and table titles, acknowledgments, and references, and the irrelevant content in the text data, including titles, author information, titles, acknowledgments, and references, is deleted. Finally, by identifying sentence-ending characters such as full stops, question marks, and exclamation marks, the remaining paragraphs are segmented into sentence-level text for later use.

[0094] S2. Pre-define entity types and relationship types, and use the sentence-level geological text data to restrict the entity types and relationships in geological sentences; traverse all pairs of entity types to implement the construction of the geological ontology model;

[0095] The construction of the geological ontology model includes:

[0096] Pre-define entity types and relationship types, and extract ore-forming conditions and ore-forming anomaly information from the sentence-level geological text data; the entity types include tectonic units, ore-forming units, faults, tectonic deformations, ore deposits, strata, alteration, metamorphic rocks, sedimentary rocks, magmatic rocks, ore bodies, minerals, geophysical characteristics, chemical element characteristics, remote sensing markers, spatial positions, time, genetic types, grades, scales, ore species, and mineral zoning;

[0097] The relationship types include tectonic actions, ore-forming actions, sedimentary actions, metamorphic actions, magmatic actions, wall-rock alteration actions, control actions, limiting actions, indicating actions, ore-forming carriers or material sources, occurrence relationships, intrusion relationships, inclusion relationships, attribute relationships, spatial relationships, and time relationships;

[0098] Restrict the entity types and relationships in the sentence-level geological text data; traverse all pairs of entity types to determine whether a specified relationship type exists in each sentence, and obtain the geological ontology model; the geological ontology model consists of knowledge rules in the form of <head entity type, relationship type, tail entity type>, and this geological ontology model represents geological knowledge and is used to guide geological text annotation and information extraction.

[0099] In order to extract metallogenic conditions and metallogenic anomaly information from geological texts, taking the granite pegmatite type lithium ore as an example, 22 entity types and 16 relationship types are predefined. The entity types include tectonic element (TECU), metallogenic unit (MENU), fault (FAUL), tectonic deformation (STRU), ore deposit (DEPO), stratum (STRA), alteration (ALTE), metamorphic rock (META), sedimentary rock (SEDI), magmatic rock (MAG), ore body (OREB), mineral (MINE), geophysical feature (GPC), chemical element feature (GCC), remote sensing signature (RSC), spatial position (SPAP), time (TIME), genetic type (GENE), grade (GRA), scale (SCAL), ore type (CATE), mineral zoning (ZONA). The relationship types include: tectonic process, metallogenic process, sedimentation process, metamorphism process, magmatism process, wall rock alteration process, control effect, delimitation effect, indication effect, metallogenic carrier (or material source), occurrence relationship, intrusion relationship, inclusion relationship, attribute relationship, spatial relationship and time relationship.

[0100] To ensure that the relationships between entities conform to the understanding of granite pegmatite type lithium ore, ore deposit geologists have imposed restrictions on the possible entity types and relationships in geological sentences based on their professional knowledge. All pairs of entity types are traversed to determine whether a specified relationship type exists in each sentence. This process has produced a geological ontology model, which consists of 184 knowledge rules in the form of <head entity type, relationship type, tail entity type>, as shown in Table 1. For example, in a granite pegmatite type lithium ore deposit, magmatic rock is the metallogenic carrier or material source of the ore deposit, and the corresponding rule is "<ore deposit, metallogenic carrier (or material source), magmatic rock>". The same pair of entity types can have multiple relationship types such as <magmatic rock, intrusion relationship, magmatic rock> and <magmatic rock, inclusion relationship, magmatic rock>. This geological ontology model represents geological knowledge and can be used to guide geological text annotation and information extraction.

[0101] Table 1 Partial Geological Knowledge Rules

[0102]

[0103] S3. Annotate the sentence-level geological text data according to the geological ontology model to obtain a geological entity relationship data set, which is used for the training of subsequent named entity recognition models and relationship extraction models;

[0104] Annotating the sentence-level geological text data according to the geological ontology model includes:

[0105] The constructed geological ontology model is used to automatically check the marked entity relationships. The entity relationships that do not conform to the rules are automatically marked and modified again. Through two rounds of iterative work of annotation, inspection, and revision, sentences containing relationships are obtained, and a geological entity relationship dataset is formed.

