A project end-to-end monitoring method and system based on artificial intelligence

Abstract

This invention discloses an AI-based method and system for monitoring the entire project process, relating to the field of project monitoring technology. It constructs a document citation graph of the entire project process and identifies each complete citation path from the graph as a cross-reference link. Based on the document content corresponding to the cross-reference link, it calculates the content completeness value and the information gain value of the cited content, thereby calculating the authenticity value of the cross-reference link. The authenticity value of the cross-reference link is used as a credibility correction factor for task completion status identification in project monitoring, assisting in determining whether the task has been truly completed. This allows for a quantitative assessment of the authenticity of cross-reference links, increasing the accuracy of automatic identification of the true status of the entire project. It also enables existing AI monitoring systems to accurately judge and clearly identify risks when processing cross-referenced text, ensuring that the results of the entire project process monitoring match the actual situation and guaranteeing accurate monitoring results.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of project monitoring, and in particular to a project whole-process monitoring method and system based on artificial intelligence. BACKGROUND

[0002] With the in-depth application of artificial intelligence technology in the field of energy and power and the continuous improvement of the digitalization degree of project management, more and more engineering projects introduce an automatic monitoring system based on artificial intelligence to judge whether a task has been completed or a problem has been rectified. The prior art usually analyzes the content of project documents such as project daily reports, meeting minutes, supervision opinions and rectification reports by AI, and judges the task status based on keywords, sentence structures, historical similar expressions and other means. However, in actual engineering projects, a large number of documents do not directly describe the task completion situation, but cross-document and cross-version reference other content through means such as "see the previous version of the daily report", "according to the meeting minutes" and "quote the supervision opinion" to form a "cross-reference chain".

[0003] However, the existing AI system generally regards the reference content as equivalent information for judgment, ignoring a series of problems that may be caused by the reference process. For example, the reference content may only be a partial excerpt of the original text, resulting in the loss of key information; some references are just a repetition of the original document, without adding any actual progress. These problems can cause the AI system to misjudge that the task has been completed, thereby affecting the accuracy of automatic identification of the real status of the project whole process. At present, there is a lack of a method for quantitatively evaluating the authenticity of cross-reference content, resulting in the key technical problems that the existing AI monitoring system is inaccurate in judgment and unclear in risk identification when processing cross-reference texts, thereby causing the results of project whole-process monitoring to be inconsistent with the actual situation, resulting in inaccurate monitoring results. SUMMARY

[0004] The purpose of the present application is to solve the above-mentioned problems and provide a project whole-process monitoring method and system based on artificial intelligence.

[0005] In the first aspect of the present application, a project whole-process monitoring method based on artificial intelligence is first proposed, which comprises:

[0006] Based on the reference language existing in the project documents, a document reference graph of the project whole process is constructed, and each complete reference path is identified from the document reference graph as a cross-reference link;

[0007] According to the document content corresponding to the cross-reference link, the content completeness value and the information gain value of the reference content are calculated; the content completeness value and the information gain value of the reference content are respectively used to measure whether the information contained in the reference segment is complete and retains the original content of the referenced segment, and to measure whether the reference segment introduces new semantic information compared with the referenced segment;

[0008] The truth value of the cross-reference link is calculated based on the content completeness value and the information gain value of the referenced content.

[0009] The authenticity value of cross-reference links is used as a credibility correction factor for task completion status identification in project monitoring, to help determine whether the task has been truly completed.

[0010] Optionally, the steps for calculating the content integrity value are as follows:

[0011] Each node in a cross-reference chain represents a document fragment; for each pair of document fragments in a cross-reference chain... The structure retention rate is calculated by comparing the smallest semantic units. The formula for the calculation is: In the formula, Indicates the first Structural information retention rate in segment references; Represents a document segment The set of semantic units in Represents a document segment The content mentioned in the text corresponds to a set of semantic units; Indicates how much of the original structure is accurately preserved in the quoted section; This indicates the total number of semantic units in the original segment;

[0012] Calculate the product of all structure retention rates in the entire cross-reference link, and use it as the information structure retention rate of the entire cross-reference link;

[0013] Extracting the starting point in the cross-reference chain and the end point Corresponding semantic vector and Calculate the directional projection to obtain the semantic directional consistency between the beginning and end of the link. The calculation formula is as follows: In the formula, The semantic direction consistency between the beginning and end of the link;

[0014] Multiply the information structure residual rate of the entire cross-reference link by the information structure residual rate of the entire cross-reference link, and use the result of the multiplication as the content integrity value.

[0015] Optionally, the calculation steps for the information gain value are as follows:

[0016] For the initially referenced document fragment and the ultimately referenced document fragment in the cross-reference link, extract the corresponding semantic vectors; perform discrete semantic block division on these two vectors, dividing each vector into several semantic sub-blocks according to the nesting position;

[0017] Calculate the similarity between each semantic sub-block corresponding to the semantic vector of the final referenced document fragment and each semantic sub-block corresponding to the semantic vector of the initial referenced document fragment, and compare the similarity with a preset similarity threshold. If the similarity between each semantic sub-block corresponding to the semantic vector of the final referenced document fragment and all semantic sub-blocks corresponding to the semantic vector of the initial referenced document fragment is less than the preset similarity threshold, then the corresponding semantic sub-block is recorded as a newly added semantic information block.

[0018] The percentage of newly added semantic sub-blocks is obtained by dividing the total number of newly added semantic information blocks by the total number of semantic sub-blocks corresponding to the semantic vector of the initially referenced document fragment.

[0019] Calculate the similarity between each pair of all newly added semantic sub-blocks, and subtract the similarity from the value 1 to obtain the information diversity value between the corresponding two newly added semantic sub-blocks. Calculate the average of all information diversity values ​​to obtain the information diversity degree.

[0020] The information gain value is obtained by multiplying the information diversity value by the proportion of newly added semantic sub-blocks.

[0021] Optionally, the steps for calculating the truth value of a cross-reference link based on the content completeness value and the information gain value of the referenced content are as follows:

[0022] A two-dimensional coordinate system is established by calculating the truth value of the cross-reference link based on the content completeness value and the information gain value of the referenced content. The truth value of the cross-reference link calculated based on the content completeness value and the information gain value of the referenced content is mapped to the two-dimensional coordinate system; the coordinate points correspond to the content completeness value and the information gain value of the referenced content, respectively.

