Automatic extraction method and system for photo-assisted electro-catalysis literature information

Abstract

The invention discloses an automatic extraction method and system for photo-assisted electro-catalysis literature information. The method comprises the steps that photo-assisted electro-catalysis related literatures are obtained and analyzed to obtain clean texts; identifying a material entity and a reaction type in the clean text; based on the reaction type, extracting a performance parameter from the clean text; acquiring a corresponding relation between the material entity and the performance parameter according to the material entity and the performance parameter; performing multi-dimensional standardization processing on the corresponding relation to obtain standardized data; and storing the standardized data in a structured database to complete automatic extraction of the literature. According to the method, the problems of scattered information, non-uniform description, low manual extraction efficiency, lack of a standardized data processing mechanism and the like of the existing photo-assisted electro-catalysis literature can be solved.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of data processing and materials science, and in particular relates to an automatic extraction method and system for photo-assisted electrocatalysis literature information. Background Technology

[0002] With the continued growth of global energy demand and the increasing severity of environmental problems, the importance of various clean energy conversion technologies is becoming increasingly prominent. In energy devices such as metal-air batteries, fuel cells, and water electrolyzers, photo-assisted electrocatalysis has become one of the key technical routes for improving the overall performance of the system because it can reduce reaction energy consumption and improve catalytic efficiency under light conditions.

[0003] In materials development, photo-assisted electrocatalysis systems typically involve complex material compositions, interfacial structures, and variable testing conditions. Different publications describe catalyst structures, electrochemical parameters, and illumination conditions in significantly different ways, lacking a unified standard, making direct comparisons between research results difficult. Furthermore, current techniques often rely on manual extraction of performance parameters from literature, such as Tafel slope, overpotential, light / dark voltage, and current density. This approach is not only inefficient but also prone to omissions or inaccurate extraction, making it unsuitable for supporting structure-property relationship analysis and large-scale data-driven material screening.

[0004] Although some literature analysis tools can parse chemical texts, they are mostly designed for general chemical data and lack the ability to identify specific features in photo-assisted electrocatalysis systems, such as composite materials, parallel parameter expressions, and illumination condition annotations. At the same time, existing tools lack standardized processing mechanisms such as unit unification and illumination condition normalization for material performance data, making it difficult to meet the requirements of consistency and comparability of high-quality data in photo-assisted electrocatalysis research.

[0005] Therefore, at present, there is a need for a technical solution that can automatically analyze literature, identify catalyst structures, extract performance parameters, label illumination conditions, and standardize data in the field of photo-assisted electrocatalysis, so as to build a structured, searchable, and comparable large-scale professional database to support material screening, structure-activity analysis, and data-driven catalyst design. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes an automatic extraction method and system for information from photo-assisted electrocatalysis literature. This method can automatically identify, standardize, and output structured information such as material structure, performance parameters, and illumination conditions from the literature, thereby improving the efficiency and accuracy of data acquisition and building a high-quality data foundation with strong comparability that can be used for material analysis and database construction.

[0007] To achieve the above objectives, the present invention provides an automatic extraction method for literature information on photo-assisted electrocatalysis, comprising:

[0008] Relevant literature on photo-assisted electrocatalysis was obtained and parsed to produce clean text.

[0009] Identify the material entities and reaction types in the clean text;

[0010] Based on the reaction type, performance parameters are extracted from the clean text;

[0011] Based on the material entities and the performance parameters, when a parallel description structure is detected, the parallel result reorganization module is connected to the performance parameter extraction module and the material entity identification module. Its input is the set of material entities output by the material entity identification module and the set of performance parameters (including parameter type, value and reaction type label) output by the performance parameter extraction module. Its output is a one-to-one correspondence between material entities and performance parameters. The parallel result reorganization includes: (1) splitting the material entities in the parallel segments to obtain a list of material entities; (2) splitting the performance parameter values ​​in the parallel segments to obtain a list of parameter values; (3) under the constraint of the reaction type identification result, redistributing and aligning the split material entities and performance parameter values ​​to construct a one-to-one correspondence between material entities and performance parameters. When no parallel description structure is detected, the material entities and performance parameters directly establish a correspondence according to the default order.

[0012] The correspondence is standardized in multiple dimensions to obtain standardized data;

[0013] The standardized data is stored in a structured database to complete the automatic extraction of documents.