[0106] The obtained geological entity relationship dataset is the basic dataset for training the named entity recognition model and the relationship extraction model. By annotating the entity relationships in geological texts, the constructed dataset provides accurate and comprehensive annotation samples for model training, ensuring that the training and optimization of the model in subsequent steps can accurately reflect the professional knowledge in the geological field.

[0107] The method of manual annotation is used to annotate the entity relationships in the collected text data. The constructed geological ontology model is used to automatically check the marked entity relationships, and those entity relationships that do not conform to the rules are automatically marked and modified again. After two rounds of iterative work of annotation, inspection, and revision, 3112 sentences containing relationships are obtained, forming a geological entity relationship dataset. As shown in Table 2, this dataset contains 3112 sentences, 16486 entities, and 14545 entity relationships. On average, each sentence contains 5.30 entities and 4.67 relationships.

[0108] Table 2 Statistical table of relationship types in the geological entity relationship dataset

[0109]

[0110]

[0111] S4. Build a geological named entity recognition model based on spans, use the geological entity relationship dataset to train the model, establish a geological named entity recognition model, and realize the recognition of geological entities in the input geological text;

[0112] The named entity recognition model based on spans is used to recognize 22 defined geological entities from the input sentence. The model results are as Figure 2 shown, and it mainly consists of three modules: text vector encoding, span generation and encoding, and span classification.

[0113] S41: Text vector encoding module. When the input sentence is X = [x1, x2, …, x n , where x1 represents the first character, x2 represents the second character, x n represents the nth character, [·] represents a list, and the longest entity length is L characters. Use a pre-trained language model such as RobERTa to generate the vector encoding e i of the ith character x i in the input sentence;

[0114] S42: Span generation and encoding module, which traverses all characters of the input sentence X to generate all potential spans S = {s 1,1 , s 1,2 , …, s 1,L , s 2,2 …, s n,n}, where s 1,1 = [x1], s 1,2 = [x1, x2], s 1,L = [x1, x2, …, x L , s 2,2 = [x2], s n,n = [x n , {} represents a set; for the given i-th span s i , the vector representation is h e (s i ) = [e STRAT(i) ; e END(i) ; w(s i )], where e STRAT(i) represents the vector encoding of the starting character of the span s i , e END(i) represents the vector encoding of the ending character of the span s i , and w(s i ) represents the length encoding of the span s i .

[0115] S43: Span classification module, which inputs h e (s i ) into a feed-forward neural network to calculate the probability that the span s i belongs to a non-entity or a certain entity type, thereby obtaining the entity prediction result.

[0116] S5. Use the geological ontology model to generate geological entity pairs;

[0117] After obtaining the entities in the input sentence, the next step is to convert the entities into entity pairs to predict the semantic relationship between two entities in the sentence, and the expression form is <head entity, entity type 1, tail entity, entity type 2, sentence>. Assume that the entity list identified in a geological sentence X is E = [(e1, T1), (e2, T2), …, (e n , T n )], where e1 and T1 represent the first entity in the sentence and the entity type corresponding to this entity respectively, e2 and T2 represent the second entity in the sentence and the entity type corresponding to this entity respectively, e n , T n represent the n-th entity in the sentence and the entity type corresponding to this entity respectively, and n represents the total number of entities.

[0118] S51: Generate all entity pairs [(e t1 , T h1 , e t1 , T t1 ), (e h2 , T h2 , e t2 , T t2 ), …, (e hs , T hs , e ts , T ts )] by traversing the entities in the entity list E. (e h1 , T h1 , e t1 , T t1 ) represents the first group of entity pairs. e h1 , T h1 represent the head entity and the type corresponding to this entity respectively. e t1 , T t1 represent the tail entity and the type corresponding to this entity respectively. (e h2 , T h2 , e t2 , T t2 ) represents the second group of entity pairs. e h2 , T h2 represent the head entity and the type corresponding to this entity respectively. e t2 , T t2 represent the tail entity and the type corresponding to this entity respectively. (e hs , T hs , e ts , T ts ) represents the s-th group of entity pairs. e hs , T hs represent the head entity and the type corresponding to this entity respectively. e ts , T ts represent the tail entity and the type corresponding to this entity respectively. Among them, the value of s is n×(n - 1);

[0119] S52: For each group of entity pairs, make a judgment. For the i-th group of entity pairs (e hi , T hi , e ti , T ti ), if the combination of the head entity type T hi and the tail entity type T t is in the defined geological ontology model, then generate a relationship prediction task (e hi , T hi , e ti , T ti , X), otherwise, delete this entity pair;

[0120] S53: Obtain combinations of entity pairs for which a series of prediction relationships are required.