[0023] Calculate the distance from the coordinate point to the centroid in the two-dimensional coordinate system to obtain the truth value of the corresponding cross-reference link.

[0024] Optionally, the authenticity value of the cross-reference link can be used as a credibility correction factor for task completion status identification in project monitoring. The steps to assist in determining whether a task has been truly completed are as follows:

[0025] Compare the authenticity value of the cross-reference link with the preset authenticity value threshold. If the authenticity value is not less than the preset authenticity value threshold, it means that the authenticity of the cross-reference link in the entire project process is high and can fully reflect the actual problems. Analyzing the cross-reference link can directly and accurately monitor the entire project process.

[0026] If the accuracy value is less than the preset accuracy value threshold, it indicates that the accuracy of cross-reference links in the entire project process is low and cannot fully reflect the actual problems. The monitoring results are directly judged to be erroneous, an alarm is issued, and manual actual monitoring is carried out.

[0027] In a second aspect of this invention, a project end-to-end monitoring system based on artificial intelligence is proposed, the system comprising:

[0028] Cross-reference link module: Based on the reference language existing in the project documents, construct a document reference graph of the entire project process, and identify each complete reference path from the document reference graph as a cross-reference link;

[0029] Calculation module: Based on the document content corresponding to the cross-reference link, calculate the content integrity value and the information gain value of the referenced content; the content integrity value and the information gain value of the referenced content are used to measure whether the information contained in the referenced fragment completely retains the original content of the referenced fragment and to measure whether the referenced fragment introduces new semantic information compared to the referenced fragment, respectively.

[0030] Authenticity module: Calculates the authenticity value of cross-reference links based on the content completeness value and the information gain value of the referenced content;

[0031] Monitoring module: The authenticity value of cross-reference links is used as a credibility correction factor for task completion status identification in project monitoring, to help determine whether the task has been truly completed.

[0032] Optionally, the computing module includes:

[0033] Structural information retention module: Each node in the cross-reference link represents a document fragment; for each pair of document fragments in the cross-reference link... The structure retention rate is calculated by comparing the smallest semantic units. The formula for the calculation is: In the formula, Indicates the first Structural information retention rate in segment references; Represents a document segment The set of semantic units in Represents a document segment The content mentioned in the text corresponds to a set of semantic units; Indicates how much of the original structure is accurately preserved in the quoted section; This indicates the total number of semantic units in the original segment;

[0034] Information Structure Residual Rate Module: Calculates the product of all structural residual rates in the entire cross-reference link, which is used as the information structure residual rate of the entire cross-reference link;

[0035] Directional Consistency Module: Extracts the starting point in the cross-reference link. and the end point Corresponding semantic vector and Calculate the directional projection to obtain the semantic directional consistency between the beginning and end of the link. The calculation formula is as follows: In the formula, The semantic direction consistency between the beginning and end of the link;

[0036] Content integrity module: Multiply the information structure residual rate of the entire cross-reference link by the information structure residual rate of the entire cross-reference link, and use the result of the multiplication as the content integrity value.

[0037] Optionally, the computing module further includes:

[0038] Semantic sub-block module: Extracts the corresponding semantic vectors from the initially referenced document fragments and the ultimately referenced document fragments in the cross-reference link; performs discrete semantic block operation on these two vectors, dividing each vector into several semantic sub-blocks according to the nesting position;

[0039] The new semantic information block module calculates the similarity between each semantic sub-block corresponding to the semantic vector of the final referenced document fragment and each semantic sub-block corresponding to the semantic vector of the initial referenced document fragment, and compares the similarity with a preset similarity threshold. If the similarity between each semantic sub-block corresponding to the semantic vector of the final referenced document fragment and all semantic sub-blocks corresponding to the semantic vector of the initial referenced document fragment is less than the preset similarity threshold, then the corresponding semantic sub-block is recorded as a new semantic information block.

[0040] The module for the proportion of newly added semantic sub-blocks: divide the total number of newly added semantic information blocks by the total number of semantic sub-blocks corresponding to the semantic vector of the initially referenced document fragment to obtain the proportion of newly added semantic sub-blocks;

[0041] Information Diversity Module: Calculate the similarity between all pairs of newly added semantic sub-blocks, subtract the similarity from the value 1 to obtain the information diversity value between the corresponding two newly added semantic sub-blocks, and average all information diversity values ​​to obtain the information diversity score.

[0042] Information gain module: Multiply the information diversity and the proportion of newly added semantic sub-blocks to obtain the information gain value.

[0043] Optionally, the realism module includes:

[0044] Coordinate System Module: A two-dimensional coordinate system is established by calculating the truth value of the cross-reference link based on the content completeness value and the information gain value of the referenced content. The truth value of the cross-reference link calculated from the content completeness value and the information gain value of the referenced content is mapped to the two-dimensional coordinate system; the coordinate points correspond to the content completeness value and the information gain value of the referenced content, respectively.

[0045] The authenticity value module calculates the distance from a coordinate point in a two-dimensional coordinate system to the centroid, thus obtaining the authenticity value of the corresponding cross-reference link.

[0046] Optionally, the monitoring module includes:

[0047] The first monitoring module compares the authenticity value of the cross-reference link with the preset authenticity value threshold. If the authenticity value is not less than the preset authenticity value threshold, it means that the authenticity of the cross-reference link in the entire project process is high and can fully reflect the actual problems. Analyzing the cross-reference link can directly and accurately monitor the entire project process.

[0048] The second monitoring module: If the authenticity value is less than the preset authenticity value threshold, it means that the authenticity of the cross-reference links in the entire project process is low and cannot fully reflect the actual problems. It directly judges that the monitoring results are wrong, issues an alarm, and conducts actual manual monitoring.

[0049] The beneficial effects of this invention are:

[0050] This invention proposes an AI-based method and system for monitoring the entire project process. It constructs a document citation graph of the entire project process based on citations found in project documents, and identifies each complete citation path from this graph as a cross-reference link. Based on the document content corresponding to each cross-reference link, it calculates the content completeness value and the information gain value of the cited content. Furthermore, it calculates the authenticity value of the cross-reference link based on these values. The authenticity value of the cross-reference link is used as a credibility correction factor for task completion status identification during project monitoring, assisting in determining whether tasks have been truly completed. This allows for a quantitative assessment of the authenticity of cross-reference links, increasing the accuracy of automatic identification of the project's true status throughout the entire process. It also enables existing AI monitoring systems to accurately judge and clearly identify risks when processing cross-referenced text, ensuring that the results of the entire project process monitoring match the actual situation and guaranteeing accurate monitoring results. Attached Figure Description

[0051] The invention will now be further described with reference to the accompanying drawings.

