A smart mapping method and data processing device for bill of quantities data

By using intelligent mapping methods and employing classification rule trees and semantic vector retrieval technology, the problem of heterogeneity in bill of quantities data in engineering cost management was solved, improving mapping efficiency and accuracy, and reducing mapping failures and manual intervention.

CN123115985APending Publication Date: 2026-11-17SHANGHAI SHANGZI ENG ZAOJIA CONSULTATION CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611376507.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-07
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies in engineering cost management struggle to effectively handle the heterogeneity and non-standard descriptions of bill of quantities data from different projects and stages, leading to mapping failures or mismatches and reducing the efficiency of bill of quantities data mapping processing.

Method used

The intelligent mapping method is adopted. By extracting the project feature text, unit of measurement and sub-item hierarchical name, the initial matching is performed using the classification rule tree. Then, the semantic vector retrieval is transferred and the hierarchical weight verification is introduced. The supplementary subject code input by the user is received, and feature words are extracted from the difference between the project feature text and the standard text. These feature words are then classified and added to the inclusion or exclusion word set of the classification rule tree, and mapping experience is continuously accumulated.

Benefits of technology

It improves the efficiency of inventory data mapping processing, reduces the probability of mapping failure, and lowers the time overhead of rule matching and the need for manual intervention.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115985A_ABST
    Figure CN123115985A_ABST
Patent Text Reader

Abstract

A smart mapping method and data processing device for bill of quantities data, relating to the field of computer technology, is disclosed to improve the efficiency of mapping large amounts of bill of quantities data. In this method, the data processing device extracts the project feature text, unit of measurement, and sub-item hierarchical names of bill of quantities entries. Based on the included and excluded word sets of the classification rule tree nodes, it performs rule matching on the word segmentation results of the project features. When a match fails, semantic vector retrieval is initiated, using hierarchical weights to correct the professional similarity of candidate subjects, and filtering candidates with inconsistent measurement dimensions through standard unit of measurement verification. Entries that still cannot be automatically mapped are manually supplemented with subject codes. Simultaneously, feature words are extracted by comparing project features with standard text, updating the word set of the corresponding node in the classification rule tree, and continuously iterating the rule base to reduce the probability of subsequent matching failures for similar entries, thereby improving the efficiency of batch mapping of bill of quantities data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an intelligent mapping method and data processing device for bill of quantities data. Background Technology

[0002] In the field of construction cost management, cost engineers use the bill of quantities (BOQ) as the core carrier to carry out full-process investment control. This requires mapping the BOQ data from each stage of the same project—from estimation, preliminary budget, bidding, contract to settlement—as well as the BOQ data between different projects, to a unified standard subject system for horizontal comparison. Because the BOQ entries for different projects and stages are independently compiled by each participating party, there are significant differences in the descriptive text for the same engineering entity in terms of word order, symbols, specifications, and abbreviations. Therefore, it is necessary to map these scattered and heterogeneous BOQ texts to a unified standard subject.

[0003] Related technologies typically employ an automated processing scheme combining rule matching and keyword retrieval to automate the alignment and mapping of batch list data. Specifically, a rule base corresponding to standard items, standard keywords, and regular expressions is pre-established. When heterogeneous list text is input, the text string is automatically extracted, and the pre-defined rule base is traversed for character comparison or regular expression validation. When the string features of the input text match the conditions in the rule base, the corresponding standard item is output, thereby achieving automated alignment and mapping of batch list data.

[0004] However, with the expansion of engineering projects and the popularization of multi-business complexes, the same standard subject has generated a large number of synonymous and non-standard descriptions and abbreviations in different projects. The preset rule base of related technologies is difficult to exhaust and cover the complex semantic changes and long-tail combinations in multi-source unstructured text. A large number of non-standard description texts cannot be matched with fixed rules, resulting in mapping failure or mismatch, which reduces the efficiency of mapping and processing a large amount of list data. Summary of the Invention

[0005] This application provides an intelligent mapping method and data processing device for bill of quantities data, which can improve the efficiency of mapping and processing large amounts of bill of quantities data.

[0006] Firstly, an intelligent mapping method for bill of quantities data is provided, applied to data processing equipment. This method includes: extracting the project feature text, unit of measurement, and sub-item hierarchical names of the bill of quantities entries to be processed; traversing each node in a preset classification rule tree, where each node is configured with a standard subject code and corresponding inclusion and exclusion word sets; when the word segmentation set of the project feature text covers all inclusion word sets of a node and does not contain any exclusion word set of that node, the node is determined as the target node, and the standard subject code corresponding to the target node is output; when the project feature text fails to match any node in the classification rule tree, the project feature text is converted into a feature vector, and the hierarchical weights corresponding to the sub-item hierarchical names are obtained; based on the hierarchy... The system calculates the weighted similarity between the feature vector of the project feature text and the candidate text vector in the preset standard subject library, and selects target candidate subjects with a weighted similarity greater than a preset threshold. If the standard unit of measurement bound to the target candidate subject is consistent with the extracted unit of measurement, the bill of quantities item to be processed is mapped to the standard subject code corresponding to the target candidate subject. If the standard unit of measurement bound to the target candidate subject is inconsistent with the extracted unit of measurement or no target candidate subject is selected, the system receives the supplementary subject code input for the bill of quantities item to be processed. The system extracts the difference feature words between the project feature text and the standard text corresponding to the supplementary subject code, and classifies the difference feature words according to a preset strategy to supplement the included word set and excluded word set of the corresponding node of the supplementary subject code in the classification rule tree.