[0121] S6. Construct sentences containing geological knowledge based on the geological ontology model, use the geological relation extraction model for relation prediction, and implement the extraction of geological information from the input text;

[0122] The geological relation extraction model is used to predict the semantic relationship of the input entity pair in the sentence. The model results are as Figure 3 shown, and it mainly consists of four modules: potential relation type retrieval, relation type and entity type fusion, entity pair vector expression, and relation classification.

[0123] S61: Potential relation type retrieval module. For a given input (e h , T h , e t , T t , X), judge the relationship between the head entity e h and the tail entity e t in the sentence X. According to the head entity type T h and the tail entity type T t , retrieve the relation types {pr1, pr2,..., pr m} that meet the conditions from the geological ontology model, where pr1 represents the first relation type that meets the conditions, pr2 represents the second relation type that meets the conditions, and pr m represents the mth relation type that meets the conditions;

[0124] S62: Relation type and entity type fusion module. The potential relation types and entity types are merged into the sentence X. Use "#" to separate the entity types and splice them in front of the input sentence, with the form "#pr1##pr2#...#pr m #". Integrate the entity type information near the entity, with the forms "<S:T h >e h < / S:T h >" and "<O:T t >e t < / O:T t >" to obtain the processed sentence where S represents the label of the head entity and O represents the label of the tail entity, as Figure 4 shown;

[0125] S63: Entity pair vector expression module. Use a pre-trained language model such as RobERTa to obtain the vector encoding e of the ith character in the sentence i ; Create the vector expression of the entity pair as where The vector encoding representing <S:T h > The vector encoding representing <O:T t >

[0126] S64: The relationship classification module inputs h r (e h , e t ) into a feed-forward neural network to calculate the probability that there is no relationship or a certain relationship type between the head entity e h and the tail entity e t , thus obtaining the relationship prediction result.

[0127] S7: Complete geological information extraction.

[0128] According to the relationship extraction model, for the relationship type prediction r of the input (e h , T h , e t , T t , X), a set of structured text information (e h , T h , r, e t, T t ) can be obtained, and entity and relationship extraction are performed on all input sentences to complete the geological information extraction of the input text.

[0129] Example 2. The text information extraction system with geological knowledge constraints provided by the embodiments of the present invention includes:

[0130] The geological text data acquisition and processing module is used to determine the data source of information extraction, identify and extract the irrelevant content in the text data after the chapter, and segment the remaining paragraphs into sentence-level texts;

[0131] The geological ontology model construction module is used to pre-define entity types and relationship types, and determine whether there is a specified relationship type in each sentence by traversing, thereby obtaining the geological ontology model;

[0132] The entity relationship annotation module is used to annotate the sentence-level geological text data according to the geological ontology model, and obtain the geological entity relationship data set through two rounds of iterative work of annotation, inspection, and revision;

[0133] The geological entity recognition module is used to construct a span-based named entity recognition model, and the span-based named entity recognition model is used to recognize defined geological entities from the input sentences;

[0134] The geological entity pair generation module is used to generate geological entity pairs by using the geological ontology model;

[0135] A relationship prediction module, which is used to construct sentences containing geological knowledge based on a geological ontology model and predict the semantic relationship of an input entity pair in the sentence using a relationship extraction model;

[0136] A geological information extraction module, which is used to obtain structured text information according to the semantic relationship of an input entity pair in a sentence; perform entity and relationship extraction on the input sentence to complete the geological information extraction of the input text.

[0137] To evaluate the effectiveness of the method proposed by the present invention, this model was compared with several text information extraction models, including: (1) Multi-head Selection, (2) UniRE, (3) PFN, (4) PURE, (5) PL-Maker, (6) ATG. Three conventional evaluation metrics, namely Precision, Recall, and F1-score, were used to measure the performance of the model of the present invention in geological information extraction.

[0138] The calculation formulas for the three evaluation metrics are as follows:

[0139]

[0140] Among them, TP (True positive) refers to the number of correctly predicted positive samples, FP (False positive) refers to the number of incorrectly predicted positive samples, and FN (False Negatives) refers to the number of positive samples not predicted. Precision reflects the ability of the model to avoid false positives, while Recall indicates the degree to which the model identifies true positive samples. The F1-score is a balanced metric that takes both Precision and Recall into account. The higher the F1-score, the better the overall classification performance of the model.