[0014] Optionally, obtain and parse clean texts of relevant literature on photo-assisted electrocatalysis, including:

[0015] Obtain document files in PDF, HTML, XML, or LaTeX formats from publicly available literature databases, institutional databases, or other accessible literature collections;

[0016] The document file is parsed to obtain the text to be processed, which contains structural information;

[0017] The text to be processed is formatted and denoised to obtain the clean text.

[0018] Optionally, identifying material entities in the clean text includes:

[0019] The clean text is matched based on a pre-built material dictionary to identify elemental materials, compounds, composite materials, heterojunction materials, supported materials, and doped materials.

[0020] Regular rules are used to identify chemical formula expressions, complex structure expressions, doped expressions, carrier expressions, and multi-component combined table expressions in clean text;

[0021] The clean text was annotated using a sequence labeling model;

[0022] The dictionary matching results, regular rule recognition results, and model annotation results are fused together, and the material entity is output through a voting or confidence weighting strategy.

[0023] Optionally, identifying the reaction type in the clean text includes:

[0024] Extract the method section, test section, chart title, and text near the reaction formula from the clean text as reaction-related sentence segments;

[0025] Identify reaction trigger words in the reaction-related sentence segments, and determine the reaction type R based on the reaction trigger words. When the reaction type cannot be identified, the reaction type R is marked as unknown.

[0026] Optionally, extracting performance parameters from the clean text based on the reaction type includes:

[0027] Based on the reaction type R, the corresponding parameter trigger dictionary is invoked to locate the parameter name;

[0028] Extract numerical values, units, and conditional information from the sentence containing the parameter name or adjacent sentences. When the unit or conditional information appears across sentences, complete it by backtracking across sentences and associating it with the nearest neighbor.

[0029] The output includes performance parameters with parameter type, value, unit, and conditions.

[0030] Optionally, obtaining the correspondence between the material entity and the performance parameters based on the material entity and the performance parameters includes:

[0031] Parallel sentences are detected in the clean text. When parallel sentences exist, the material entities in the detected parallel sentence segments are split into ordered lists, and the performance parameters are split into ordered numerical lists. Under the constraint of the reaction type R, the ordered lists and the ordered numerical lists are matched one by one in sequence to obtain the correspondence between the material entities and the performance parameters. When the number of material entities is inconsistent with the number of performance parameters, a default padding or nearest neighbor backoff strategy is used to complete the matching.

[0032] When no parallel syntax exists, the material entity and performance parameters are directly established in a preset order.

[0033] Optionally, the correspondence can be standardized in multiple dimensions to obtain standardized data, including:

[0034] The correspondence is standardized by unit standardization, reference electrode conversion, and illumination condition normalization to obtain the standardized data.

[0035] Optionally, storing the standardized data in a structured database to complete the automatic extraction of documents includes:

[0036] The standardized data is validated and anomaly detected to obtain a structured data set of target quality.

[0037] The target quality structured data set is stored in a structured database to complete the automatic extraction of documents.

[0038] This invention also provides an automatic extraction system for photo-assisted electrocatalysis literature information, including: a literature acquisition and parsing module, a material entity identification module, a performance parameter extraction module, a correspondence construction module, a data standardization module, a data verification module, and a database construction module;

[0039] The document acquisition and parsing module is used to acquire relevant literature on photo-assisted electrocatalysis and parse it to obtain clean text.

[0040] The material entity recognition module is used to identify material entities in the clean text;

[0041] The performance parameter extraction module is used to identify the reaction type in the cleaning text and extract performance parameters from the cleaning text based on the reaction type.

[0042] The correspondence construction module is used to obtain the correspondence between the material entity and the performance parameters based on the material entity and the performance parameters.

[0043] The data standardization module is used to perform multi-dimensional standardization processing on the correspondence to obtain standardized data.

[0044] The data verification module is used to perform data verification and anomaly detection on the standardized data;

[0045] The database construction module is used to store the standardized data into a structured database and complete the automatic extraction of documents.

[0046] Compared with the prior art, the present invention has the following advantages and technical effects:

[0047] (1) This invention addresses the complex naming and diverse structural expressions of materials in the field of photo-assisted electrocatalysis. It can stably identify composite, heterojunction, supported, doped and substrate material entities and complete synonym merging, thereby improving the completeness and accuracy of material extraction.