[0007] By adopting the above technical solution, after the data processing equipment extracts the project feature text, unit of measurement, and sub-item hierarchical names of the items in the bill of quantities to be processed, it first performs rule matching on the word segmentation set of the project feature text using the included and excluded word sets configured in the node of the classification rule tree as matching conditions. For items that fail to match the rules, it switches to the semantic vector retrieval process, introduces hierarchical weights to correct the professional affiliation of candidate subjects based on their similarity, and uses the standard unit of measurement as the final verification condition to filter candidate subjects with inconsistent measurement dimensions. For items that still cannot be automatically mapped, it receives supplementary subject codes input by humans, and extracts feature words from the differences between the project feature text and the standard text, classifying and supplementing them into the included and excluded word sets of the corresponding nodes in the classification rule tree. This allows the rule tree to continuously accumulate mapping experience of non-standard descriptions, reducing the probability of subsequent mapping failures for similar items, thereby improving the efficiency of mapping large amounts of bill of quantities data.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the step of traversing each node in the preset classification rule tree specifically includes: extracting the standard sub-names bound to each top-level node in the classification rule tree; calculating the text matching degree between the extracted sub-item level names and each standard sub-name, and determining the top-level node corresponding to the standard sub-name with the highest text matching degree as the guide node; defining all lower-level nodes belonging to the guide node in the classification rule tree as candidate traversal intervals; and traversing each lower-level node within the candidate traversal intervals to match the included word set and excluded word set configured for each lower-level node.

[0009] By adopting the above technical solution, before traversing the classification rule tree, text matching is performed using the sub-item level names of the input entries and the standard sub-item names bound to each top-level node. This limits the traversal range to the candidate traversal interval of the top-level node with the highest matching degree, skipping other subtrees that do not match the professional affiliation of the input entry. This reduces the number of nodes required for each match to the range of subtrees related to the professional affiliation of the input entry. When the number of nodes in the classification rule tree is large, this reduces invalid traversal operations in a single match and lowers the time overhead of rule matching.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, when the word segmentation set of the project feature text covers the entire word set of a certain node and does not contain any of the excluded word sets of that node, the step of determining the node as the target node and outputting the standard subject code corresponding to the target node specifically includes: if there are at least two candidate nodes that satisfy the condition that the word segmentation set of the project feature text covers the entire word set of the candidate node and does not contain any of the excluded word sets of the candidate node, then calculate the character coverage rate of the word set of the candidate node in the project feature text respectively; determine the candidate node with the largest character coverage rate as the pending node; if there are multiple pending nodes with the same character coverage rate, compare the level depth of the pending node in the classification rule tree, determine the pending node with the deepest level depth as the target node, and output the standard subject code corresponding to the target node.

[0011] By adopting the above technical solution, when the word segmentation set simultaneously satisfies the matching conditions of the included and excluded word sets of multiple nodes, candidate nodes are filtered sequentially based on character coverage and hierarchical depth. The candidate node with the highest character coverage means that its included word set provides a more comprehensive description of the project feature text, while the node with the deepest hierarchical depth means that its corresponding standard subject classification is more detailed. Through sequential filtering based on two levels of criteria, a unique target node can still be determined even when multiple nodes are matched, avoiding the problem of uncertain mapping results caused by multiple nodes being matched simultaneously, and improving the uniqueness of the rule matching results.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, if the standard unit of measurement bound to the target candidate item is consistent with the extracted unit of measurement, the step of mapping the bill of quantities item to be processed to the standard item code corresponding to the target candidate item specifically includes: mapping the standard unit of measurement bound to the target candidate item and the extracted unit of measurement to a preset physical dimension dictionary respectively to obtain the corresponding standard dimension identifier and extracted dimension identifier. The physical dimension dictionary records a many-to-one mapping relationship between the unit of measurement and the dimension identifier; if the standard dimension identifier and the extracted dimension identifier are the same, it is determined that the unit of measurement is consistent, and the bill of quantities item to be processed is mapped to the standard item code corresponding to the target candidate item.

[0013] By adopting the above technical solution, when verifying the consistency of measurement units, instead of directly performing character comparison on the original unit strings, the standard measurement units and the extracted measurement units are mapped to a physical dimension dictionary, and equivalence is determined by comparing their corresponding dimension identifiers. The physical dimension dictionary records a many-to-one mapping relationship between measurement units and dimension identifiers, allowing units with different writing styles but the same measurement dimension to be judged as consistent, such as m and meter, m² and square meter, etc. This reduces verification misjudgments caused by differences in unit writing styles and lowers the probability of needing manual intervention.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, the step of extracting difference feature words between the project feature text and the standard text corresponding to the supplementary subject code specifically includes: performing lexical analysis on the project feature text and the standard text respectively to obtain the original word sequence and the standard word sequence; comparing the original word sequence and the standard word sequence based on the longest common subsequence algorithm, and extracting the word sequence of the original word sequence that did not match as a candidate difference word set; removing stop words and modifiers from the candidate difference word set, and determining the retained core nouns as difference feature words.

[0015] By adopting the above technical solution, when extracting differential feature words, the project feature text and the standard text are first subjected to lexical analysis, and compared at the unit of word sequence rather than string. Based on the longest common subsequence algorithm, the common part between the two word sequences is identified, and the word elements in the original word sequence that do not enter the common subsequence are determined as the candidate differential word set. Then, the core noun words are retained after filtering by stop words and modifiers. Thus, the text is processed at the word element granularity, which has better robustness to word order changes and format differences. The extracted differential feature words are more focused on the substantive content that reflects the actual differences of the project, and the inclusion of meaningless word elements in the differential word set is reduced.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, the step of classifying and supplementing the differential feature words into the included word set and excluded word set of the supplementary subject code corresponding node in the classification rule tree based on a preset strategy specifically includes: obtaining the peer neighbor nodes of the supplementary subject code corresponding node in the classification rule tree; if differential feature words already exist in the included word set of the peer neighbor nodes, determining that the differential feature words are easily confused features, and supplementing the differential feature words into the excluded word set of the supplementary subject code corresponding node; if differential feature words do not exist in the included word set of the peer neighbor nodes, determining that the differential feature words are unique features, and supplementing the differential feature words into the included word set of the supplementary subject code corresponding node.

