A Method for Compiling a Lifecycle Unit Process Data Inventory Based on Entity Recognition

Abstract

This application provides a method for compiling a lifecycle unit process data list based on entity recognition. The method includes: acquiring a document to be recognized and a table name corpus; determining target tables in the document to be recognized based on table names in the table name corpus; based on an entity recognition model, recognizing entities in the target tables according to preset material names to obtain multiple target material names; determining a first target cell and a second target cell in the target table; if the second target cell meets the content acquisition conditions, acquiring the target data in the second target cell; and generating a lifecycle unit process data list for the target product based on the target material names and target data. This application can reduce data collection and processing costs and improve the efficiency and accuracy of data collection for lifecycle unit processes, thereby improving the efficiency and accuracy of compiling a lifecycle unit process data list based on entity recognition.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method for compiling a lifecycle unit process data inventory based on entity recognition. Background Technology

[0002] In today's world, sustainable development and environmental awareness play a crucial role in decision-making, and Life Cycle Assessment (LCA) has become an important tool for assessing the environmental impact (including greenhouse gas emissions and resource consumption) of products, services, or processes throughout their life cycles. Taking carbon footprint calculation (life cycle greenhouse gas emissions) as an example, electricity production unit process data from internationally developed life cycle databases (such as Ecoinvent in Switzerland and Gabi in Germany) are typically used to calculate the product's carbon footprint. However, these international databases suffer from data lag and data scarcity, leading to low accuracy and significant uncertainty in the calculated carbon footprint. To improve the accuracy of product carbon footprint calculations, it is necessary to obtain life cycle process data corresponding to the region where the product is located. However, without utilizing international databases, current methods for obtaining life cycle unit process data mainly rely on literature reviews and surveys of key enterprises, which are inefficient and costly. Summary of the Invention

[0003] This application provides a method for compiling a lifecycle unit process data list based on entity recognition, which aims to improve the efficiency and accuracy of lifecycle unit process data collection, thereby improving the accuracy of the lifecycle unit process data list and reducing data processing costs.

[0004] In a first aspect, this application provides a method for compiling a lifecycle unit process data inventory based on entity recognition, the method comprising the following steps:

[0005] Obtain the document to be identified and the table name corpus, wherein the table name corpus includes various table names;

[0006] The target table is determined in the document to be identified based on the table names in the table name corpus. The target table contains life cycle unit process data of the target product.

[0007] Based on the entity recognition model, the entities in the target table are identified according to the preset material names to obtain multiple target material names;

[0008] In the target table, a first target cell is determined that records the name of the target material, and a second target cell is determined that is adjacent to the first target cell;

[0009] If the second target cell meets the content acquisition conditions based on the identification result obtained by identifying the data type of the second target cell, the target data in the second target cell is acquired.

[0010] Generate a lifecycle unit process data list for the target product based on the target material name and the target data.

[0011] In one embodiment, if it is determined that the second target cell meets the content acquisition conditions based on the recognition result obtained by data type recognition of the second target cell, the target data in the second target cell is acquired, including:

[0012] If the identification result of the data type of the second target cell includes single-precision floating-point type, determine the position information of the second target cell;

[0013] If the second target cell meets the preset location conditions based on the location information of the second target cell, the target data in the second target cell is obtained.

[0014] In one embodiment, when it is determined that the second target cell meets a preset position condition based on the position information of the second target cell, obtaining the target data in the second target cell includes:

[0015] When the second target cell and the first target cell are in a first adjacent relationship, the column containing the second target cell in the table is determined as the first target column;

[0016] The cell at the target position in the first target column is identified as the third target cell, and character recognition processing is performed on the third target cell to obtain the keywords in the third target cell;

[0017] If the keyword in the third target cell matches a preset keyword, retrieve the target data from the second target cell; or

[0018] When the second target cell and the first target cell are in a second adjacent relationship, the row containing the second target cell in the target table is determined as the first target row;

[0019] The cell at the target position in the first target row is identified as the fourth target cell, and character recognition processing is performed on the fourth target cell to obtain the keywords in the fourth target cell;

[0020] If the keyword in the fourth target cell matches the preset keyword, the target data in the second target cell is obtained.