[0048] (2) Before parameter extraction, the present invention identifies the literature reaction type R and selects the corresponding parameter category set and extraction rules according to the reaction type, so that the performance parameter extraction has domain constraints and improves the pertinence and reliability of parameter extraction under different reaction systems.

[0049] (3) This invention splits and aligns parallel sentences under reaction type constraints, constructs a one-to-one correspondence between multiple materials and multiple parameters, and suppresses mismatches by verifying reaction legality and semantic consistency, thereby improving the accuracy of material-performance mapping.

[0050] (4) The present invention provides multi-dimensional standardization of the execution unit, reference electrode, illumination conditions and test environment, and different standardization strategies can be adopted according to the reaction type, so that data across literature and across systems are comparable under a unified benchmark.

[0051] (5) Based on the reaction type, the present invention configures different reasonable intervals and conflict rules, and combines anomaly detection and missing completion mechanisms to achieve automated quality control and ensure the reliability and consistency of database data;

[0052] (6) Through processes such as extraction, correspondence, standardization and verification, this invention forms a sustainably expandable photo-assisted electrocatalysis structured database, which supports multidimensional retrieval and data retrieval, and provides a high-quality data foundation for material screening, structure-activity analysis and machine learning modeling. Attached Figure Description

[0053] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0054] Figure 1 This is a flowchart of an automatic extraction method for literature information by photo-assisted electrocatalysis according to an embodiment of the present invention;

[0055] Figure 2 This is a schematic diagram of parallel recognition and structured conversion under reaction type constraints in an embodiment of the present invention. Detailed Implementation

[0056] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0057] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0058] This embodiment proposes an automatic extraction method for literature information on photo-assisted electrocatalysis, such as... Figure 1 As shown, the specific steps include:

[0059] Relevant literature on photo-assisted electrocatalysis was obtained and parsed to produce clean text.

[0060] Identify the material entities and reaction types in the clean text;

[0061] Based on the reaction type, performance parameters are extracted from the clean text;

[0062] Based on the material entities and the performance parameters, when a parallel description structure is detected, the parallel result reorganization module is connected with the performance parameter extraction module and the material entity recognition module. Its input is the set of material entities output by the material entity recognition module and the set of performance parameters (including parameter type, value and reaction type label) output by the performance parameter extraction module. Its output is the one-to-one correspondence between material entities and performance parameters.

[0063] The correspondence is standardized in multiple dimensions to obtain standardized data;

[0064] The standardized data is stored in a structured database to complete the automatic extraction of documents.

[0065] Specifically, S1 is document acquisition and text parsing; S2 is material entity identification; S3 is performance parameter identification and extraction; S4 is correspondence construction; S5 is data standardization processing; S6 is data verification and anomaly detection; and S7 is structured database construction.

[0066] Furthermore, the clean texts obtained from relevant literature on photo-assisted electrocatalysis include:

[0067] Obtain document files in PDF, HTML, XML, or LaTeX formats from publicly available literature databases, institutional databases, or other accessible literature collections;

[0068] The document file is parsed to obtain the text to be processed, which contains structural information;

[0069] The text to be processed is formatted and denoised to obtain the clean text.

[0070] Specifically, in this embodiment, the first step is to acquire relevant literature on photo-assisted electrocatalysis from the target database. The target database can be a publicly available literature database, an institution's self-built database, or other accessible collections of literature, wherein the document carrier can include, but is not limited to, PDF, HTML, XML, or other structured / semi-structured document formats.

[0071] After obtaining the literature, the document is parsed to obtain the text to be processed. The parsing result should at least contain the main text, and preferably also include structural information such as figure captions, table text, and their paragraph positions. The parsing method can be:

[0072] PDF-based text extraction and parsing;

[0073] Parsing based on HTML / web page structure tags;

[0074] Extraction based on structured tags (such as XML, TEI, LaTeX, or database built-in fields);

[0075] Any combination of the above methods.

[0076] To reduce noise, the parsed text undergoes formatting and noise reduction processing, including but not limited to:

[0077] Remove headers, footers, reference numbers, and meaningless symbols;

[0078] Standardize the representation of subscripts, superscripts, special characters, Greek letters, and hyphens;

[0079] Encode paragraphs, sentences, and table cells sequentially while preserving their positional information;

[0080] For scanned documents, the initial text can be obtained first through character recognition (OCR) before performing the above-mentioned noise reduction.