[0017] By adopting the above technical solution, before adding differential feature words to the classification rule tree, deduplication is first performed in the word sets of neighboring nodes at the same level. The affiliation of a word is determined based on whether it has already been included in the word sets of neighboring nodes at the same level. Words already included are added to the exclusion word set as easily confused features, while words not included are added to the included word set as unique features. This ensures that the direction of differential feature word inclusion is related to its distribution among neighboring nodes, avoiding the mistaken inclusion of common words among neighboring nodes into the included word set of the target node, thereby reducing the probability of matching conflicts between neighboring nodes during the automatic expansion of the rule tree.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of receiving the supplementary subject code input for the bill of quantities item to be processed if the standard unit of measurement bound to the target candidate subject is inconsistent with the extracted unit of measurement or if no target candidate subject is selected, the method further includes: when a supplementary subject code is received because the standard unit of measurement bound to the target candidate subject is inconsistent with the extracted unit of measurement, extracting the standard text corresponding to the supplementary subject code and the mismatched standard text corresponding to the target candidate subject that caused the unit verification failure; extracting mutually exclusive feature words between the standard text and the mismatched standard text; and supplementing the mutually exclusive feature words to the exclusion word set of the corresponding node of the target candidate subject in the classification rule tree.

[0019] By employing the above technical solution, when a mapping failure is detected due to inconsistent units of measurement, and the operator has already entered a supplementary subject code, the standard text corresponding to the supplementary subject code is further extracted, along with the mismatched standard text corresponding to the target candidate subject that caused the unit verification failure. Mutually exclusive feature words are extracted by comparing the two, and these mutually exclusive feature words are added to the exclusion word set of the erroneous node. This transforms the unit verification failure event into a supplement to the exclusion word set of the erroneous node, enabling the erroneous node to proactively avoid input text with the same mutually exclusive features in subsequent matching, reducing the recurrence of similar unit verification failures.

[0020] In a second aspect, embodiments of this application provide a data processing apparatus, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the data processing apparatus to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a data processing device, cause the data processing device to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a data processing device, cause the data processing device to perform the method described in the first aspect and any possible implementation thereof.

[0023] Understandably, the data processing apparatus provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the methods provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0025] 1. After extracting the project feature text, unit of measurement, and sub-item hierarchical names from the items in the bill of quantities to be processed, the data processing equipment first performs rule matching on the word segmentation set of the project feature text using the included and excluded word sets configured in the node of the classification rule tree as matching conditions. For items that fail to match the rules, the process switches to the semantic vector retrieval process, introducing hierarchical weights to correct the professional affiliation of candidate subjects based on their similarity, and using the standard unit of measurement as the final verification condition to filter candidate subjects with inconsistent measurement dimensions. For items that still cannot be automatically mapped, the system receives supplementary subject codes input manually, extracts feature words from the differences between the project feature text and the standard text, and classifies and supplements them into the included and excluded word sets of the corresponding nodes in the classification rule tree. This allows the rule tree to continuously accumulate mapping experience of non-standard descriptions, reducing the probability of subsequent mapping failures for similar items, thereby improving the efficiency of mapping large amounts of bill of quantities data.

[0026] 2. Before traversing the classification rule tree, text matching is performed using the sub-item level names of the input entries and the standard sub-item names bound to each top-level node. This limits the traversal range to the candidate traversal interval of the top-level node with the highest matching degree, skipping other subtrees that do not match the professional affiliation of the input entry. This reduces the number of nodes required for each match to the range of subtrees related to the input entry's professional affiliation. When the number of nodes in the classification rule tree is large, this reduces invalid traversal operations in a single match and lowers the time overhead of rule matching.

[0027] 3. Before adding differential feature words to the classification rule tree, deduplication is performed on the word sets of neighboring nodes at the same level. The affiliation of a word is determined by whether it has already been included in the word sets of neighboring nodes at the same level. Words already included are added to the exclusion word set as easily confused features, while words not included are added to the included word set as unique features. This ensures that the direction of differential feature word inclusion is related to its distribution among neighboring nodes, avoiding the mistaken inclusion of shared words among neighboring nodes into the included word set of the target node, thereby reducing the probability of matching conflicts between neighboring nodes during the automatic expansion of the rule tree. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating an intelligent mapping method for bill of quantities data in an embodiment of this application.

[0029] Figure 2 This is another flowchart illustrating an intelligent mapping method for bill of quantities data in an embodiment of this application.

[0030] Figure 3 This is a schematic diagram of the physical device structure of a data processing device in the embodiments of this application. Detailed Implementation

[0031] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0032] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0033] This application provides an intelligent mapping method and data processing device for bill of quantities data, which can improve the efficiency of mapping and processing large amounts of bill of quantities data.

[0034] Please see Figure 1 This is a flowchart illustrating an intelligent mapping method for bill of quantities data in an embodiment of this application.

[0035] S101. Extract the project characteristic text, unit of measurement, and sub-item level name of the bill of quantities items to be processed.

[0036] In the bill of quantities, an item refers to the smallest data unit describing a single engineering entity in the engineering cost document. It typically includes fields such as item name, item description, unit of measurement, and quantity. The item description text refers to the natural language description of the engineering entity's attributes such as material, specifications, and construction process, for example, "galvanized steel pipe - DN50 - threaded connection." The unit of measurement refers to the unit of measurement used to measure the quantity of the item, such as m, m... 2 The item level name refers to the professional and sub-item information to which the item belongs in the bill of quantities system. For example, fire protection engineering / water extinguishing system is used to identify the engineering professional category to which the item belongs.