[0141] The geological entity relationship dataset was divided into a training set, a validation set, and a test set in a ratio of 7:1.5:1.5, consisting of 9935, 2123, and 2487 triples respectively. The training set was used to train the model, the validation set was used for parameter tuning, and the test set was used for model performance evaluation. Table 3 shows the experimental results of 7 models on the constructed geological dataset. The accuracy, recall, and F1-score of the model proposed by the present invention are 75.37%, 76.28%, and 75.82% respectively, which are better than several other models. These results indicate that the model has better performance in geological entity and relationship extraction compared with the baseline model.

[0142] Table 3 Results of various entity and relationship extraction models on the test set

[0143]

[0144]

[0145] In the present invention, a geological ontology model is introduced as additional information to constrain the extraction of geological entities and relationships. To verify the effectiveness of the geological ontology model, the module related to the geological ontology model was removed from the proposed model. The model after removing the geological ontology model is called the backbone model. The precision of this model on the test set is 67.01%, the recall rate is 62.56%, and the F1 score is 64.71%. This means that the introduction of the geological ontology model has brought a performance improvement of more than 10 percentage points, highlighting the effectiveness of the geological ontology model for geological information extraction.

[0146] To further analyze the role of the filtering component, the present invention calculated the number of entity pairs generated before and after adding the geological ontology model and the calculation speed, as shown in Table 4. The backbone model generated 60,678 pairs of entities by traversing all entities in the training set, and the training speed was 0.22 sentences per second. After adding the geological ontology model module, the number of entity pairs decreased to 38,195, and the training speed increased to 0.35 sentences per second. The geological ontology model module eliminated 37% of the redundant entity pairs during the training process, thus increasing the training speed by approximately 57%. On the test set, the backbone model generated 17,154 pairs of entity pairs, and the inference speed was 1.63 sentences per second. After adding the geological ontology model module, 10,895 pairs of entity pairs were generated, and the inference speed was 2.47 sentences per second. During the inference process, the geological ontology model module eliminated 36% of the redundant entity pairs, increasing the inference speed by approximately 51%. These results show that the geological ontology model effectively filters redundant entity pairs, reduces the computational overhead, and thus improves the accuracy and efficiency of the model.

[0147] Table 4 Influence of the introduction of the geological ontology model on entity pair generation and calculation speed

[0148]

[0149] As described above, it is only a relatively preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for extracting text information constrained by geological knowledge, characterized in that The method includes the following steps: S1. Pretreatment of geological text materials: Use a Python program to determine the data source for information extraction, delete the irrelevant content in the text data, and obtain sentence-level geological text data after processing, so as to realize the acquisition and processing of geological text materials; S2. Pre-define entity types and relationship types, and use the sentence-level geological text data to restrict the entity types and relationships in geological sentences; Traverse all pairs of entity types to realize the construction of a geological ontology model; S3. Annotate the sentence-level geological text data according to the geological ontology model to obtain a geological entity relationship data set, which is used for the training of subsequent named entity recognition models and relationship extraction models; S4. Construct a span-based geological named entity recognition model, use the geological entity relationship data set to train the model, establish a geological named entity recognition model, and realize the recognition of geological entities in the input geological text; S5. Generate geological entity pairs using the geological ontology model; S6. Construct sentences containing geological knowledge based on the geological ontology model, use the geological relationship extraction model to predict relationships, and realize the extraction of geological information from the input text.

2. The method for extracting text information constrained by geological knowledge according to claim 1, wherein In step S1, the acquisition and processing of geological text materials include: Use a Python program to extract and separate text content, graphics, and tables from geological text materials, and select the text content as the data source for information extraction; Apply a set of pattern matching rules to identify and extract chapters, and delete the irrelevant content in the text data; By identifying the sentence-ending characters, split the remaining paragraphs into sentence-level geological text data.

3. The method for extracting text information constrained by geological knowledge according to claim 2, wherein The chapters include titles, abstracts, chart and table titles, acknowledgments, and references; The irrelevant content in the text data includes titles, author information, titles, acknowledgments, and references.