[0081] The output obtained by S1 is clean text and its structural location information, which serves as the input for subsequent recognition.

[0082] Furthermore, identifying material entities in the clean text includes:

[0083] The clean text is matched based on a pre-built material dictionary to identify elemental materials, compounds, composite materials, heterojunction materials, supported materials, and doped materials.

[0084] Regular rules are used to identify chemical formula expressions, complex structure expressions, doped expressions, carrier expressions, and multi-component combined table expressions in clean text;

[0085] The clean text was annotated using a sequence labeling model;

[0086] The dictionary matching results, regular rule recognition results, and model annotation results are fused together, and the material entity is output through a voting or confidence weighting strategy.

[0087] Specifically, in this embodiment, material entity recognition is performed on the clean text obtained in S1 to extract material name entities appearing in the literature. Material entities include at least: elements, compounds, composite materials, heterojunction materials, supported materials, doped materials and their carriers / substrates, etc.

[0088] Material entity recognition can be achieved using one or more of the following methods:

[0089] Dictionary / knowledge base matching method:

[0090] A materials dictionary or materials knowledge base is pre-built. The dictionary sources can include existing catalytic materials databases, domain reviews, public materials libraries, and manually expanded entries. During recognition, candidate segments in the text are matched against the dictionary, and the matched segments are output as material entities.

[0091] Rule / regular expression recognition method:

[0092] A set of rules is constructed based on the characteristics of material expression. The rules may cover, but are not limited to:

[0093] Chemical formula expression (e.g., CdS, Ni3S2, CoFe2O4);

[0094] Complex / heterogeneous structure expression (e.g., A@B, A / B, A–B, A|B);

[0095] Doping or solid solution expression (e.g., A:Cu, Cu-doped A, Al₋) x B x );

[0096] Vector / substrate expression (e.g., A / NF, AonTi, A-supported B);

[0097] Multiple components can be listed in a combined table (e.g., A–B–C, A / B / C).

[0098] Text is scanned, located, merged, and deduplicated using rules.

[0099] Sequence labeling or neural network recognition methods (optional implementation):

[0100] The sentence is represented as a sequence input model, material entity boundaries are labeled, and candidate material fragments are output. In one embodiment, the sequence labeling model is an improved sequence labeling model, which, based on existing sequence labeling models, has undergone structural improvements to address the complex naming and diverse symbols of chemical material entities, thereby enhancing the accuracy of material entity boundary identification. The improvements include at least:

[0101] (1) Introduce feature enhancement processing of material structure and symbol features in the model input stage to enhance the ability to represent chemical formulas, composite materials and load-bearing material names;

[0102] (2) In the process of sequence modeling, contextual semantic information is combined to jointly model material entities composed of cross words or cross symbols;

[0103] (3) Introduce label validity constraints in the label decoding stage to reduce the generation of label sequences that do not conform to the material entity labeling rules. Therefore, the structure of the improved sequence labeling model can be abstractly represented as: input text → feature enhancement module → sequence modeling module → constraint decoding module → output label sequence.

[0104] Fusion recognition method:

[0105] The dictionary matching, rule recognition and model recognition are integrated, and the final material entity is output using voting, confidence weighting or priority strategies.

[0106] Post-processing of the recognition results:

[0107] Merge synonym / alias expressions (such as "Nifoam", "NF", etc.);

[0108] Unify capitalization, hyphens, and subscript / superscript representation;

[0109] Preserve the mapping between the original form and the standardized form.

[0110] S2 yields the set of material entities and their position indices in the text.

[0111] Furthermore, identifying the reaction type in the clean text includes:

[0112] Extract the method section, test section, chart title, and text near the reaction formula from the clean text as reaction-related sentence segments;

[0113] Identify reaction trigger words in the reaction-related sentence segments, and determine the reaction type R based on the reaction trigger words. When the reaction type cannot be identified, the reaction type R is marked as unknown.