[0037] Specifically, the data processing equipment locates the target list item from the input bill of quantities file, sequentially parses and extracts the corresponding project feature description field, unit of measurement field, and hierarchical path information of the sub-item, storing them as project feature text, unit of measurement, and sub-item hierarchical name, respectively. For structured list files with standardized formats, each field can be directly read by mapping the column header. For unstructured text, regular expressions or named entity recognition models can be used to extract the above information.

[0038] In some embodiments, project feature text, units of measurement, and sub-item hierarchical names can be extracted in various ways. Optionally, for structured spreadsheet-formatted list files, the column header identifiers in each column of the table can be identified first, and a mapping relationship between column header identifiers and field types can be established. Then, the list data can be traversed row by row, and the cell contents of the project feature column, unit column, and sub-item hierarchical path column in each row can be extracted according to the mapping relationship. Finally, the extracted content is cleaned of whitespace characters and the encoding is standardized to obtain standardized project feature text, units of measurement, and sub-item hierarchical names. Optionally, for unstructured PDF or image-formatted list files, the file content can be converted into a processable text string using optical character recognition technology. Then, a pre-trained sequence labeling model can be used to perform entity recognition and boundary localization on the project feature descriptions, unit of measurement markers, and hierarchical path keywords in the text. Finally, the recognition results are classified according to field types to obtain project feature text, units of measurement, and sub-item hierarchical names.

[0039] S102. Traverse each node in the preset classification rule tree. The nodes of the classification rule tree are configured with standard subject codes and the included and excluded word sets corresponding to the standard subject codes.

[0040] The classification rule tree refers to a standard subject classification system organized in a tree-like data structure. The top-level node corresponds to the major engineering category, lower-level nodes are refined layer by layer along professional divisions and sub-divisions, and leaf nodes correspond to specific standard subjects. The standard subject code is a unique identifier for a specific standard subject, uniformly defined by the standard subject system. The included word set is the set of necessary feature words for the standard subject corresponding to the node; the project feature text must cover all words in the included word set to match the node. The excluded word set is the set of prohibited feature words used by the node to distinguish similar subjects; if the project feature text contains any word from the excluded word set, the node will not match.

[0041] Specifically, the data processing device visits each node layer by layer from the top-level node downwards according to the hierarchical structure of the classification rule tree. For each node, it reads the configured inclusion and exclusion word sets and performs matching judgments with the project feature text extracted in step S101. The traversal of the classification rule tree can be further implemented through dynamic pruning. First, the standard sub-node names bound to each top-level node are extracted. The text matching degree between the sub-item hierarchical names extracted in step S101 and each standard sub-node name is calculated. The top-level node with the highest matching degree is determined as the guiding node. The traversal range is then limited to the candidate traversal interval formed by all the subordinate nodes of this guiding node. Matching of the inclusion and exclusion word sets is only performed within this interval, avoiding invalid traversal across professional nodes and thus improving traversal efficiency.

[0042] S103. When the word segmentation set of the project feature text covers all the included word sets of a certain node, and does not contain any excluded word sets of that node, the node is determined as the target node, and the standard subject code corresponding to the target node is output.

[0043] Here, the word segmentation set refers to the set of tokens obtained after lexical segmentation of the project feature text. For example, segmenting "galvanized steel pipe-DN50-threaded connection" yields {galvanized, steel pipe, DN50, threaded connection}. "Covering the entire included word set" means that the word segmentation set of the project feature text includes every token in the included word set of a given node. "Excluding any excluded word set" means that the word segmentation set of the project feature text does not contain any token from the excluded word set of that node.

[0044] Specifically, the project feature text extracted in step S101 is segmented to obtain a segmented set. Then, for each node accessed in step S102, two checks are performed sequentially: First, it is checked whether the segmented set completely contains the node's included word set. Second, it is checked whether the segmented set does not contain any word from the node's excluded word set. If both checks pass, the node is identified as the target node, and its corresponding standard subject code is output, completing the mapping of this list item. When at least two candidate nodes simultaneously meet the above matching conditions, the character coverage rate of each candidate node's included word set in the project feature text is calculated, and the node with the highest character coverage rate is selected as the pending node. If multiple pending nodes with the same character coverage rate still exist, the depth of each pending node in the classification rule tree is compared, and the node with the deepest depth is identified as the target node, and its corresponding standard subject code is output, thus achieving unique node determination when multiple nodes are matched.

[0045] S104. When the project feature text fails to match all nodes in the classification rule tree, the project feature text is converted into a feature vector, and the hierarchical weights corresponding to the sub-item hierarchical names are obtained.

[0046] In this context, "matching failure" refers to the situation where, during the traversal in steps S102 to S103, the word segmentation set of the project feature text fails to meet the matching conditions of any node in the classification rule tree. A feature vector is a numerical vector obtained by mapping the project feature text to a high-dimensional real-number space using text vectorization methods; it is used to measure the semantic similarity between texts in the vector space. Hierarchical weights are pre-configured weight coefficients based on the hierarchical names of the sub-items; these are used to assign different weights to candidate subjects belonging to the same professional level in subsequent similarity calculations, reflecting the degree of influence of professional affiliation on the mapping results.

[0047] Specifically, when no target node meeting the conditions is found after the traversal of steps S102 to S103, the rule matching for this list item is deemed to have failed, and the process transitions to the vector retrieval process. The data processing device calls a pre-trained text vectorization model to encode the item feature text into a fixed-dimensional feature vector. Simultaneously, based on the sub-item hierarchical names extracted in step S101, the device queries the preset hierarchical weight configuration table for the corresponding weight coefficient, preparing for subsequent weighted similarity calculations. The introduction of hierarchical weights allows similarity calculations to prioritize candidate subjects with consistent professional affiliations, reducing the probability of cross-professional mismatches.