[0021] In one embodiment, when it is determined that the second target cell meets the content acquisition conditions based on the recognition result obtained by data type recognition of the second target cell, the method further includes:

[0022] Based on the entity recognition model, the entities in the target table are recognized according to a preset string to determine the second target column. The unit of measurement recorded in the cell in the second target column that is in the same row as the second target cell is determined as the unit of measurement corresponding to the target data. All cells in the second target column are used to record the unit of measurement; or

[0023] Based on the entity recognition model, according to the entity recognition result of the target table based on the preset string, a second target row is determined, and the unit of measurement recorded in the cell in the second target row that is in the same column as the second target cell is determined as the unit of measurement corresponding to the target data, wherein all cells in the second target row are used to record the unit of measurement; or

[0024] Based on a preset unit of measurement standard, string matching is performed on the third target cell or the fourth target cell to obtain the unit of measurement corresponding to the target data.

[0025] In one embodiment, based on an entity recognition model, entities in the target table are identified according to preset material names to obtain multiple target material names, including:

[0026] Based on the entity recognition model, the entities in each cell of the target table are vectorized to obtain the first word vector corresponding to each entity in each cell;

[0027] Determine the second word vector corresponding to each of the preset material names;

[0028] The target material name is determined based on the first word vector and the second word vector.

[0029] In one embodiment, determining the target material name based on the first word vector and the second word vector includes:

[0030] Determine the second word vector that matches each of the first word vectors;

[0031] The preset material name corresponding to the matched second word vector is determined as the target material name.

[0032] In one embodiment, determining a second word vector that matches each of the first word vectors includes:

[0033] Based on the cosine similarity calculation model, the cosine similarity between the first word vector and each of the second word vectors is determined;

[0034] The second word vector whose cosine similarity is greater than or equal to the cosine similarity threshold is determined as the second word vector that matches the first word vector.

[0035] In one embodiment, the method for compiling a lifecycle unit process data inventory based on entity recognition further includes:

[0036] Obtain the document to be mined, which is used to determine the materials corresponding to the life cycle unit process of the target product;

[0037] The target material name and the target data generate a lifecycle unit process data list for the target product, including:

[0038] A lifecycle unit process data list for the target product is generated based on the material name obtained from the document to be mined, the target material name, and the target data.

[0039] In one embodiment, generating a lifecycle unit process data list for the target product based on the product name obtained from the document to be mined, the target material name, and the target data includes:

[0040] Identify the target material name that matches the material name obtained in the document to be mined;

[0041] The target material name that matches the material name obtained in the document to be mined, and the target data corresponding to the matched target material name are filled into the preset list template to obtain the life cycle unit process data list of the target product.

[0042] In one embodiment, the method for compiling a lifecycle unit process data list based on entity recognition further includes: performing word segmentation on the document to be mined based on a text mining model to obtain multiple corpus units;

[0043] Based on preset semantic rules, string matching processing is performed on multiple corpus units to obtain at least one table name, and the table name corpus is established based on the table name.

[0044] This application provides a method for compiling a product lifecycle unit process data list based on entity recognition. This method utilizes a table name corpus to identify target tables in the document to be identified, and then uses an entity recognition model to identify entities in the target tables, obtaining multiple target material names. It also identifies second target cells in the target tables that meet the content retrieval criteria, and obtains the target data from these second target cells. Based on the target material names and the obtained target data, the product lifecycle unit process list is compiled. By extracting entity information from the document to be identified using an entity recognition model, the method achieves automatic compilation of the product lifecycle unit process list, improves the efficiency and accuracy of lifecycle unit process data collection, and reduces data processing costs. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart illustrating a method for compiling a lifecycle unit process data list based on entity recognition, as provided in an embodiment of this application. Detailed Implementation

[0047] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0048] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0049] This application provides a method for compiling a lifecycle unit process data list based on entity recognition. This method can be applied to terminal devices, such as tablet computers, laptops, and desktop computers.

[0050] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0051] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for compiling a lifecycle unit process data list based on entity recognition, provided as an embodiment of this application.

[0052] It should be noted that unit process data is the smallest unit in lifecycle inventory analysis. It quantifies the inputs (raw materials, energy, resources, etc.) and outputs (products, by-products, environmental emissions, waste, etc.) of a unit process, thus reflecting the data corresponding to the aforementioned processes. The products mentioned include, but are not limited to, products from various industries such as chemical products and agricultural products. Therefore, product lifecycle unit process data can provide users with complete and intuitive data, providing a theoretical and data foundation for assessing the environmental impact of products.