[0052] Figure 1 A flowchart of a project end-to-end monitoring method based on artificial intelligence;

[0053] Figure 2 This is a framework diagram of a project end-to-end monitoring system based on artificial intelligence. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] This invention provides a method for monitoring the entire project process based on artificial intelligence. See also... Figure 1 , Figure 1 A flowchart illustrating a project end-to-end monitoring method based on artificial intelligence, provided as an embodiment of the present invention. The method includes the following steps:

[0056] S1: Based on the referenced language in the project documents, construct a document reference graph for the entire project process, and identify each complete reference path from the document reference graph as a cross-reference link;

[0057] S2: Calculate the content integrity value and the information gain value of the referenced content based on the document content corresponding to the cross-reference link; the content integrity value and the information gain value of the referenced content are used to measure whether the information contained in the referenced fragment completely retains the original content of the referenced fragment and to measure whether the referenced fragment introduces new semantic information compared to the referenced fragment, respectively.

[0058] S3: Calculate the truth value of the cross-reference link based on the content completeness value and the information gain value of the referenced content;

[0059] S4: Use the authenticity value of cross-reference links as a credibility correction factor for task completion status identification in project monitoring to help determine whether the task has been truly completed.

[0060] Based on the embodiments of the present invention, an artificial intelligence-based project full-process monitoring method is provided. Through the above method, the authenticity of cross-reference links can be quantitatively evaluated, the accuracy of automatic identification of the true state of the project throughout the entire process can be increased, the existing AI monitoring system can accurately judge and clearly identify risks when processing cross-reference texts, and the results of project full-process monitoring can be consistent with the actual situation, thus ensuring the accuracy of monitoring results.

[0061] In one embodiment, S1: Based on the referential language present in the project documents, construct a document reference graph for the entire project process, and identify each complete reference path from the document reference graph as a cross-reference link;

[0062] Specifically, in step S1, the various documents involved in the entire project (including daily project reports, meeting minutes, supervision opinions, rectification reports, etc.) are first pre-processed in a structured manner. Each document is automatically divided into document fragment units according to number, title level, or natural paragraph, with each fragment unit as an independent node. Subsequently, the text is matched and identified based on preset citation language rules. These rules include, but are not limited to, phrases with clear directional meanings such as "see ××", "based on ××'s opinion", "see the previous ×× report", and "as before". These citation languages ​​are located using regular expressions and syntax templates to extract citation relationship pairs. Each pair of "citation fragment" and "cited fragment" constitutes a directed edge, ultimately forming a citation graph G(V,E) with document fragments as nodes and citation relationships as directed edges. Multi-level citation chains may be formed between nodes in the graph. For example, "Supervision opinion in March 2023" cites "Rectification log in February 2023", which in turn cites "Meeting minutes in January 2023", thus forming a citation path. Graph search algorithms such as depth-first search or breadth-first search are employed to identify all reference paths from the starting point to the ending point in the graph. Paths where all nodes are valid semantic references are defined as complete reference paths, serving as instances of cross-reference links. This graph structure not only preserves the semantic transmission relationships between documents but also provides a path-level structured foundation for subsequent evaluation of the authenticity of referenced content. Specifically, document reference graph construction and the identification of complete reference paths from the document reference graph as cross-reference links are existing technologies, and more than one method can achieve this; detailed examples and descriptions are omitted here.

[0063] In one embodiment, S2: Calculate the content integrity value and the information gain value of the referenced content based on the document content corresponding to the cross-reference link; the content integrity value and the information gain value of the referenced content are used to measure whether the information contained in the referenced fragment completely retains the original content of the referenced fragment, to measure the degree of ambiguity in the referenced fragment, and to measure whether the referenced fragment introduces new semantic information compared to the referenced fragment.

[0064] In one implementation, a content integrity value is calculated based on the document content corresponding to the cross-reference link. This value is used to measure whether the information contained in the referenced fragment completely retains the original content of the referenced fragment. Specifically, the steps for calculating the content integrity value are as follows:

[0065] For cross-reference links , This indicates the original information document fragment (the starting point of the link). Indicates the final referenced segment (link endpoint), each This indicates a single reference propagation; each node represents a document fragment; all All of these are document fragments generated by document numbering / paragraph segmentation; the link is generated by regular expression matching and graph traversal, among other methods.

[0066] For each pair of document fragments in the cross-reference link The structure retention rate is calculated by comparing the smallest semantic units. The formula for the calculation is: In the formula, Indicates the first Structural information retention rate in segment references; Represents a document segment A collection of semantic units (such as sentences or semantic fragments) in a sentence. Represents a document segment The content mentioned in the text corresponds to a set of semantic units; Indicates how much of the original structure is accurately preserved in the quoted section; This indicates the total number of semantic units in the original segment;

[0067] Calculate the product of all structure retention rates in the entire cross-reference link, and use it as the information structure retention rate of the entire cross-reference link;

[0068] Extracting the starting point in the cross-reference chain and the end point The corresponding semantic vector (which can be generated using models such as BERT). and Calculate the directional projection to obtain the semantic directional consistency between the beginning and end of the link. The calculation formula is as follows: In the formula, This refers to the semantic consistency between the beginning and end of the link. This measures the degree of semantic direction preservation at both ends of the cross-reference link. If the final reference is highly consistent with the original content in terms of semantic direction, then... Approaching 1;

[0069] Multiply the information structure residual rate of the entire cross-reference link by the information structure residual rate of the entire cross-reference link, and use the result of the multiplication as the content integrity value.