[0048] In some embodiments, the vectorization of project feature text can be achieved in several ways. Optionally, a vectorization method based on a pre-trained language model can be used. The project feature text is input into a BERT-like model fine-tuned on an engineering cost corpus, and the output of the [CLS] position of the last layer of the model is extracted as the feature vector of the text. This method can capture the contextual semantic information in the text and has good expressive power for engineering description texts containing professional compound words. Alternatively, a vectorization method based on TF-IDF weighted word vectors can be used. First, a pre-trained word vector model is used to obtain the word vector of each word in the segmented set of the project feature text. Then, the TF-IDF value of each word in the engineering cost corpus is used as the weight to perform a weighted average of the word vectors to obtain the feature vector of the project feature text. This method has low computational cost and is suitable for scenarios with high real-time requirements.

[0049] S105. Based on hierarchical weights, calculate the weighted similarity between the feature vector of the project feature text and the candidate text vector in the preset standard subject library, and select target candidate subjects with a weighted similarity greater than a preset threshold.

[0050] The standard subject library refers to a pre-built database containing all standard subjects and their corresponding standard description texts. Each standard subject in the library has been vectorized and its corresponding candidate text vectors are stored. Candidate text vectors are feature vectors obtained by vectorizing the description text of each standard subject entry in the standard subject library. Weighted similarity is a similarity value obtained by introducing hierarchical weight coefficients on top of the basic vector similarity, used to measure the semantic similarity of the texts while also considering the consistency of professional attribution.

[0051] Specifically, the data processing device traverses all candidate text vectors in the standard subject library. For each candidate subject, it first calculates the basic similarity between the feature vector of the project feature text and the candidate text vector. Then, it adjusts the basic similarity based on the hierarchical weights obtained in step S104 to obtain a weighted similarity. Finally, all candidate subjects with a weighted similarity greater than a preset threshold are output as target candidate subjects for use in subsequent unit of measurement verification steps.

[0052] In some embodiments, similarity can be calculated using multiple weighted methods. Optionally, a multiplicative weighted method can be used, directly multiplying the cosine similarity by the hierarchical weight to obtain a weighted similarity. When the target item and the candidate subject belong to the same professional level, a larger hierarchical weight is used; when they belong to different professional levels, a smaller value is used, thus amplifying the similarity of candidate subjects within the same professional level during calculation. Optionally, an additive bias method can also be used, superimposing a fixed bias value corresponding to the hierarchical weight on top of the cosine similarity. A positive bias is accumulated when the professional levels are the same, and a negative bias is accumulated when the professional levels are different. This bias adjusts the final similarity ranking result, giving priority to candidate subjects within the same professional level during the screening process.

[0053] S106. If the standard unit of measurement bound to the target candidate item is consistent with the extracted unit of measurement, then the bill of quantities item to be processed will be mapped to the standard item code corresponding to the target candidate item.

[0054] Here, the standard unit of measurement refers to the official legal unit of measurement pre-bound to each standard subject entry in the standard subject library. Consistent units of measurement mean that the standard unit of measurement bound to the target candidate subject is equivalent in measurement dimension to the unit of measurement extracted in step S101.

[0055] Specifically, after selecting the target candidate item in step S105, the data processing device extracts the standard unit of measurement bound to the target candidate item in the standard item library and performs a consistency check with the unit of measurement extracted in step S101. The consistency check of the unit of measurement can be further implemented using a preset physical dimension dictionary. That is, the standard unit of measurement and the extracted unit of measurement are mapped to the physical dimension dictionary respectively to obtain their corresponding dimension identifiers. If the dimension identifiers of the two are the same, the unit of measurement is determined to be consistent. The physical dimension dictionary records a many-to-one mapping relationship between units of measurement and dimension identifiers. For example, m and meter are both mapped to length dimension identifiers, thereby eliminating misjudgments caused by different writing forms of units of measurement. After the verification is passed, the standard item code field of the bill of quantities item to be processed is assigned the standard item code corresponding to the target candidate item, completing the mapping of the item.

[0056] In some embodiments, consistency verification of units of measurement can be performed in multiple ways. Optionally, a mapping verification method based on a physical dimension dictionary can be used. A dimension dictionary covering commonly used units of measurement in the field of engineering cost estimation is pre-constructed. The dictionary records the standard symbol, alias, and corresponding dimension category for each unit of measurement. During verification, the dimension categories of the standard unit of measurement and the extracted unit of measurement are queried separately. If the dimension categories are the same, consistency is determined; if any unit is not found in the dictionary, it is considered questionable and transferred to manual processing. Alternatively, a verification method based on unit standardization conversion can be used. A standardization rule set for units of measurement is pre-defined. The extracted unit of measurement and the standard unit of measurement are converted into a unified standard symbol form according to the rule set. Then, the converted standard symbols are subjected to precise string matching. If a match is successful, consistency is determined.

[0057] S107. If the standard unit of measurement bound to the target candidate item is inconsistent with the extracted unit of measurement or if the target candidate item is not selected, then the supplementary item code entered for the bill of quantities item to be processed is received.

[0058] Among them, the supplementary subject code refers to the target standard subject code manually specified by the operator for the pending bill of quantities items that failed to be automatically mapped.

[0059] Specifically, when the unit verification result in step S106 is inconsistent, or when step S105 fails to filter out any target candidate items that meet the threshold conditions, the item in the bill of quantities to be processed is marked as awaiting manual intervention, and the original information of the item is presented to the operator, who then assigns the correct standard item code based on their business experience. The supplementary item code entered by the operator is received and stored by the system, serving as the input basis for updating the rule base in step S108.

[0060] S108. Extract the difference feature words between the project feature text and the standard text corresponding to the supplementary subject code, and classify and supplement the difference feature words into the included word set and excluded word set of the supplementary subject code corresponding node in the classification rule tree based on the preset strategy.

[0061] Here, "Supplementary Subject Code Corresponding Standard Text" refers to the official description text of the standard subject corresponding to the supplementary subject code in the standard subject library. "Distinguishing Feature Words" refers to words in the item feature text that have substantial distinguishing significance compared to the standard text, reflecting the non-standardized descriptive features of the target list items. "Preset Strategy" refers to the classification rules used to determine whether distinguishing feature words should be added to the included or excluded word set.