[0053] The data required for compiling a product lifecycle unit process data list is mainly recorded in tabular form in environmental impact assessment reports and / or energy assessment reports. Current text recognition technology cannot effectively identify and extract the data recorded in these tables. Furthermore, because material names may be abbreviated in the reports, the accuracy of lifecycle unit process data lists derived directly from table data is low. Therefore, the compilation of existing lifecycle unit process data lists mostly relies on manual data acquisition and comparison, resulting in high labor costs and low efficiency. The lifecycle unit process data list compilation method based on entity recognition provided in this application can extract data from tables, thereby improving data processing efficiency, reducing data processing costs, and increasing the accuracy of the list.

[0054] like Figure 1 As shown, the method for compiling a lifecycle unit process data list based on entity recognition includes steps S101 to S106.

[0055] Step S101: Obtain the document to be identified and the table name corpus, wherein the table name corpus includes a variety of table names.

[0056] For example, the table name corpus can be a pre-constructed corpus of table names. Specifically, the table names in the corpus can be mined from multiple documents to be mined. These documents include, but are not limited to, product environmental impact assessment reports, which can be obtained from publicly available information. It should be understood that the table name corpus includes various table names, such as raw material consumption tables, energy consumption tables, pollutant tables, and waste gas tables, thereby enabling the identification of tables in the documents to be identified based on the table names contained in the corpus.

[0057] Step S102: Determine the target table in the document to be identified based on the table names in the table name corpus. The target table contains life cycle unit process data of the target product.

[0058] For example, matching the table names in the table name corpus to the document to be identified is performed to determine the target table within the document. It should be noted that compiling the lifecycle unit process data for the target product includes the materials associated with the lifecycle unit processes of the target product and their corresponding consumption or emission data. The materials associated with the lifecycle unit processes of the target product include, but are not limited to, the raw materials required for manufacturing the target product and the pollutants emitted. Related data on raw material consumption and / or pollutant emissions are generally recorded in tabular format, such as in environmental impact assessment (EIA) documents. Therefore, it is necessary to locate the target table containing the lifecycle unit process data of the target product in the EIA document to obtain the data and / or information from the target table, and then obtain the data and / or information required to compile the lifecycle unit process data list for the target product from the target table.

[0059] In some embodiments, determining the target table in the document to be identified based on the table names in the table name corpus includes: traversing the document to be identified based on the table names in the table name corpus; and identifying the table in the document to be identified whose table name is the same as the table name in the table name corpus as the target table.

[0060] For example, the document to be identified is traversed by table name to identify and locate the table in the document. In this step, if it is determined that there is a string in the document to be identified whose text similarity with the table name is less than or equal to a preset text similarity threshold, then the location of the string is determined as the location of the target table. Specifically, the page number or line number of the string in the document to be identified is determined to determine the location of the table corresponding to the string, and the table corresponding to the string is determined as the target table. This enables the extraction of data or information from the target table to complete the compilation of the life cycle unit process data list based on entity recognition.

[0061] Step S103: Based on the entity recognition model, the entities in the target table are identified according to the preset material names to obtain multiple target material names.

[0062] For example, the preset material name can be determined based on the input information of the material name, or it can be obtained from a preset industry corpus, so as to use the preset material name to identify the entity in the target table and determine the target material name in the table.

[0063] It should be understood that identifying entities in the table by pre-setting material names can reduce the matching difficulty caused by the high complexity of industry material names and improve the accuracy of entity matching.

[0064] It should be noted that the material names in this application include, but are not limited to, energy names, material names, and compound names, such as, but not limited to, the names of materials such as water, electricity, plastics, and nitrogen oxides.

[0065] In some embodiments, the step of identifying entities in the target table based on the entity recognition model and obtaining multiple target material names by means of: vectorizing the entities in each cell of the target table based on the entity recognition model to obtain a first word vector corresponding to each entity in each cell; determining a second word vector corresponding to each of the preset material names; and determining the target material name based on the first word vector and the second word vector.

[0066] For example, before performing entity recognition on the target table, non-numeric cells in the target table are identified for entity recognition processing to improve the accuracy and efficiency of entity recognition. Based on the entity recognition model, the entities in the non-numeric cells of the target table are vectorized to obtain a first word vector corresponding to the entity in the non-numeric cell, and a second word vector for determining the preset material name. The target material name is then determined based on the first and second word vectors.

[0067] The following section uses the material name as an example to illustrate the process of determining the target material name. In the specific implementation process, determining the second word vector of the preset chemical product name includes steps S1031 to S1035.