[0070] It should be noted that in the process of calculating the complete value of the content, the cross-reference links must first be extracted from the project document. The original document is then structured and divided into multiple document fragment units according to paragraph numbering, heading level, or semantic boundaries. For example, the sentence "Item 2: Basic construction has been completed and is awaiting acceptance" in the "Daily Report of March 15, 2024" is divided into an independent fragment. Then, citation statements are identified by regular expressions and rule matching, such as "See Item 2 of the Daily Report of March 15, 2024" for details, and citation pairs are constructed. Combining all citation relationships, the complete citation path is extracted by a graph traversal algorithm. For example, the "Daily Report of March 17" cites the "Meeting Minutes of March 16", which in turn cites the "Daily Report of March 15", forming a three-segment citation chain. For each pair of adjacent document segments in the link, a syntactic analysis tool is used to segment the cited segment into the smallest set of semantic units, such as the two sentences "basic construction has been completed" and "awaiting acceptance". Semantic analysis is then performed on the cited segment. If only "basic construction has been completed" is retained while "awaiting acceptance" is omitted, then the semantic unit accurately retained by the cited segment is identified as one, thus calculating a structural retention rate of 1 / 2. The structural retention rates of all segments in the entire link are multiplied together. For example, in a three-segment link, the retention rates of three adjacent segments are 0.8, 0.75, and 0.9, respectively, resulting in an overall residual degree of 0.8 × 0.75 × 0.9 = 0.54. Subsequently, a pre-trained language model (such as BERT) is used to encode the first and last segments of the link to obtain an overall semantic vector. By normalizing the vector and calculating the directional projection, the semantic consistency between the first and last segments is obtained. For example, if the first segment vector is "basic construction + acceptance preparation" and the last segment vector is "construction completed, awaiting inspection", and their directions are similar, then the consistency degree can reach 0.88. Finally, the structural residual rate is multiplied by the semantic consistency value and the square root is taken to calculate the content integrity value, such as √(0.54×0.88)=0.689. The above data such as document structure, semantic units, semantic vectors, and citation relationships can all be automatically extracted through natural language processing toolkits, semantic vector models, and rule engines, ensuring that each step is repeatable, executable, and can be implemented in the AI ​​monitoring system.

[0071] It's important to note that the content integrity value is a comprehensive numerical measure used to quantify whether the content of the ultimately referenced document in a cross-reference chain fully retains the structural information and semantic direction of the original document fragment (the starting point of the chain). The value ranges from 0 to 1, with a higher value indicating a higher degree of preservation. This value is primarily used to address the critical issue of "misjudging task completion" in existing AI systems' automatic monitoring of project documents. In actual engineering projects, different documents often indirectly reference each other using phrases such as "see last week's report" or "based on meeting minutes," forming long-chain semantic propagation paths. If the system relies solely on the current document's description when determining task completion, ignoring the fact that its content may have been formed through multiple paraphrases, restates, or excerpts, misjudgments are likely to occur. For example, if the initial original description is "Construction plan delayed due to heavy rain, expected to be completed on March 15th," but the final fragment of the reference chain only writes "Construction expected to be completed," omitting the "delayed" and specific date information, then without considering the missing, ambiguous, or distorted information, the AI ​​system may incorrectly determine that the task has been completed on schedule. The content integrity value was introduced precisely to reveal and measure the existence of this "semantic reduction" phenomenon. It obtains the degree of structural preservation through semantic unit comparison and judges whether information has been distorted or shifted by the semantic vector directions of the starting and ending points, thus providing a specific and comparable integrity value. If the value is small, it indicates that the cited content has either omitted important information (such as progress details, responsible parties, etc.) or has shifted in semantic expression (e.g., "suggested rectification" becomes "rectification completed"). In this case, the AI ​​system needs to treat the link with caution and avoid using it as credible evidence of task completion. Conversely, if the value is close to 1, it indicates that the cited fragment is basically consistent with the original content in terms of information structure and semantic direction, and can be used as a relatively reliable basis for task status judgment. Therefore, the content integrity value not only improves the verifiability and traceability of cross-referenced information but also provides an AI project monitoring system with a credibility correction mechanism based on the authenticity of the link, thereby significantly improving the accuracy of automatic judgment and reducing omissions and misjudgments in risk identification.

[0072] It's important to note that the core advantage of calculating content integrity using the above method lies in its comprehensive integration of two dimensions: "information structure preservation" and "semantic direction consistency." This enables a multi-faceted and multi-layered quantitative assessment of the authenticity of cited fragments, avoiding the risk of misjudgment inherent in traditional methods that rely solely on keyword matching, semantic similarity, or sentence structure features. Traditional methods, when dealing with complex cross-reference scenarios, often only determine whether two texts "appear similar," failing to identify critical issues such as information pruning, meaning reduction, or semantic distortion during the citation process. This is particularly problematic in engineering projects where practices like "quoting only partial content," "deleting unfavorable expressions," and "replacing key states with vague descriptions" are common. This lack of structural examination easily leads to AI misjudgments. In contrast, the structure preservation rate calculation in this solution emphasizes precise coverage of semantic units, revealing the degree of "deletion" and "omission" during the citation process. Semantic direction consistency further analyzes whether the final cited fragment faithfully reflects the original fragment's intended meaning, revealing semantic distortion issues such as "inappropriate restatement" or "directional reversal." More importantly, the results of the two methods are fused through multiplication to form non-linear compression. This automatically weakens the influence of high-risk references with low retention rates or low directional consistency in the final value, making the entire value more sensitive to key outliers and improving the ability to prevent false positives. Therefore, compared with conventional weighted average or simple similarity methods, this scheme not only has higher semantic expression judgment capabilities but also stronger engineering interpretability and risk identification capabilities. It can support more accurate task completion status judgment and truly improve the ability of the project's full-process monitoring system to process complex reference text.

[0073] In one embodiment, based on the document content corresponding to the cross-reference link, an information gain value is calculated for the referenced content. This value measures whether the referenced fragment introduces new semantic information compared to the referenced fragment. Specifically, the steps for calculating the information gain value are as follows:

[0074] The initial document segment (starting point) in a cross-reference chain is denoted as... The final quoted document fragment (end point) is denoted as Extract the corresponding semantic vectors and input them into a pre-trained language model (such as BERT), respectively. and Perform discrete semantic block partitioning on these two vectors, dividing each vector into several semantic sub-blocks (such as topic units or subtask blocks) according to their nesting positions. , In the formula, This represents the semantic vector corresponding to the initial referenced document fragment. A semantic sub-block, This indicates the total number of corresponding semantic sub-blocks; This represents the semantic vector corresponding to the final referenced document fragment. A semantic sub-block, This represents the total number of corresponding semantic sub-blocks; such semantic segmentation can be achieved by extracting CLS vectors at syntactic boundaries using BERT features.