[0062] Specifically, the standard text corresponding to the supplementary subject codes is extracted and compared with the project feature text extracted in step S101. A difference analysis is performed to identify word sequences that appear in the project feature text but not in the standard text. Stop words and modifiers that lack distinguishing significance are removed, and core nouns are retained as difference feature words. After the difference feature words are extracted, they are categorized and added to the included or excluded word sets of the corresponding nodes of the supplementary subject codes in the classification rule tree according to a preset strategy. This allows the rule tree to be continuously expanded based on manually corrected data, reducing the probability of subsequent mapping failures for similar entries.

[0063] In some embodiments, differential feature words can be extracted in multiple ways. Optionally, an extraction method based on the longest common subsequence algorithm can be used. First, the project feature text and the standard text are processed by word segmentation in the field of engineering cost, respectively, to obtain the original word sequence and the standard word sequence. Then, the longest common subsequence of the two word sequences is calculated, and the word elements in the original word sequence that do not appear in the longest common subsequence are determined as the candidate differential word set. Finally, the candidate differential word set is filtered for stop words and part-of-speech tags, and noun word elements are retained as differential feature words. Optionally, an extraction method based on set difference operation can also be used. First, the project feature text and the standard text are processed by word segmentation, respectively, to obtain the original word set and the standard word set. Then, the difference between the original word set and the standard word set is calculated, and stop words are removed from the difference set. The retained word elements are determined as differential feature words. This method is simple to implement and suitable for short text scenarios that are not sensitive to word order.

[0064] In the above embodiments, after the data processing device extracts the project feature text, unit of measurement, and sub-item hierarchical names of the items in the bill of quantities to be processed, it first performs rule matching on the word segmentation set of the project feature text using the included and excluded word sets configured in the node of the classification rule tree as matching conditions. For items that fail to match the rules, the process proceeds to the semantic vector retrieval process, introducing hierarchical weights to correct the professional affiliation of the candidate subjects based on their similarity, and using the standard unit of measurement as the final verification condition to filter candidate subjects with inconsistent measurement dimensions. For items that still cannot be automatically mapped, the device receives supplementary subject codes input manually, and extracts feature words from the differences between the project feature text and the standard text, classifying and supplementing them into the included and excluded word sets of the corresponding nodes in the classification rule tree. This allows the rule tree to continuously accumulate mapping experience of non-standard descriptions, reducing the probability of subsequent mapping failures for similar items, thereby improving the efficiency of mapping large amounts of bill of quantities data.

[0065] However, in scenarios with dense classification rule tree structures and a large number of peer nodes under the same parent node, the extracted differential feature words may already be included in the vocabulary set of a neighboring peer node. This word is not a unique distinguishing word for the target node relative to its peer nodes. If it is added to the target node's vocabulary set, subsequent input text containing that word will simultaneously hit both the target node and that neighboring node, leading to matching conflicts between nodes and affecting the accuracy of rule tree expansion.

[0066] Please see Figure 2 This is another flowchart illustrating an intelligent mapping method for bill of quantities data in an embodiment of this application.

[0067] S201. Extract the project characteristic text, unit of measurement, and sub-item level name of the bill of quantities items to be processed.

[0068] S202. Traverse each node in the preset classification rule tree. The nodes of the classification rule tree are configured with standard subject codes and the included word set and excluded word set corresponding to the standard subject codes.

[0069] S203. When the word segmentation set of the project feature text covers all the included word sets of a certain node, and does not contain any excluded word sets of that node, the node is determined as the target node, and the standard subject code corresponding to the target node is output.

[0070] S204. When the project feature text fails to match all nodes in the classification rule tree, the project feature text is converted into a feature vector, and the hierarchical weights corresponding to the sub-item hierarchical names are obtained.

[0071] S205. Based on hierarchical weights, calculate the weighted similarity between the feature vector of the project feature text and the candidate text vector in the preset standard subject library, and select target candidate subjects with a weighted similarity greater than a preset threshold.

[0072] S206. If the standard unit of measurement bound to the target candidate item is consistent with the extracted unit of measurement, then the bill of quantities item to be processed will be mapped to the standard item code corresponding to the target candidate item.

[0073] S207. If the standard unit of measurement bound to the target candidate item is inconsistent with the extracted unit of measurement or if the target candidate item is not selected, then the supplementary item code entered for the bill of quantities item to be processed shall be received.

[0074] S208. Extract the difference feature words between the project feature text and the corresponding standard text of the supplementary subject code.

[0075] Step S201 is similar to step S101, step S202 is similar to step S102, step S203 is similar to step S103, step S204 is similar to step S104, step S205 is similar to step S105, step S206 is similar to step S106, step S207 is similar to step S107, and step S208 is similar to step S108. They will not be described again here.

[0076] S209. Obtain the sibling neighbor nodes of the node corresponding to the supplementary subject code in the classification rule tree.

[0077] Among them, neighboring nodes at the same level refer to other nodes in the classification rule tree that are at the same level as the node corresponding to the supplementary subject code and have the same parent node. These nodes belong to the same sub-section of the target node in the engineering classification system and have a high degree of similarity in engineering attributes. Therefore, they are most likely to cause feature confusion with the target node.

[0078] Specifically, the node corresponding to the supplementary subject code is located in the classification rule tree. The parent node information of this node is read, and then all other nodes sharing the same parent node are obtained. These nodes are designated as peer-neighbor nodes for use in the differential feature word classification judgment in steps S210 and S211. Obtaining peer-neighbor nodes is a preliminary step to prevent differential feature words from being blindly included in the database. By comparing the existing configurations of peer nodes horizontally, the classification of differential feature words is based on evidence, avoiding the incorrect inclusion of shared feature words between peer nodes into the target node's word set, thereby reducing the cross-node confusion rate in subsequent matching.