[0068] Step S1031: Perform word segmentation on the chemical product names in the chemical industry product corpus. Specifically, use the TF-IDF algorithm to model the segmented documents in the corpus to obtain the weight of each segmentation result in the text.

[0069] Step S1032: Train the word-segmented chemical product name text based on the Word2Vec model. For each word-segmented chemical product name, extract one or more words X from the context of the target word as needed. iEach word is represented by a one-hot encoding, forming a matrix of size [1*V], where V represents the number of word segments for each industry product name. Taking the chemical product PBAT as an example, it is synthesized from three chemical monomers: [polybutylene terephthalate (PBAT) (chemical name), adipic acid, butanediol, terephthalic acid]. If the target word is terephthalic acid, then the one-hot encoding of PBAT is: [1,0,0,0.....]. 100 (Dimension 100); For each industry product name data, the one-hot matrix of each word is multiplied by the input weight matrix W[V*N] to obtain the word vector corresponding to each word. The word vector is a matrix of size [1*N], where N refers to the dimension of the output word vector.

[0070] Step S1033: Calculate the average value of the word vectors corresponding to each word to determine the hidden layer vector h, and multiply the hidden layer vector h by the output weight matrix W'[N*V] to obtain the vector y, where the size of the vector y is [1*V]. Then, process the vector y using the softmax activation function to obtain the V-dim probability distribution.

[0071] Step S1034: The input word for each industry product name is an ont-hot encoding, that is, each dimension represents a chemical product name. In the V-dim probability distribution, the chemical product referred to by the index with the highest probability is the predicted middle word. And by continuously adjusting the model error, the word vectors of each word of the chemical product name are updated at the same time as W.

[0072] Step S1035: Store the model after training is completed and the word vectors of each word in all chemical product names, so as to use the word vectors of each word as the second word vector.

[0073] After determining the second word vector corresponding to the preset material name, the target material name is determined based on the first word vector and the second word vector, thereby completing the acquisition of data and / or information from the target table.

[0074] In some embodiments, determining the target material name based on the first word vector and the second word vector includes: determining a second word vector that matches each of the first word vectors; and determining a preset material name corresponding to the matched second word vector as the target material name.

[0075] For example, the first word vector is matched one by one to determine the second word vector matched by each first word vector, and the preset material name corresponding to the matched second word vector is determined as the target material name corresponding to the first word vector, that is, the information corresponding to the entity in the cell where the first word vector is located, thereby completing the extraction of table information.

[0076] In some embodiments, determining the second word vector that matches each of the first word vectors includes: determining the cosine similarity between the first word vector and each of the second word vectors based on a cosine similarity calculation model; and determining the second word vector whose cosine similarity is greater than or equal to a cosine similarity threshold as the second word vector that matches the first word vector.

[0077] For example, based on the cosine similarity calculation model, cosine similarity is calculated for each first word vector and each second word vector to determine the cosine similarity between each first word vector and each second word vector. The second word vector with a cosine similarity greater than or equal to the cosine similarity threshold is determined as the second word vector that matches the first word vector. In some specific implementations, after determining the cosine similarity between a certain first word vector and each second word vector, the second word vector with the largest cosine similarity is taken as the second word vector that matches the first word vector. Thus, the target material name can be obtained based on the preset material name corresponding to the second word vector with the largest cosine similarity as the entity recognition result of the first word vector.

[0078] Step S104: Determine a first target cell in the target table that records the name of the target material, and a second target cell adjacent to the first target cell.

[0079] For example, after determining the target material name, a first target cell is determined based on the cell where the target material name is located in the target table. That is, the first target cell is the cell that records the target material name. After determining the first cell, the cell adjacent to the first target cell is determined as the second target cell, so as to obtain the data recorded in the second target cell and complete the compilation of the life cycle unit process data list of the target product based on the target material name and the data recorded in the second target cell.

[0080] Step S105: If the second target cell meets the content acquisition conditions based on the recognition result obtained by data type recognition of the second target cell, the target data in the second target cell is acquired.

[0081] For example, after determining the second target cell, it is determined whether the second target cell meets the content acquisition conditions based on the data type of the second target cell, and if the second target cell meets the content acquisition conditions, the target data recorded in the second target cell is acquired, so as to improve the efficiency of compiling the product life cycle unit process data list.