[0075] Calculate the similarity between each semantic sub-block corresponding to the semantic vector of the final referenced document fragment and each semantic sub-block corresponding to the semantic vector of the initial referenced document fragment, and compare the similarity with a preset similarity threshold. If the similarity between each semantic sub-block corresponding to the semantic vector of the final referenced document fragment and all semantic sub-blocks corresponding to the semantic vector of the initial referenced document fragment is less than the preset similarity threshold, then the corresponding semantic sub-block is recorded as a newly added semantic information block.

[0076] The percentage of newly added semantic sub-blocks is obtained by dividing the total number of newly added semantic information blocks by the total number of semantic sub-blocks corresponding to the semantic vector of the initially referenced document fragment.

[0077] Calculate the similarity between each pair of all newly added semantic sub-blocks, and subtract the similarity from the value 1 to obtain the information diversity value between the corresponding two newly added semantic sub-blocks. Calculate the average of all information diversity values ​​to obtain the information diversity degree.

[0078] The information gain value is obtained by multiplying the information diversity value by the proportion of newly added semantic sub-blocks.

[0079] It should be noted that each piece of data involved in the information gain value calculation process can be automatically obtained using existing natural language processing techniques and pre-trained language models, making it highly feasible for engineering implementation. Specifically, semantic vectors are extracted from the document start and end points in the cross-reference link. Language models such as BERT can be used to encode the original text segment to obtain the vector representation of the entire segment; for example, inputting the sentence "This work has been rectified" into the BERT model will extract the entire segment's 768-dimensional CLS vector representation. Subsequently, for semantic block processing, the entire text is segmented according to sentence or topic boundaries (which can be achieved using syntactic analysis or sentence segmentation tools), and each clause is encoded separately to obtain multiple sub-block vectors, forming a set of semantic sub-blocks; for example, "This work has been rectified" and "Rectification measures have been implemented" are divided into two sub-blocks, and CLS vectors are extracted for each; then, the cosine similarity between each semantic sub-block in the end point and all semantic sub-blocks in the start point is calculated. If all are less than a set threshold such as 0.8, the semantic sub-block is marked. This is denoted as newly added semantics. Similarity calculation can be achieved using conventional vector library comparison. The number of all newly added semantic sub-blocks is counted and divided by the number of semantic sub-blocks at the starting point to obtain the proportion of newly added semantics. For example, if there are 4 sub-blocks at the starting point and 3 newly added sub-blocks are identified at the ending point, the proportion is 0.75. Then, the cosine similarity between each pair of these 3 newly added sub-blocks is calculated and 1 is taken to obtain the semantic difference. The average of all differences is then calculated to obtain the information diversity. Finally, the proportion of newly added semantics is multiplied by the diversity to obtain the information gain value. The entire process does not involve manual intervention. All text segmentation, vector extraction, similarity calculation, and difference evaluation can be automatically completed by mature toolchains such as BERT, Spacy, and SentencePiece, ensuring the objectivity, repeatability, and feasibility of engineering deployment of the indicator extraction.

[0080] It's important to note that the information gain value measures whether the final cited document fragment introduces new, non-redundant semantic information compared to the original document fragment it references, thus reflecting whether the cross-reference link semantically conveys substantial content expansion. Its core meaning is that if a cited fragment not only fully inherits the core information of the original text but also supplements additional, topic-related, but non-repetitive semantic content, then the citation link is not merely a simple restatement or transfer, but rather achieves incremental semantic updates within the task context, possessing higher information gain. A higher information gain value indicates that the cited content substantially advances the original content, reflecting the citation's initiative, proactivity, and updateability; therefore, its "authenticity" is also higher. In other words, the AI ​​monitoring system can more reliably determine that the task has indeed made new progress or undergone processing actions. For example, if the original paragraph is "civil construction preparation work has been completed," while the cited paragraph is "construction preparation has been completed and site leveling has begun, with equipment foundation erection proceeding simultaneously," although the first half is repetitive, the second half significantly expands the construction content and adds temporal information. The system should therefore determine that the task status has evolved. Without this metric, the AI ​​system might misjudge "not updated" due to high language repetition rates, thus overlooking potential new behaviors or risks. By combining the "proportion of newly added semantic sub-blocks" and the "diversity among newly added information," this value can eliminate meaningless expansions, linguistic redundancy, or "clichéd" citations, truly identifying citation behaviors with actual incremental value. This enhances the granularity and reliability of task status recognition, solving the problem that traditional keyword or semantic similarity-based mechanisms often fall into the trap of "overfitting repetitive information and ignoring effective increments" when processing cross-referenced text. It significantly improves the AI ​​monitoring system's perception accuracy and responsiveness to the real status of the entire project process.

[0081] It should be noted that the advantage of calculating information gain using the above method is that it can simultaneously measure the semantic increment in citation behavior in a fine-grained and structured manner from two dimensions: "existence of new content" and "richness of new content." This overcomes the misjudgment problems common in traditional methods, such as "surface difference judgment" or "keyword splicing expansion." On the one hand, traditional information gain calculation is often based on methods such as full-text vector differences or TF-IDF word frequency changes. Although these methods have a certain discriminative power in long texts, in engineering project documents, due to the short passages and highly templated sentence structures, simply comparing the overall vector or keywords can easily misjudge empty repetitions or meaningless clichés as useful information. In contrast, this method decomposes the entire passage into multiple relatively independent semantic units through semantic segmentation, avoiding the semantic dilution problem caused by the "averaging" of the overall vector, and can accurately identify every small piece of new content. On the other hand, this method not only determines whether new sub-blocks have appeared, but also extracts their semantic diversity by calculating the similarity between the new sub-blocks. This allows it to assess whether these new contents are truly different and meaningful additions, rather than redundant expressions with different forms but similar meanings. For example, if a supervisor's opinion citation adds three supplementary phrases: "Civil engineering acceptance completed," "Hidden works filing submitted," and "Equipment foundation construction started," these three phrases represent different construction nodes semantically, indicating high information gain. However, if the three added phrases are "Project progress has been advanced," "Work content is continuously improved," and "On-site operations are proceeding in an orderly manner," although the language differs, they are essentially repetitive, and this method can identify them as low gain. Therefore, by combining "new addition ratio" and "semantic diversity" into an information gain value, this method can effectively avoid the defects of traditional methods such as overfitting template words and phrases or omitting the actual progress of new tasks. It ensures that the actual value of the cited information is accurately identified, improves the logical clarity and credibility of AI in judging the progress of tasks in engineering project scenarios, and is especially suitable for dealing with scenarios where information redundancy and structural complexity coexist in cross-reference links. It has high adaptability and practicality in engineering practice.