[0079] In some embodiments, sibling neighbor nodes can be obtained in multiple ways. Optionally, a tree-based parent node indexing method can be used. Each node in the classification rule tree maintains a reference to its parent node. The parent node is located directly through the parent node reference of the node corresponding to the supplementary subject code. Then, the entire list of child nodes of the parent node is traversed. After excluding the node corresponding to the current supplementary subject code itself, the remaining child nodes are determined as sibling neighbor nodes. Optionally, a hierarchical coding-based interval retrieval method can also be used. The standard subject codes of each node in the classification rule tree have a hierarchical prefix pattern. The hierarchical prefix of the supplementary subject code is extracted, and all other nodes with the same hierarchical prefix are retrieved in the coding index of the rule tree. The retrieval results are determined as sibling neighbor nodes.

[0080] S210. If there are already differential feature words in the word set of neighboring nodes at the same level, determine that the differential feature words are easily confused features, and add the differential feature words to the exclusion word set of the node corresponding to the supplementary subject code.

[0081] Among them, easily confused features refer to a difference feature word that has been included in the word set of neighboring nodes at the same level. This indicates that the word is not unique in distinguishing nodes at the same level. If it is added to the word set of the target node, it will cause ambiguous matching between the two nodes in the subsequent input text.

[0082] Specifically, for each differential feature word extracted in step S208, the word sets of each peer-neighboring node obtained in step S209 are searched one by one to determine whether the differential feature word already exists in the word set of any peer-neighboring node. If it exists, it indicates that the word is a common feature word among peer-neighboring nodes. Adding it to the word set of the target node will not play a distinguishing role and may even cause matching conflicts. Therefore, the differential feature word is added to the exclusion word set of the node corresponding to the supplementary subject code, so that the target node actively avoids input text containing the word in subsequent matching, thereby reducing the mismatch rate between peer-neighboring nodes.

[0083] S211. If there are no different feature words in the word set of neighboring nodes at the same level, determine that the different feature words are unique features and add the different feature words to the word set of the node corresponding to the supplementary subject code.

[0084] Among them, unique features refer to a certain differential feature word that is not included in the word set of any peer-neighboring nodes. This indicates that the word is a unique distinguishing word of the target node relative to peer-neighboring nodes. Adding it to the word set of the target node can effectively improve the accuracy of subsequent matching.

[0085] Specifically, if the search results in step S210 indicate that the differential feature word does not exist in the word set of any neighboring node at the same level, the word is determined to be a unique feature of the target node. It is then added to the word set of the node corresponding to the supplementary subject code, enriching the node's matching capability. This allows subsequent similar list entries containing the word to directly match the node through rule matching, reducing the probability of switching to vector retrieval due to rule matching failure. Steps S210 and S211 together implement the differential feature word classification and supplementation strategy. By distinguishing between easily confused features and unique features, the rule tree avoids matching conflicts between nodes caused by blindly adding entries during automatic expansion.

[0086] In some embodiments, when manual intervention is triggered in step S207 due to the inconsistency between the standard unit of measurement bound to the target candidate subject and the extracted unit of measurement, and a supplementary subject code input by the operator is received, the standard text corresponding to the supplementary subject code and the mismatched standard text corresponding to the target candidate subject that caused the unit verification failure can be further extracted. A word sequence comparison is performed on both, and words that can distinguish the standard subjects to which the two belong are extracted from the standard text relative to the mismatched standard text. These words are identified as mutually exclusive feature words, and the mutually exclusive feature words are added to the exclusion word set of the corresponding node of the target candidate subject in the classification rule tree. In subsequent matching, this erroneous node will actively exclude input text containing mutually exclusive feature words, ensuring that entries with similar descriptive features to the standard subject corresponding to the supplementary subject code are no longer hit by the erroneous node, reducing the recurrence of unit verification failures caused by combinations of similar feature words.

[0087] In the above embodiments, before supplementing the differential feature words to the classification rule tree, all sibling neighbor nodes corresponding to the supplementary subject code node are first located. For each differential feature word, it is checked whether it already exists in the word set of any sibling neighbor node. Existing words are determined to be easily confused features and are added to the exclusion word set of the target node. Non-existent words are determined to be unique features and are added to the word set of the target node. In addition, when the mapping failure is due to inconsistent units of measurement, mutually exclusive feature words between the standard text corresponding to the supplementary subject code and the text corresponding to the incorrect target candidate subject are further extracted and added to the exclusion word set of the incorrect node, so that the incorrect node can actively avoid input text with the same features in subsequent matching. In this way, the rule tree can determine the direction of differential word entry into the database based on the existing configuration of sibling nodes during the automatic expansion process, reducing the probability of matching conflicts between sibling nodes caused by blind entry into the database, and forming a continuous correction constraint on historical mismatch nodes.

[0088] The above describes an intelligent mapping method for bill of quantities data in the embodiments of this application. The following describes an exemplary data processing device 300 provided in the embodiments of this application.

[0089] Figure 3This is an exemplary hardware structure diagram of the data processing device 300 provided in an embodiment of this application. In some embodiments, the data processing device 300 is a computer device. The computer device includes a processor, a memory, and a network interface connected via a system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores data. The network interface of the computer device is used to communicate with other external terminals or servers via a network connection. In some embodiments, the network interface can be a wired network interface; in some embodiments, the network interface can also be a wireless network interface. When the computer program is executed by the processor, it implements an intelligent mapping method for bill of quantities data according to an embodiment of this application.

[0090] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0091] In some embodiments of this application, a computer-readable storage medium is also provided, including instructions that, when executed on the data processing device 300, cause the data processing device 300 to perform an intelligent mapping method for bill of quantities data according to an embodiment of this application.