4. The method for extracting text information constrained by geological knowledge according to claim 1, wherein In step S2, the construction of the geological ontology model includes: Pre-define entity types and relationship types, and extract metallogenic conditions and metallogenic anomaly information from the sentence-level geological text data; The entity types include tectonic units, metallogenic units, faults, tectonic deformations, ore deposits, strata, alterations, metamorphic rocks, sedimentary rocks, magmatic rocks, ore bodies, minerals, geophysical characteristics, chemical element characteristics, remote sensing markers, spatial positions, time, genetic types, grades, scales, ore types, and mineral zoning; The relationship types include tectonic action, metallogenic action, sedimentary action, metamorphic action, magmatic action, wall rock alteration action, control action, limiting action, indicating action, metallogenic carrier or material source, occurrence relationship, intrusion relationship, inclusion relationship, attribute relationship, spatial relationship, and time relationship; Restrict the entity types and relationships in the sentence-level geological text data; Traverse all pairs of entity types to determine whether a specified relationship type exists in each sentence, and obtain a geological ontology model; The geological ontology model consists of knowledge rules in the form of <head entity type, relationship type, tail entity type>, and this geological ontology model represents geological knowledge and is used to guide geological text annotation and information extraction.

5. The method for extracting text information constrained by geological knowledge according to claim 1, wherein In step S3, annotating the sentence-level geological text data according to the geological ontology model includes: The constructed geological ontology model is used to automatically check the marked entity relationships. The entity relationships that do not conform to the rules are automatically marked and modified again. Through two rounds of iterative work of annotation, inspection and revision, sentences containing relationships are obtained, and a geological entity relationship dataset is formed.

6. The method for extracting text information constrained by geological knowledge according to claim 1, wherein In step S4, a span-based geological named entity recognition model is constructed, and the model is trained using the geological entity relationship dataset, including: The span-based geological named entity recognition model is used to identify defined geological entities from the input sentence. The model mainly consists of three modules: text vector encoding, span generation and encoding, and span classification; Text vector encoding module. When the input sentence is X = [x1, x2, …, x n , where x1 represents the 1st character, x2 represents the 2nd character, x n represents the nth character, [·] represents a list, and the longest entity length is L characters. Use a pre-trained language model such as RobERTa to generate the vector encoding e i of the ith character x i in the input sentence; Span generation and encoding module, which traverses all characters of the input sentence X to generate all potential spans S = {s 1,1 , s 1,2 , …, s 1,L , s 2,2 …, s n,n}, where s 1,1 = [x1], s 1,2 = [x1, x2], s 1,L = [x1, x2, …, x L , s 2,2 = [x2], s n,n = [x n , {·} represents a set; for the given i-th span s i , the vector representation is h e (s i ) = [e STRAT(i) ; e END(i) ; w(s i )], where e STRAT(i) represents the vector encoding of the starting character of the span s i , e END(i) represents the vector encoding of the ending character of the span s i , and w(s i ) represents the length encoding of the span s i . Span classification module, input h e (s i ) into the feed-forward neural network to calculate the probability that the span s i belongs to a non-entity or a certain entity type, so as to obtain the entity prediction result.

7. The method for extracting text information constrained by geological knowledge according to claim 1, wherein In step S5, the geological ontology model is used to generate geological entity pairs, including: After obtaining the entities in the input sentence, the entities are transformed into entity pairs to predict the semantic relationship between the two entities in the sentence, and the expression form is <head entity, entity type 1, tail entity, entity type 2, sentence>; when the entity list identified in the geological sentence X is E = [(e1, T1), (e2, T2), …, (e n , T n )], where e1 and T1 respectively represent the first entity in the sentence and the entity type corresponding to this entity, e2 and T2 respectively represent the second entity in the sentence and the entity type corresponding to this entity, e n , T n respectively represent the nth entity in the sentence and the entity type corresponding to this entity, and n represents the total number of entities.