[0082] In some embodiments, the step of obtaining target data in the second target cell when the second target cell meets the content acquisition conditions based on the identification result obtained by identifying the data type of the second target cell includes: determining the position information of the second target cell when the identification result of the data type of the second target cell includes single-precision floating-point type; and obtaining the target data in the second target cell when the second target cell meets the content acquisition conditions based on the position information of the second target cell.

[0083] In the specific implementation process, the data type of the second target cell is identified. If the identification result shows that the data type of the second target cell is single-precision floating point, the target data recorded in the second target cell is obtained.

[0084] In some embodiments, the step of obtaining target data in the second target cell when the second target cell meets preset position conditions based on the position information of the second target cell includes: when the second target cell and the first target cell are in a first adjacent relationship, determining the column where the second target cell is located as a first target column in the table where the second target cell is located; determining the cell at the target position in the first target column as a third target cell, and performing character recognition processing on the third target cell to obtain the keyword in the third target cell; and obtaining the target data in the second target cell when the keyword in the third target cell matches the preset keyword.

[0085] For example, the first adjacency relationship is a left-right adjacency relationship. Specifically, if the second target cell is left-right adjacent to the first target cell, the column where the second target cell is located is determined as the first target column, and the cell at the top of the first target column in the target table is determined as the third target cell. Character recognition processing is performed on the third target cell to obtain the keywords in the third target cell. If the keywords in the third target cell match the preset keywords, it is determined that the second target cell meets the preset position conditions, and the target data recorded in the second target cell is obtained.

[0086] The following explanation uses Table 1, a summary table of pollutant emissions from a factory, as an example.

[0087] Summary Table 1 of Pollutant Emissions from the Entire Plant

[0088]

[0089]

[0090] In the specific implementation process, if the cell containing nitrogen oxides is identified as the first target cell, and the cell adjacent to the first target cell is identified as the second target cell, it should be understood that if the identification result of the second target cell directly above and below the first target cell is not a single-precision floating-point number, but the identification result of the second target cell horizontally adjacent to the first target cell is a single-precision floating-point number, then it is determined that the second target cell directly above and below the first target cell does not meet the content acquisition conditions, and the second target cell horizontally adjacent to the first target cell is processed in the next step.

[0091] After determining that the data type of the second target cell, which is adjacent to the first target cell on the left and right, is a single-precision floating-point type, the column containing the second target cell is determined as the first target column, and the cell at the top of the first target column is determined as the third target cell, that is, the cell in Table 1 that records "emissions of this project" is the third target cell.

[0092] The third target cell is processed for character recognition to obtain keywords in the third target cell, including but not limited to emissions and emissions amount; if the preset keywords also include emissions, it is determined that the keywords in the third target cell match the preset keywords, and the target data recorded in the second target cell which is adjacent to the first target cell on the left and right is obtained: 42.4, thereby determining the emissions of nitrogen oxides.

[0093] In other embodiments, when the second target cell and the first target cell are in a second adjacent relationship, in the target table where the second target cell is located, the row where the second target cell is located is determined as the first target row; the cell at the target position in the first target row is determined as the fourth target cell, and character recognition processing is performed on the fourth target cell to obtain the keyword in the fourth target cell; if the keyword in the fourth target cell matches the preset keyword, the target data in the second target cell is obtained.

[0094] For example, the second adjacency relationship is an upper and lower adjacency relationship. Specifically, if the second target cell is upper and lower adjacency to the first target cell, the row where the second target cell is located is determined as the first target row, and the cell at the beginning of the first target row is determined as the fourth target cell. Character recognition processing is performed on the fourth target cell to obtain the keywords in the fourth target cell. Then, the keywords in the four target cells are matched with preset keywords. After a successful match, the target data in the second target cell is obtained.

[0095] The following explanation uses Table 2, a summary table of pollutant emissions from a factory, as an example.

[0096] Summary Table 2 of Pollutant Emissions from the Entire Plant

[0097]

[0098] In the specific implementation process, if the cell containing sulfur dioxide is identified as the first target cell, and the cell adjacent to the first target cell is identified as the second target cell, it should be understood that the identification result of the second target cell adjacent to the first target cell to the left and right is not a single-precision floating-point number, while the identification result of the second target cell adjacent to the first target cell vertically is a single-precision floating-point number. Therefore, the second target cell adjacent to the first target cell vertically is determined to meet the content acquisition condition, and the next step of processing is carried out on the second target cell adjacent to the first target cell vertically.