[0092] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0093] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0094] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0095] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for intelligent mapping of bill of quantities data, characterized in that, Applied to a data processing device, the method includes: Extract the project characteristic text, unit of measurement, and sub-item hierarchical name of the bill of quantities items to be processed; Traverse each node in the preset classification rule tree, wherein each node of the classification rule tree is configured with a standard subject code and a set of included words and a set of excluded words corresponding to the standard subject code; When the word segmentation set of the project feature text covers all the included word set of a certain node and does not contain any excluded word set of that node, the node is determined as the target node, and the standard subject code corresponding to the target node is output. When the project feature text fails to match any node in the classification rule tree, the project feature text is converted into a feature vector, and the hierarchical weight corresponding to the sub-item hierarchical name is obtained. Based on the hierarchical weights, the weighted similarity between the feature vector of the project feature text and the candidate text vector in the preset standard subject library is calculated, and target candidate subjects with a weighted similarity greater than a preset threshold are selected. If the standard unit of measurement bound to the target candidate item is consistent with the extracted unit of measurement, then the item in the bill of quantities to be processed is mapped to the standard item code corresponding to the target candidate item; If the standard unit of measurement bound to the target candidate item is inconsistent with the extracted unit of measurement or the target candidate item is not filtered out, then the supplementary item code input for the bill of quantities item to be processed is received. Extract the difference feature words between the project feature text and the standard text corresponding to the supplementary subject code, and classify and supplement the difference feature words into the included word set and the excluded word set of the node corresponding to the supplementary subject code in the classification rule tree based on a preset strategy.

2. The method according to claim 1, characterized in that, The steps of traversing each node in the preset classification rule tree specifically include: Extract the standard division name bound to each top-level node in the classification rule tree; Calculate the text matching degree between the extracted sub-item hierarchical names and each of the standard sub-item names, and determine the top-level node corresponding to the standard sub-item name with the highest text matching degree as the guide node; All subordinate nodes belonging to the guiding node in the classification rule tree are defined as candidate traversal intervals; Within the candidate traversal interval, each lower-level node is traversed to match the included and excluded word sets configured for each lower-level node.

3. The method according to claim 1, characterized in that, The step of determining a node as a target node and outputting the standard subject code corresponding to the target node when the word segmentation set of the project feature text covers the entire included word set of a certain node and does not contain any excluded word set of that node specifically includes: If there are at least two candidate nodes that satisfy the condition that the word segmentation set of the project feature text covers all the word sets of the candidate node and does not contain any of the excluded word sets of the candidate node, then the character coverage rate of the word sets of the candidate nodes in the project feature text is calculated respectively. The candidate node with the highest character coverage is identified as the node to be determined. If there are multiple undetermined nodes with the same character coverage, the node with the deepest level in the classification rule tree is determined as the target node, and the standard subject code corresponding to the target node is output.

4. The method according to claim 1, characterized in that, The step of mapping the unprocessed bill of quantities item to the standard item code corresponding to the target candidate item if the standard unit of measurement bound to the target candidate item is consistent with the extracted unit of measurement specifically includes: The standard units of measurement bound to the target candidate subject and the extracted units of measurement are mapped to a preset physical dimension dictionary to obtain the corresponding standard dimension identifier and extracted dimension identifier. The physical dimension dictionary records a many-to-one mapping relationship between units of measurement and dimension identifiers. If the standard dimension identifier is the same as the extracted dimension identifier, it is determined that the units of measurement are consistent, and the item in the bill of quantities to be processed is mapped to the standard subject code corresponding to the target candidate subject.

5. The method according to claim 1, characterized in that, The step of extracting the difference feature words between the project feature text and the standard text corresponding to the supplementary subject code specifically includes: Lexical analysis is performed on the project feature text and the standard text respectively to obtain the original word sequence and the standard word sequence; The original word sequence is compared with the standard word sequence based on the longest common subsequence algorithm, and the word sequence that does not match in the original word sequence is extracted as a candidate difference word set; Stop words and modifiers are removed from the candidate difference word set, and the core nouns retained are determined as the difference feature words.

6. The method according to claim 5, characterized in that, The step of classifying and supplementing the differential feature words into the included word set and the excluded word set of the corresponding node of the supplementary subject code in the classification rule tree based on a preset strategy specifically includes: Obtain the sibling neighbor node of the node corresponding to the supplementary subject code in the classification rule tree; If the difference feature word already exists in the word set of the neighboring nodes at the same level, the difference feature word is determined to be a confusing feature, and the difference feature word is added to the exclusion word set of the node corresponding to the supplementary subject code; If the difference feature word is not present in the word set of the neighboring nodes at the same level, the difference feature word is determined to be a unique feature, and the difference feature word is added to the word set of the node corresponding to the supplementary subject code.

7. The method according to claim 1, characterized in that, After the step of receiving the supplementary subject code input for the bill of quantities item to be processed if the standard unit of measurement bound to the target candidate subject is inconsistent with the extracted unit of measurement or if the target candidate subject is not screened, the method further includes: When the supplementary subject code is received because the standard unit of measurement bound to the target candidate subject is inconsistent with the extracted unit of measurement, the standard text corresponding to the supplementary subject code is extracted and compared with the mismatched standard text corresponding to the target candidate subject that caused the unit verification failure. Extract mutually exclusive feature words between the standard text and the mismatched standard text; The mutually exclusive feature words are added to the exclusion word set of the corresponding node of the target candidate subject in the classification rule tree.

8. A data processing device, characterized in that, The data processing device includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors invoke the computer instructions to cause the data processing device to perform the method as described in any one of claims 1-7.

9. A computer program product containing instructions, characterized in that, When the computer program product is run on a data processing device, the data processing device performs the method as described in any one of claims 1-7.

10. A computer-readable storage medium comprising instructions, characterized in that, When the instructions are executed on a data processing device, the data processing device performs the method as described in any one of claims 1-7.