[0082] In one embodiment, S3: The step of calculating the truth value of a cross-reference link based on the content integrity value and the information gain value of the referenced content is as follows:

[0083] A two-dimensional coordinate system is established by calculating the truth value of the cross-reference link based on the content completeness value and the information gain value of the referenced content. The truth value of the cross-reference link calculated based on the content completeness value and the information gain value of the referenced content is mapped to the two-dimensional coordinate system; the coordinate points correspond to the content completeness value and the information gain value of the referenced content, respectively.

[0084] Calculate the distance from the coordinate point to the centroid in the two-dimensional coordinate system to obtain the truth value of the corresponding cross-reference link.

[0085] It's important to note that by mapping content integrity and information gain values ​​to a two-dimensional coordinate system and calculating their Euclidean distance from the origin as the authenticity value of the cross-reference link, this approach simultaneously reflects the two core dimensions of "whether the reference faithfully preserves the original content" and "whether it brings new and valid information" within a unified numerical space. This allows for a comprehensive evaluation of the authenticity of the reference chain. Compared to weighted averages or linear combinations, this method is more stable and objective, avoiding interference from human weighting. It is particularly suitable for scenarios where dimensions have asymmetrical contributions or mutual compensation relationships. For example, when a reference perfectly preserves the original content but adds no new information (high integrity, low gain) or vice versa (low integrity, high gain), its projection in the two-dimensional space still accurately reflects its "partially true" state. Only when both are high is its distance from the origin far, indicating that the reference chain has high semantic transmission authenticity. This approach not only enhances the ability to identify complex semantic evolution paths in cross-reference chains, but also gives the truth value good visualization characteristics and distinguishability, which helps to quickly identify risky links and potential information distortion points in project monitoring, and enhances the AI ​​monitoring system's comprehensive perception of the semantic logical consistency of multi-round references.

[0086] In one embodiment, S4: The step of using the truth value of the cross-reference link as a credibility correction factor for task completion status identification in project detection, to assist in determining whether the task has been truly completed, is as follows:

[0087] Compare the authenticity value of the cross-reference link with the preset authenticity value threshold. If the authenticity value is not less than the preset authenticity value threshold, it means that the authenticity of the cross-reference link in the entire project process is high and can fully reflect the actual problems. Analyzing the cross-reference link can directly and accurately monitor the entire project process.

[0088] Conversely, if the authenticity value is less than the preset authenticity value threshold, it indicates that the authenticity of the cross-reference links in the entire project process is low and cannot fully reflect the actual problems. The monitoring results are directly judged to be erroneous, an alarm is issued, and manual actual monitoring is carried out.

[0089] It should be noted that during the entire project monitoring process, using the authenticity value of cross-reference links as a credibility correction factor for task completion status identification can effectively enhance the system's ability to judge the authenticity of semantic transmission in text. Specifically, by comparing this authenticity value with a preset threshold, if it is not less than the threshold, it indicates that the cross-reference chain performs well in terms of both content completeness and information gain, and the credibility of the semantic propagation process is high. It can be considered that the cited content truly and effectively reflects the original problem and its processing status, so the system can automatically determine that the task has been truly completed. For example, if a construction task is cited multiple times in the daily report and its citation chain authenticity value is high, it indicates that the completion status of the task has been stably and consistently transmitted to the final report, and the system can automatically identify the task as "completed". Conversely, if the authenticity value is low, it indicates that there may be problems such as taking things out of context, deleting content, or incorrect summarizing in the citation. In this case, the system no longer trusts the text information of the link, but triggers a manual verification process, and determines whether the task has been truly completed through on-site review or supervisor confirmation. By introducing this credibility correction mechanism, the risk of traditional AI systems misjudging task status due to one-sided reliance on referenced content can be avoided, thereby improving the robustness and accuracy of the monitoring system. It is particularly suitable for real-world engineering management environments with dense cross-references of multiple documents and complex task chains.

[0090] Based on the same inventive concept, this invention also provides an artificial intelligence-based project end-to-end monitoring system. See also Figure 2 , Figure 2 A framework diagram of an AI-based project end-to-end monitoring system is provided for embodiments of the present invention. The system includes:

[0091] Cross-reference link module: Based on the reference language existing in the project documents, construct a document reference graph of the entire project process, and identify each complete reference path from the document reference graph as a cross-reference link;

[0092] Calculation module: Based on the document content corresponding to the cross-reference link, calculate the content integrity value and the information gain value of the referenced content; the content integrity value and the information gain value of the referenced content are used to measure whether the information contained in the referenced fragment completely retains the original content of the referenced fragment and to measure whether the referenced fragment introduces new semantic information compared to the referenced fragment, respectively.

[0093] Authenticity module: Calculates the authenticity value of cross-reference links based on the content completeness value and the information gain value of the referenced content;

[0094] Monitoring module: The authenticity value of cross-reference links is used as a credibility correction factor for task completion status identification in project monitoring, to help determine whether the task has been truly completed.

[0095] Based on the embodiments of the present invention, an artificial intelligence-based project end-to-end monitoring system can quantitatively evaluate the authenticity of cross-reference links through the above-mentioned method, increase the accuracy of automatic identification of the true status of the entire project, make existing AI monitoring systems accurate in judging and clearly identify risks when processing cross-reference texts, and ensure that the results of the project end-to-end monitoring are consistent with the actual situation, thus ensuring the accuracy of the monitoring results.