[0099] After determining that the data type of the second target cell, which is adjacent to the first target cell, is single-precision floating-point, the column where the second target cell is located is determined as the first target row, and the cell at the beginning of the first target row is determined as the fourth target cell, that is, the cell containing "organized emissions" and the cell containing "unorganized emissions" are the fourth target cells.

[0100] The fourth target cell undergoes character recognition processing to obtain keywords including, but not limited to, "organized," "unorganized," and "emissions." If the preset keywords also include "emissions," then the keywords in the fourth target cell match the preset keywords. The data recorded in the second target cell, which is in the same column as the first target cell and in the same row as the fourth target cell, is then retrieved to determine the sulfur dioxide emissions. It should be understood that in this table, the second target cells include cells recording 0.623 and cells recording 0; therefore, the sulfur dioxide emissions are the sum of the values ​​recorded in these two cells. It should be noted that the number of second target cells can be more or less, depending on the specific table format, and this application does not limit this.

[0101] In some embodiments, when the second target cell is determined to meet the content acquisition conditions based on the recognition result obtained by data type recognition of the second target cell, the method further includes: based on the entity recognition model, determining the second target column by recognizing the entity of the target table according to the preset string, and determining the unit of measurement recorded in the cell in the second target column that is in the same row as the second target cell as the unit of measurement corresponding to the target data, wherein all cells in the second target column are used to record the unit of measurement.

[0102] For example, in the process of compiling a product lifecycle data list, in addition to obtaining the value of a certain compound emission, it is also necessary to obtain the corresponding unit of measurement to obtain accurate data. The following uses Table 3, a summary table of pollutant emissions from a factory, as an example to illustrate this.

[0103] Summary Table 3 of Pollutant Emissions from the Entire Plant

[0104] Pollutant Name unit Emissions from this project Sulfur dioxide ton 9.422 nitrogen oxides ton 42.4 Volatile organic compounds ton 55.13

[0105] In the specific implementation process, based on the entity recognition model, entities in the target table are recognized according to a preset string, such as "unit," to obtain the corresponding recognition results. If the recognition results include cells filled with unit entities, the column containing that cell is determined as the second target column (e.g., the second column in Table 3 is the second target column), and the cell in the same row as the second target cell is determined. According to the table's filling rules, the unit of measurement filled in that cell is the unit of measurement corresponding to the value in the second target cell, thus completing the determination of the unit of measurement for the target data.

[0106] In other embodiments, based on the entity recognition model, a second target row is determined according to the entity recognition result of the target table based on the preset string, and the unit of measurement recorded in the cell in the second target row that is in the same column as the second target cell is determined as the unit of measurement corresponding to the target data, wherein all cells in the second target row are used to record the unit of measurement.

[0107] It should be understood that the table format in Table 2 can be referred to. In the table format in Table 2, the units of measurement are filled in the target table in the form of rows. The units of measurement corresponding to the target data are determined by a similar processing method as in the embodiment where the units of measurement are filled in the target table in the form of columns. This will not be repeated here.

[0108] In other embodiments, based on a preset unit of measurement standard, string matching processing is performed on the third target cell or the fourth target cell to obtain the unit of measurement corresponding to the target data.

[0109] For example, the unit of measurement can also be filled in the starting cell of a column or the starting cell of a row in a table. For instance, the third target cell in Table 1 includes the unit of measurement "ton", and the fourth target cell in Table 2 includes the unit of measurement "ton". Therefore, by performing string matching processing on the third or fourth target cell with the preset unit of measurement standard, the corresponding unit of measurement can be obtained.

[0110] In the specific implementation process, after determining the third or fourth target cell, the keywords in the third or fourth target cell can be matched with the keywords in the preset unit of measurement standard to determine whether the third or fourth target cell includes a unit of measurement. If it is determined that the third or fourth target cell does not include a unit of measurement, then the string matching processing based on the preset string in the target table is performed as described above to obtain the corresponding unit of measurement, thereby improving the accuracy of the product lifecycle process data list.

[0111] Step S106: Generate a lifecycle unit process data list for the target product based on the target material name and the target data.

[0112] For example, a lifecycle unit process data list for the target product is generated based on the determined target material name, target data, and the correspondence between the target material name and the target data. For instance, the lifecycle unit process data list for the target product is generated based on the determined nitrogen oxides and the corresponding target data: 42.4. It should be understood that this lifecycle unit process data list for the target product includes information that the nitrogen oxide emission is 42.4, where nitrogen oxides are the pollutants corresponding to the lifecycle of the target product, nitrogen oxides are used to characterize the target material name, and 42.4 is the emission data corresponding to the nitrogen oxides.