[0114] Specifically, the input consists of reaction-related segments from clean text, including at least the method segment, test segment, graph title, and text near the reaction formula. The reaction type is determined by identifying reaction trigger words, reaction formula expressions, or test potential windows. Reaction trigger words include at least OER / oxygen evolution, HER / hydrogen evolution, and ORR / oxygen reduction. The output is a reaction type label R (e.g., OER / HER / ORR / other). If the reaction type cannot be identified, R is marked as unknown.

[0115] After obtaining the reaction type label R, the corresponding performance parameter set and its trigger words / extraction rule base are invoked based on the reaction type to perform differentiated performance parameter identification and extraction. When the reaction type cannot be identified, extraction is performed based on the general parameter set, and the reaction type is marked as unknown for subsequent steps to verify or complete.

[0116] The output of S3 is a set of candidate records containing parameter type, value, unit, test condition and its text position index, and carries the reaction type label R as the input constraint of S4.

[0117] Furthermore, based on the reaction type, extracting performance parameters from the clean text includes:

[0118] Based on the reaction type R, the corresponding parameter trigger dictionary is invoked to locate the parameter name;

[0119] Extract numerical values, units, and conditional information from the sentence containing the parameter name or adjacent sentences. When the unit or conditional information appears across sentences, complete it by backtracking across sentences and associating it with the nearest neighbor.

[0120] The output includes performance parameters with parameter type, value, unit, and conditions.

[0121] Specifically, in this embodiment, the performance parameters and their values ​​of the material are identified and extracted from clean text. The performance parameters include at least: Tafel slope, overpotential, voltage (light / dark state), current density, band gap, conductivity, etc., and the parameter set can be expanded to other equivalent or related indicators depending on the field.

[0122] Parameter identification and numerical extraction can be performed using a combination of pattern matching and context completion, including at least the following:

[0123] 1. Parameter Name Location: Identify parameter trigger words based on parameter keywords or a thesaurus. These trigger words may include, for example, overpotential, η, overpotential, @10 mA cm⁻¹. -2 Tafel slope, mV dec-1 , under light, dark, illumination, etc.;

[0124] 2. Numerical and Unit Extraction: Identify the corresponding numerical values ​​and units near the parameter name or within the same sentence. The numerical values ​​include at least single values, interval values, error expressions, and multi-valued lists.

[0125] 3. Conditional information completion: For the extracted parameter values, further conditional information such as reference electrode type, electrolyte / environment, test current density or potential point is obtained from the context; when units or conditions appear across sentences, completion is performed by backtracking across sentences and associating with nearest neighbors.

[0126] Further, obtaining the correspondence between the material entity and the performance parameters based on the material entity and the performance parameters includes:

[0127] Parallel sentences are detected in the clean text. When parallel sentences exist, the material entities in the detected parallel sentence segments are split into ordered lists, and the performance parameters are split into ordered numerical lists. Under the constraint of the reaction type R, the ordered lists and the ordered numerical lists are matched one by one in sequence to obtain the correspondence between the material entities and the performance parameters. When the number of material entities is inconsistent with the number of performance parameters, a default padding or nearest neighbor backoff strategy is used to complete the matching.

[0128] When no parallel syntax exists, the material entity and performance parameters are directly established in a preset order.

[0129] Specifically, based on the material entities and performance parameters, when a parallel description structure is detected, the parallel result reconstruction module is connected to the performance parameter extraction module and the material entity recognition module. Its inputs are the set of material entities output by the material entity recognition module and the set of performance parameters (including parameter type, value, and reaction type label) output by the performance parameter extraction module. Its output is a one-to-one correspondence between material entities and performance parameters. The parallel result reconstruction includes:

[0130] (1) Split the material entities in the parallel sentence segments to obtain a list of material entities;

[0131] (2) The performance parameter values ​​in the parallel segments are split to obtain a list of parameter values;

[0132] (3) Under the constraints of the reaction type identification results, the split material entities and performance parameter values ​​are redistributed and aligned to construct a one-to-one correspondence between material entities and performance parameters.

[0133] When no parallel description structure is detected, the material entity and performance parameters are directly established in the default order.

[0134] like Figure 2 As shown, in this embodiment, for parallel descriptive sentences containing multiple materials and multiple parameter values, a one-to-one correspondence between material entities and parameter values ​​is constructed. First, parallel sentence structure detection is performed on the text, and the detection rules include at least: parallel trigger words, parallel structures, and tabular parallel structures; the parallel trigger words include at least "separately," "each respectively," "corresponding," "in sequence," "each reaches," "each is," etc.