[0096] In one embodiment, the calculation module includes: a structure information retention rate module: each node in the cross-reference link represents a document fragment; for each pair of document fragments in the cross-reference link... The structure retention rate is calculated by comparing the smallest semantic units. The formula for the calculation is: In the formula, Indicates the first Structural information retention rate in segment references; Represents a document segment The set of semantic units in Represents a document segment The content mentioned in the text corresponds to a set of semantic units; Indicates how much of the original structure is accurately preserved in the quoted section; This indicates the total number of semantic units in the original segment;

[0097] Information Structure Residual Rate Module: Calculates the product of all structural residual rates in the entire cross-reference link, which is used as the information structure residual rate of the entire cross-reference link;

[0098] Directional Consistency Module: Extracts the starting point in the cross-reference link. and the end point Corresponding semantic vector and Calculate the directional projection to obtain the semantic directional consistency between the beginning and end of the link. The calculation formula is as follows: In the formula, The semantic direction consistency between the beginning and end of the link;

[0099] Content integrity module: Multiply the information structure residual rate of the entire cross-reference link by the information structure residual rate of the entire cross-reference link, and use the result of the multiplication as the content integrity value.

[0100] In one embodiment, the computing module further includes:

[0101] Semantic sub-block module: Extracts the corresponding semantic vectors from the initially referenced document fragments and the ultimately referenced document fragments in the cross-reference link; performs discrete semantic block operation on these two vectors, dividing each vector into several semantic sub-blocks according to the nesting position;

[0102] The new semantic information block module calculates the similarity between each semantic sub-block corresponding to the semantic vector of the final referenced document fragment and each semantic sub-block corresponding to the semantic vector of the initial referenced document fragment, and compares the similarity with a preset similarity threshold. If the similarity between each semantic sub-block corresponding to the semantic vector of the final referenced document fragment and all semantic sub-blocks corresponding to the semantic vector of the initial referenced document fragment is less than the preset similarity threshold, then the corresponding semantic sub-block is recorded as a new semantic information block.

[0103] The module for the proportion of newly added semantic sub-blocks: divide the total number of newly added semantic information blocks by the total number of semantic sub-blocks corresponding to the semantic vector of the initially referenced document fragment to obtain the proportion of newly added semantic sub-blocks;

[0104] Information Diversity Module: Calculate the similarity between all pairs of newly added semantic sub-blocks, subtract the similarity from the value 1 to obtain the information diversity value between the corresponding two newly added semantic sub-blocks, and average all information diversity values ​​to obtain the information diversity score.

[0105] Information gain module: Multiply the information diversity and the proportion of newly added semantic sub-blocks to obtain the information gain value.

[0106] In one embodiment, the realism module includes:

[0107] Coordinate System Module: A two-dimensional coordinate system is established by calculating the truth value of the cross-reference link based on the content completeness value and the information gain value of the referenced content. The truth value of the cross-reference link calculated from the content completeness value and the information gain value of the referenced content is mapped to the two-dimensional coordinate system; the coordinate points correspond to the content completeness value and the information gain value of the referenced content, respectively.

[0108] The authenticity value module calculates the distance from a coordinate point in a two-dimensional coordinate system to the centroid, thus obtaining the authenticity value of the corresponding cross-reference link.

[0109] In one embodiment, the monitoring module includes:

[0110] The first monitoring module compares the authenticity value of the cross-reference link with the preset authenticity value threshold. If the authenticity value is not less than the preset authenticity value threshold, it means that the authenticity of the cross-reference link in the entire project process is high and can fully reflect the actual problems. Analyzing the cross-reference link can directly and accurately monitor the entire project process.

[0111] The second monitoring module: If the authenticity value is less than the preset authenticity value threshold, it means that the authenticity of the cross-reference links in the entire project process is low and cannot fully reflect the actual problems. It directly judges that the monitoring results are wrong, issues an alarm, and conducts actual manual monitoring.

[0112] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An artificial intelligence-based project whole-process monitoring method, characterized in that, The method comprises the following steps: Based on the reference language existing in the project document, a document reference graph of the whole process of the project is constructed, and each complete reference path is identified as a cross-reference link from the document reference graph; According to the document content corresponding to the cross-reference link, the content completeness value and the information gain degree value of the referenced content are calculated; the content completeness value and the information gain degree value of the referenced content are respectively used to measure whether the information contained in the reference fragment is complete and whether the reference fragment introduces new semantic information compared with the referenced fragment; According to the content completeness value and the information gain degree value of the referenced content, the authenticity value of the cross-reference link is calculated; The authenticity value of the cross-reference link is used as a credibility correction factor for task completion status identification in project monitoring, to assist in determining whether the task has been truly completed; The content completeness value is calculated by: Each node in the cross-reference link represents a document segment; for each pair of document segments in the cross-reference link The structural preservation rate is calculated by comparing the minimum semantic units, and the formula is: , wherein, represents the structural information preservation rate in the segment reference; represents the semantic unit set in the document segment , represents the corresponding semantic unit set of the content mentioned in the document segment ; represents how much of the original structure is accurately preserved in the reference segment; represents the total number of semantic units of the original segment; Calculating the product of the structure retention rates of all the cross-reference links as the information structure residual rate of the whole cross-reference link; Extracting the start point of the cross-reference link and the end point corresponding semantic vectors and , calculating the direction projection to obtain the consistency degree of the semantic direction of the head and tail of the link, and the formula is: , wherein is the consistency degree of the semantic direction of the head and tail of the link; The information structure residual rate of the whole cross-reference link is multiplied by the information structure residual rate of the whole cross-reference link, and the result of the multiplication is taken as the content completeness value; The information gain degree value is calculated by: The semantic vectors corresponding to the initial referenced document fragment and the final referenced document fragment in the cross-reference link are extracted; discrete semantic block operation is performed on the two vectors, and each vector is divided into several semantic sub-blocks according to the nesting position; The similarity between each semantic sub-block corresponding to the semantic vector of the final referenced document fragment and each semantic sub-block corresponding to the semantic vector of the initial referenced document fragment is calculated, and the similarity is compared with the preset similarity threshold value; if the similarity between each semantic sub-block corresponding to the semantic vector of the final referenced document fragment and all the semantic sub-blocks corresponding to the semantic vector of the initial referenced document fragment is less than the preset similarity threshold value, the corresponding semantic sub-block is recorded as a new semantic information block; The total number of new semantic information blocks is divided by the total number of semantic sub-blocks corresponding to the semantic vector of the initial referenced document fragment to obtain the proportion of new semantic sub-blocks; The similarity between all pairs of new semantic sub-blocks is calculated, and the value 1 is subtracted from the similarity to obtain the information diversity value between the corresponding two new semantic sub-blocks; the average of all information diversity values is obtained to obtain the information diversity degree; The information diversity degree and the proportion of new semantic sub-blocks are multiplied to obtain the information gain degree value. 2.The project whole-process monitoring method based on artificial intelligence according to claim 1, characterized in that, The steps of calculating the authenticity value of the cross-reference link according to the content completeness value and the information gain degree value of the referenced content are: A two-dimensional coordinate system is established according to the content completeness value and the information gain degree value of the referenced content, and the authenticity value of the cross-reference link is mapped into the two-dimensional coordinate; the coordinate points correspond to the content completeness value and the information gain degree value of the referenced content respectively; The distance from the origin to the coordinate point in the two-dimensional coordinate system is calculated to obtain the authenticity value corresponding to the cross-reference link. 3.The project whole-process monitoring method based on artificial intelligence according to claim 1, characterized in that, The steps of using the authenticity value of the cross-reference link as a credibility correction factor for task completion status identification in project monitoring to assist in determining whether the task has been truly completed are: The authenticity value of the cross-reference link is compared with a preset authenticity value threshold, if the authenticity value is not less than the preset authenticity value threshold, it indicates that the cross-reference link in the whole project process is relatively high in authenticity, and can fully reflect the actual problem, and analysis on the cross-reference link can directly and accurately monitor the whole project process; If the authenticity value is less than the preset authenticity value threshold, it indicates that the cross-reference link in the whole project process is relatively low in authenticity, and cannot fully reflect the actual problem, and the monitoring result is directly determined to be wrong, and an alarm is sent for human actual monitoring.