8. The method for extracting text information constrained by geological knowledge according to claim 7, wherein The specific process of using the geological ontology model to generate geological entity pairs includes: Generate all entity pairs [(e h1 , T h1 , e t1 , T t1 ), (e h2 , T h2 , e t2 , T t2 ), …, (e hs , T hs , e ts , T ts )] by traversing the entities in the entity list E, where (e h1 , T h1 , e t1 , T t1 ) represents the first group of entity pairs, e h1 , T h1 represent the head entity and the type corresponding to this entity respectively, e t1 , T t1 represent the tail entity and the type corresponding to this entity respectively, (e h2 , T h2 , e t2 , T t2 ) represents the second group of entity pairs, e h2 , T h2 represent the head entity and the type corresponding to this entity respectively, e t2 , T t2 represent the tail entity and the type corresponding to this entity respectively, (e hs , T hs , e ts , T ts ) represents the s-th group of entity pairs, e hs , T hs represent the head entity and the type corresponding to this entity respectively, e ts , T ts represent the tail entity and the type corresponding to this entity respectively, where the value of s is n×(n - 1); For each group of entity pairs, for the i-th group of entity pairs (e hi , T hi , e ti , T ti ), if the head entity type T hi and the tail entity type T t in this combination of entity types exist in the defined geological ontology model, then generate a relationship prediction task (e hi , T hi , e ti , T ti , X), otherwise, delete this entity pair; Obtain a series of entity pair combinations with relationships to be predicted.

9. The method for extracting text information constrained by geological knowledge according to claim 1, characterized in that In step S6, sentences containing geological knowledge are constructed based on the geological ontology model, and a relationship extraction model is used for relationship prediction and geological information extraction of the input text, including: The geological relationship extraction model is used to predict the semantic relationship of the input entity pair in the sentence. The model mainly consists of four modules: potential relationship type retrieval, relationship type and entity type fusion, entity pair vector expression, and relationship classification; Potential relationship type retrieval module, for a given input (e h , T h , e t , T t , X), determine the relationship between the head entity e h and the tail entity e t in the sentence X. According to the head entity type T h and the tail entity type T t retrieve the relationship types {pr1, pr2, …, pr m} that meet the conditions from the geological ontology model, where pr1 represents the first relationship type that meets the conditions, pr2 represents the second relationship type that meets the conditions, and pr m represents the mth relationship type that meets the conditions; Relationship type and entity type fusion module. Potential relationship types and entity types are merged into sentence X. The entity types are separated by "#" and concatenated in front of the input sentence, with the representation form of "#pr1##pr2#…#pr m #", and the entity type information is fused near the entity, with the representation form of "<S:T h >e h < / S:T h >” and "<O:T t >e t < / O:T t >”, and the processed sentence is obtained Among them, S represents the label of the head entity, and O represents the label of the tail entity; Entity pair vector representation module, using a pre-trained language model such as RobERTa to obtain the vector encoding e of the i-th character in the sentence ; Create the vector representation of the entity pair as i ; where represents the vector encoding of <S:T , h and represents the vector encoding of <O:T t > Relationship classification module, input h r (e h , e t ) into the feedforward neural network, and calculate the probability that there is no relationship or there is a certain relationship type between the head entity e h and the tail entity e t to obtain the relationship prediction result; Predict the relationship type r of the input (e h , T h , e t , T t , X), and obtain a set of structured text information (e h , T h , r, e t , T t ). Perform entity and relationship extraction on all input sentences to complete the geological information extraction of the input text.

10. A text information extraction system constrained by geological knowledge, characterized in that, This system is used to regulate the text information extraction method constrained by geological knowledge described in any one of claims 1-9. This system includes: A geological text data acquisition and processing module, which is used to determine the data source of information extraction, identify and extract irrelevant content in the text data after chapters, and split the remaining paragraphs into sentence-level geological text data; A geological ontology model construction module, which is used to pre-define entity types and relationship types, and determine whether a specified relationship type exists in the geological sentence by traversing, so as to obtain a geological ontology model; An entity relationship annotation module, which is used to annotate the sentence-level geological text data according to the geological ontology model. Through two rounds of iterative work of annotation, inspection and revision, a geological entity relationship dataset is obtained; A geological entity recognition module, which is used to construct a span-based named entity recognition model, and the span-based named entity recognition model is used to identify defined geological entities from the input sentence; A geological entity pair generation module, which is used to generate geological entity pairs using the geological ontology model; A relationship prediction module, which is used to construct sentences containing geological knowledge based on the geological ontology model, and use a relationship extraction model to predict the semantic relationship of the input entity pair in the sentence; A geological information extraction module, which is used to obtain structured text information according to the semantic relationship of the input entity pair in the sentence; perform entity and relationship extraction on the input sentence to complete the geological information extraction of the input text.

Images