[0135] Perform parallel segmentation and alignment matching on the detected parallel segments, including at least the following steps:

[0136] 1. Material string splitting: Based on delimiters such as commas, semicolons, and, / or conjunctions, divide the material entities into ordered lists {M1…M};

[0137] 2. Parameter value string splitting: Divide the performance parameter values ​​into an ordered list of values ​​{V1…V};

[0138] 3. Alignment matching under reaction type constraints: When the output reaction type R exists and is not unknown, the corresponding parameter category set is determined according to the reaction type R, and the parameter values ​​are confirmed and assigned under the constraints of this set; the preferred method is to use the sequential alignment strategy to generate pairings (Mᵢ,Vᵢ,R); when the number of materials and the number of parameter values ​​are inconsistent, the most reasonable one-to-one correspondence result is obtained by using default completion, nearest neighbor backoff, or a secondary matching strategy based on keyword windows.

[0139] 4. Legality and consistency verification: Preferably, the legality of the corresponding parameter category is verified based on the reaction type R. If the parameter category does not match the reaction type, a back-alignment is triggered or the parameter is marked as an abnormal candidate. At the same time, the semantic consistency of the matching result is checked, including at least unit consistency and parameter rationality checks.

[0140] The material-parameter pairing record is obtained through S4. The record contains the material entity, parameter type, value, corresponding conditions, and reaction type label R.

[0141] Furthermore, the correspondence is standardized in multiple dimensions to obtain standardized data, including:

[0142] The correspondence is standardized by unit standardization, reference electrode conversion, and illumination condition normalization to obtain the standardized data.

[0143] Specifically, in this embodiment, the standardization process selects the corresponding reference conversion relationship, unit standard, and illumination labeling system based on the reaction type R.

[0144] Standardization was performed on the material-parameter paired records in S4 to ensure consistent comparability across different literature sources. This included at least the following processing:

[0145] 1. Unit standardization: Converts the units of indicators such as voltage, overpotential, current density, and Tafel slope to preset standard units;

[0146] 2. Reference electrode conversion: When the parameters involve a potential system, identify its reference electrode type and convert it to a standard reference system (such as RHE) according to a preset conversion relationship.

[0147] 3. Illumination condition normalization: Descriptions such as "illuminance / darkness / intermittent light / simulated sunlight / AM1.5G" are mapped to a standard set of illumination labels;

[0148] Standardization of test environment parameters: Identify and extract environmental information such as electrolyte type, and represent it as a structured field. Output standardized performance data records via S5.

[0149] Furthermore, storing the standardized data in a structured database to complete the automatic extraction of documents includes:

[0150] The standardized data is validated and anomaly detected to obtain a structured data set of target quality.

[0151] The target quality structured data set is stored in a structured database to complete the automatic extraction of documents.

[0152] Specifically, in this embodiment, quality verification and anomaly detection are performed on standardized performance data to ensure data reliability. Verification content includes, but is not limited to:

[0153] 1. Range Reasonableness Verification: The reasonable range of the parameters uses different preset threshold sets based on the reaction type R. The system determines whether the values ​​are abnormal based on the preset reasonable range. The reasonable range can be derived from domain knowledge, historical data statistics, or user configuration. Values ​​exceeding the range are marked as abnormal candidates.

[0154] 2. Conflict and duplicate detection: For multiple records of the same material and the same parameter: if the values ​​are consistent or the differences are within the allowable range, they can be merged into one record; if the differences are too large, they are marked as conflict records and the source information is retained.

[0155] 3. Conversion Anomaly Detection: For the results after unit standardization or reference conversion, detect whether there are illegal values, unit mismatch, or conversion failures, and perform rollback or mark them.

[0156] 4. Missing information completion and marking: For records that lack units, conditions or reference information, complete them by backtracking through the nearest neighbor context; records that cannot be completed are marked as missing for subsequent manual review.

[0157] The S6 output is a high-quality structured data set after verification.

[0158] In this embodiment, the data set obtained in S6 is written into a structured database. The database contains at least the following fields:

[0159] Material entity (original and standardized forms); performance parameter types and standardized values; standardized units; illumination labels; testing environment conditions; literature source information (such as DOI, title, author, year, etc.).