4. An artificial intelligence-based project whole-process monitoring system, characterized in that, The system comprises: A cross-reference link module: based on the reference language existing in the project document, a document reference graph of the whole project process is constructed, and each complete reference path is identified from the document reference graph as a cross-reference link; A calculation module: according to the document content corresponding to the cross-reference link, a content completeness value and an information gain degree value of the reference content are calculated; the content completeness value and the information gain degree value of the reference content are respectively used to measure whether the information contained in the reference fragment is complete and retains the original content of the referenced fragment, and to measure whether the reference fragment introduces new semantic information compared with the referenced fragment; An authenticity module: the authenticity value of the cross-reference link is calculated according to the content completeness value and the information gain degree value of the reference content; A monitoring module: the authenticity value of the cross-reference link is used as a credibility correction factor for task completion status recognition in project monitoring, to assist in determining whether the task has been truly completed and processed; The calculation module comprises: Structure information retention rate module: each node in the cross-reference link represents a document segment; for each pair of document segments in the cross-reference link , the structure retention rate is calculated by comparing the minimum semantic units, and the formula is: , wherein, represents the structure information retention rate in the first segment reference; represents the semantic unit set in the document segment , represents the semantic unit set corresponding to the content mentioned in the document segment ; represents how much of the original structure is accurately retained in the reference segment; represents the total number of semantic units of the original segment; An information structure residual rate module: the product of all structure residual rates in the whole cross-reference link is calculated as the information structure residual rate of the whole cross-reference link; The direction consistency module extracts the semantic vectors of the start point and the end point of the cross-reference link, and calculates the direction projection to obtain the semantic direction consistency of the start point and the end point of the link. The formula is as follows: The semantic direction consistency of the start point and the end point of the link.​​​​ A content integrity module: the information structure residual rate of the whole cross-reference link is multiplied by the information structure residual rate of the whole cross-reference link, and the result of the multiplication is taken as the content completeness value; The calculation module further comprises: A semantic sub-block module: the corresponding semantic vectors of the initially referenced document fragment and the finally referenced document fragment in the cross-reference link are extracted; discrete semantic blocking operations are performed on the two vectors, and each vector is divided into a plurality of semantic sub-blocks according to the nesting position; A new semantic information block module: the similarity between each semantic sub-block corresponding to the semantic vector of the finally referenced document fragment and each semantic sub-block corresponding to the semantic vector of the initially referenced document fragment is calculated, and the similarity is compared with a preset similarity threshold, if the similarity between each semantic sub-block corresponding to the semantic vector of the finally referenced document fragment and all semantic sub-blocks corresponding to the semantic vector of the initially referenced document fragment is less than the preset similarity threshold, the corresponding semantic sub-block is recorded as a new semantic information block; A new semantic sub-block proportion module: the total number of new semantic information blocks is divided by the total number of semantic sub-blocks corresponding to the semantic vector of the initially referenced document fragment to obtain the proportion of new semantic sub-blocks; An information diversity degree module: the similarity between all new semantic sub-blocks is calculated, and the value 1 is subtracted from the similarity to obtain the information diversity value between the corresponding two new semantic sub-blocks, and the average of all information diversity values is obtained as the information diversity degree. The information gain degree module: the information diversity degree and the proportion of the newly added semantic sub-blocks are multiplied to obtain an information gain degree value. 5.The project whole-process monitoring system based on artificial intelligence according to claim 4, characterized in that, The authenticity module includes: The coordinate system module: according to the content integrity value and the information gain degree value of the referenced content, the authenticity value of the cross-reference link is calculated, a two-dimensional coordinate system is established, and the authenticity value of the cross-reference link is mapped into the two-dimensional coordinate; the coordinate points correspond to the content integrity value and the information gain degree value of the referenced content respectively; The authenticity value module: the distance from the coordinate point in the two-dimensional coordinate system to the original center is calculated to obtain the authenticity value of the corresponding cross-reference link. 6.The project whole-process monitoring system based on artificial intelligence according to claim 4, wherein, The monitoring module includes: The first monitoring module: the authenticity value of the cross-reference link is compared with the preset authenticity value threshold; if the authenticity value is not less than the preset authenticity value threshold, it indicates that the authenticity of the cross-reference link in the whole project process is relatively high, and the cross-reference link can completely reflect the actual problem; the cross-reference link is analyzed to directly and accurately monitor the whole project process; The second monitoring module: if the authenticity value is less than the preset authenticity value threshold, it indicates that the authenticity of the cross-reference link in the whole project process is relatively low, and the cross-reference link cannot completely reflect the actual problem; the monitoring result is directly determined to be wrong, and an alarm is sent for human actual monitoring.

Images