[0113] In some embodiments, the method further includes: obtaining a document to be mined, the document to be mined being used to determine the material corresponding to the life cycle unit process of the target product; generating a life cycle unit process data list of the target product based on the target material name and the target data, including: generating a life cycle unit process data list of the target product based on the material name obtained from the document to be mined, the target material name and the target data.

[0114] For example, the documents to be mined include, but are not limited to, product environmental impact assessment reports, in order to identify the materials corresponding to the life cycle unit processes that generate the target product from the documents to be mined.

[0115] In the specific implementation process, based on the TF-IDF model, the materials corresponding to the life cycle unit processes of the target product are identified in the documents to be mined. Specifically, the documents to be mined are unstructured texts, which are segmented into line units to obtain a list of unit corpora corresponding to the documents to be mined. Based on a string matching network, corpus units including keywords such as "product", "byproduct", "raw material", and "pollutant" are identified, and Chinese word segmentation is performed using corpus units, thereby enabling the determination of material names based on corpus units containing the above keywords.

[0116] In some embodiments, generating the list of life cycle unit process data of the target product according to the material name obtained from the to-be-mined document, the target material name, and the target data includes: determining the target material name that matches the material name obtained from the to-be-mined document; filling the target material name that matches the material name obtained from the to-be-mined document and the target data corresponding to the matched target material name into a preset list template to obtain the list of life cycle unit process data of the target product.

[0117] Exemplarily, among the multiple target material names determined from the to-be-identified document, determine the name that is the same as the material name determined from the to-be-mined document, determine that the material corresponding to the same name is the material associated with the life cycle unit process of the target product (such as the required material or the emitted pollutant), and determine the target data corresponding to the material of these same names, so as to fill the material of these same names and the corresponding target data into the preset list template to complete the compilation of the list of life cycle unit process data of the target product.

[0118] In some embodiments, the method further includes: performing word segmentation on the to-be-mined document based on a text mining model to obtain multiple corpus units; performing string matching on the multiple corpus units based on preset semantic rules to obtain at least one table name, and establishing the table name corpus according to the table name.

[0119] Exemplarily, the table name corpus in step S101 can also be determined according to the to-be-mined document. Specifically, perform word segmentation on the to-be-mined document to obtain multiple corpus units. Specifically, screen out multiple corpus units with the character "table" as the first character, and use these corpus units as the original corpus. Based on the preset semantic rules, perform string matching on these corpus units and a preset string, where the preset string includes but is not limited to raw and auxiliary material consumption, energy consumption, material consumption, pollutants, waste gas, organized waste gas, unorganized waste gas, wastewater, solid waste, solid waste, and three wastes, to determine the names of various relevant tables for different product synthesis methods in the to-be-mined document, so as to establish the table name corpus. It should be noted that the semantic rules can be set according to the actual situation, and this application does not limit them.

[0120] The entity recognition-based lifecycle unit process data list compilation method provided in the above embodiments utilizes a table name corpus to determine target tables for information extraction in the document to be identified, and identifies entities in the target tables based on an entity recognition model to obtain multiple target material names. It also identifies second target cells in the target tables that meet the content retrieval conditions, and obtains the target data in the second target cells. Based on the target material names and the obtained target data, the product lifecycle unit process list is compiled. By extracting entity information from the document to be identified using an entity recognition model, the automatic compilation of the product lifecycle unit process list is achieved, the efficiency and accuracy of lifecycle unit process data collection are improved, and the data processing cost is reduced.