[0160] The database structure can be a relational database, a graph database, a document database, or a combination thereof.

[0161] Users can search, analyze, or export the database by dimensions such as material composition, structural type, performance range, illumination label, and test conditions to support material screening, structure-property analysis, and machine learning model training.

[0162] This embodiment selects a literature containing parallel descriptions of photo-assisted electrocatalysis performance as input literature. The literature contains parallel descriptions of the performance parameters of the same material under different reaction types.

[0163] (1) Input a document excerpt:

[0164] A certain document records the following content (rewritten for ease of explanation):

[0165] "Under illumination, the optimized NiFe-ZAIS-5 sample exhibited an overpotential of 129 mV for HER and 220 mV for OER at 10 mA·cm⁻², both of which were superior to its performance in dark electrocatalysis."

[0166] (2) Based on the processing result of S1:

[0167] The input document is parsed and denoised to obtain a clean text containing the main text and test paragraphs.

[0168] (3) Based on the material entity identification results of S2:

[0169] Material entities are identified from clean text, resulting in a material set: M={NiFe-ZAIS-5}.

[0170] (4) Based on the reaction type identification results of S3a: Identify the reaction trigger words “HER” and “OER” in the sentence segments related to the reaction, and determine the reaction type set: R={HER,OER}.

[0171] (5) Based on the performance parameter identification and extraction results of S3: Under the reaction type constraint, the corresponding parameter set is called to extract the following:

[0172] η(HER) = 129mV, the test point is 10mA·cm⁻²;

[0173] η(OER) = 220mV, the test point is 10mA·cm⁻²;

[0174] At the same time, the lighting condition information is extracted as "lighting".

[0175] (6) Based on the correspondence in S4, the following results are constructed: Since there is a parallel structure within the same segment where a single material corresponds to multiple reaction type parameters, the parameter values ​​are assigned according to the reaction type and a one-to-one correspondence is established. The corresponding records are output as follows:

[0176] (NiFe-ZAIS-5, η=129mV@10mA·cm⁻²,R=HER),

[0177] (NiFe-ZAIS-5, η=220mV@10mA·cm⁻², R=OER).

[0178] Therefore, this embodiment can first identify the reaction type and then extract and assign differentiated parameters accordingly, thereby obtaining accurate and structured material-parameter-reaction type correspondence results in the scenario of parallel description of multiple reaction types, providing reliable input for subsequent standardization and database construction.

[0179] This embodiment also provides an automatic extraction system for photo-assisted electrocatalysis literature information, including: a literature acquisition and parsing module, a material entity identification module, a performance parameter extraction module, a correspondence construction module, a data standardization module, a data verification module, and a database construction module;

[0180] The document acquisition and parsing module is used to acquire relevant literature on photo-assisted electrocatalysis and parse it to obtain clean text.

[0181] The material entity recognition module is used to identify material entities in the clean text;

[0182] The performance parameter extraction module is used to identify the reaction type in the cleaning text and extract performance parameters from the cleaning text based on the reaction type.

[0183] The correspondence construction module is used to obtain the correspondence between the material entity and the performance parameters based on the material entity and the performance parameters.

[0184] Based on the material entities and the performance parameters, when a parallel description structure is detected, the parallel result reorganization module is connected with the performance parameter extraction module and the material entity recognition module. Its input is the set of material entities output by the material entity recognition module and the set of performance parameters (including parameter type, value and reaction type label) output by the performance parameter extraction module. Its output is the one-to-one correspondence between material entities and performance parameters.

[0185] The data standardization module is used to perform multi-dimensional standardization processing on the correspondence to obtain standardized data.

[0186] The data verification module is used to perform data verification and anomaly detection on the standardized data;

[0187] The database construction module is used to store the standardized data into a structured database and complete the automatic extraction of documents.

[0188] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for automatically extracting literature information from photo-assisted electrocatalysis, characterized in that, include: Relevant literature on photo-assisted electrocatalysis was obtained and parsed to produce clean text. Identify the material entities and reaction types in the clean text; Based on the reaction type, performance parameters are extracted from the clean text; Based on the material entity and the performance parameters, obtain the correspondence between the material entity and the performance parameters; The correspondence is standardized in multiple dimensions to obtain standardized data; The standardized data is stored in a structured database to complete the automatic extraction of documents.