[0121] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0122] It should also be understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. It should be noted that, herein, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0123] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. The above descriptions are merely specific implementations of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered 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 compiling a lifecycle unit process data inventory based on entity recognition, characterized in that, Applied to terminal devices, the method for compiling a lifecycle unit process data list based on entity recognition includes: Obtain the document to be identified and the table name corpus, wherein the table name corpus includes various table names; The target table is determined in the document to be identified based on the table names in the table name corpus. The target table contains life cycle unit process data of the target product. Based on the entity recognition model, the entities in the target table are identified according to the preset material names to obtain multiple target material names; In the target table, a first target cell is determined that records the name of the target material, and a second target cell is determined that is adjacent to the first target cell; If the second target cell meets the content acquisition conditions based on the identification result obtained by identifying the data type of the second target cell, the target data in the second target cell is acquired. Generate a lifecycle unit process data list for the target product based on the target material name and the target data; Wherein, the step of obtaining the target data in the second target cell when the identification result obtained from the data type identification of the second target cell determines that the second target cell meets the content acquisition conditions includes: If the identification result of the data type of the second target cell includes single-precision floating-point type, determine the position information of the second target cell; When the second target cell and the first target cell are in a first adjacent relationship, the column containing the second target cell in the table is determined as the first target column; The cell at the target position in the first target column is identified as the third target cell, and character recognition processing is performed on the third target cell to obtain the keywords in the third target cell; If the keyword in the third target cell matches a preset keyword, retrieve the target data from the second target cell; or If the identification result of the data type of the second target cell includes single-precision floating-point type, determine the position information of the second target cell; When the second target cell and the first target cell are in a second adjacent relationship, the row containing the second target cell in the target table is determined as the first target row; The cell at the target position in the first target row is identified as the fourth target cell, and character recognition processing is performed on the fourth target cell to obtain the keywords in the fourth target cell; If the keyword in the fourth target cell matches the preset keyword, the target data in the second target cell is obtained.

2. The method for compiling a lifecycle unit process data list based on entity recognition as described in claim 1, characterized in that, When the method determines that the second target cell meets the content acquisition conditions based on the recognition result obtained by data type recognition of the second target cell, the method further includes: Based on the entity recognition model, the entities in the target table are recognized according to a preset string to determine the second target column. The unit of measurement recorded in the cell in the second target column that is in the same row as the second target cell is determined as the unit of measurement corresponding to the target data. All cells in the second target column are used to record the unit of measurement; or Based on the entity recognition model, according to the entity recognition result of the target table based on the preset string, a second target row is determined, and the unit of measurement recorded in the cell in the second target row that is in the same column as the second target cell is determined as the unit of measurement corresponding to the target data, wherein all cells in the second target row are used to record the unit of measurement; or Based on a preset unit of measurement standard, string matching is performed on the third target cell or the fourth target cell to obtain the unit of measurement corresponding to the target data.

3. The method for compiling a lifecycle unit process data list based on entity recognition as described in claim 1 or 2, characterized in that, The entity recognition model identifies entities in the target table based on preset material names, resulting in multiple target material names, including: Based on the entity recognition model, the entities in each cell of the target table are vectorized to obtain the first word vector corresponding to each entity in each cell; Determine the second word vector corresponding to each of the preset material names; The target material name is determined based on the first word vector and the second word vector.

4. The method for compiling a lifecycle unit process data list based on entity recognition as described in claim 3, characterized in that, Determining the target material name based on the first word vector and the second word vector includes: Determine the second word vector that matches each of the first word vectors; The preset material name corresponding to the matched second word vector is determined as the target material name.

5. The method for compiling a lifecycle unit process data list based on entity recognition as described in claim 4, characterized in that, The step of determining the second word vector that matches each of the first word vectors includes: Based on the cosine similarity calculation model, the cosine similarity between the first word vector and each of the second word vectors is determined; The second word vector whose cosine similarity is greater than or equal to the cosine similarity threshold is determined as the second word vector that matches the first word vector.

6. The method for compiling a lifecycle unit process data list based on entity recognition as described in claim 1 or 2, characterized in that, The method further includes: Obtain the document to be mined, which is used to determine the materials corresponding to the life cycle unit process of the target product; Generate a lifecycle unit process data list for the target product based on the target material name and the target data, including: A lifecycle unit process data list for the target product is generated based on the material name obtained from the document to be mined, the target material name, and the target data.

7. The method for compiling a lifecycle unit process data list based on entity recognition as described in claim 6, characterized in that, The step of generating a lifecycle unit process data list for the target product based on the material name obtained from the document to be mined, the target material name, and the target data includes: Identify the target material name that matches the material name obtained in the document to be mined; The target material name that matches the material name obtained in the document to be mined, and the target data corresponding to the matched target material name are filled into the preset list template to obtain the life cycle unit process data list of the target product.

8. The method for compiling a lifecycle unit process data list based on entity recognition as described in claim 6, characterized in that, The method further includes: Based on the text mining model, the document to be mined is segmented into words to obtain multiple corpus units; Based on preset semantic rules, string matching processing is performed on multiple corpus units to obtain at least one table name, and the table name corpus is established based on the table name.

Images