2. The method for automatically extracting literature information by photo-assisted electrocatalysis according to claim 1, characterized in that, The literature related to photo-assisted electrocatalysis was retrieved and parsed to obtain clean texts, including: Obtain document files in PDF, HTML, XML, or LaTeX formats from publicly available literature databases, institutional databases, or other accessible literature collections; The document file is parsed to obtain the text to be processed, which contains structural information; The text to be processed is formatted and denoised to obtain the clean text.

3. The method for automatically extracting literature information by photo-assisted electrocatalysis according to claim 1, characterized in that, Identifying material entities in the cleaning text includes: The clean text is matched based on a pre-built material dictionary to identify elemental materials, compounds, composite materials, heterojunction materials, supported materials, and doped materials. Regular rules are used to identify chemical formula expressions, complex structure expressions, doped expressions, carrier expressions, and multi-component combined table expressions in clean text; The clean text was annotated using a sequence labeling model; The dictionary matching results, regular rule recognition results, and model annotation results are fused together, and the material entity is output through a voting or confidence weighting strategy.

4. The method for automatically extracting literature information by photo-assisted electrocatalysis according to claim 1, characterized in that, Identifying the reaction type in the clean text includes: Extract the method section, test section, chart title, and text near the reaction formula from the clean text as reaction-related sentence segments; Identify reaction trigger words in the reaction-related sentence segments, and determine the reaction type R based on the reaction trigger words. When the reaction type cannot be identified, the reaction type R is marked as unknown.

5. The method for automatically extracting literature information by photo-assisted electrocatalysis according to claim 4, characterized in that, Based on the reaction type, performance parameters are extracted from the clean text, including: Based on the reaction type R, the corresponding parameter trigger dictionary is invoked to locate the parameter name; Extract numerical values, units, and conditional information from the sentence containing the parameter name or adjacent sentences. When the unit or conditional information appears across sentences, complete it by backtracking across sentences and associating it with the nearest neighbor. The output includes performance parameters with parameter type, value, unit, and conditions.

6. The method for automatically extracting literature information by photo-assisted electrocatalysis according to claim 4, characterized in that, Obtaining the correspondence between the material entity and the performance parameters based on the material entity and the performance parameters includes: Parallel sentences are detected in the clean text. When parallel sentences exist, the material entities in the detected parallel sentence segments are split into ordered lists, and the performance parameters are split into ordered numerical lists. Under the constraint of the reaction type R, the ordered lists and the ordered numerical lists are matched one by one in sequence to obtain the correspondence between the material entities and the performance parameters. When the number of material entities is inconsistent with the number of performance parameters, a default padding or nearest neighbor backoff strategy is used to complete the matching. When no parallel syntax exists, the material entity and performance parameters are directly established in a preset order.

7. The method for automatically extracting literature information by photo-assisted electrocatalysis according to claim 1, characterized in that, The correspondence is standardized in multiple dimensions to obtain standardized data, including: The correspondence is standardized by unit standardization, reference electrode conversion, and illumination condition normalization to obtain the standardized data.

8. The method for automatically extracting literature information by photo-assisted electrocatalysis according to claim 1, characterized in that, The standardized data is stored in a structured database, and the automatic extraction of documents is completed by: The standardized data is validated and anomaly detected to obtain a structured data set of target quality. The target quality structured data set is stored in a structured database to complete the automatic extraction of documents.

9. An automatic extraction system for photo-assisted electrocatalysis literature information, used to implement the method as described in any one of claims 1-8, characterized in that, include: The system includes modules for literature acquisition and parsing, material entity recognition, performance parameter extraction, correspondence construction, data standardization, data verification, and database construction. The document acquisition and parsing module is used to acquire relevant literature on photo-assisted electrocatalysis and parse it to obtain clean text. The material entity recognition module is used to identify material entities in the clean text; The performance parameter extraction module is used to identify the reaction type in the cleaning text and extract performance parameters from the cleaning text based on the reaction type. The correspondence construction module is used to obtain the correspondence between the material entity and the performance parameters based on the material entity and the performance parameters. The data standardization module is used to perform multi-dimensional standardization processing on the correspondence to obtain standardized data. The data verification module is used to perform data verification and anomaly detection on the standardized data; The database construction module is used to store the standardized data into a structured database and complete the automatic extraction of documents.

Images