A park investment process data monitoring method and system

Abstract

This invention relates to the field of information retrieval structure technology, specifically to a method and system for monitoring data during the investment promotion process in industrial parks. The method includes the following steps: extracting investment promotion text, identifying combinations of verbs and resource nouns, analyzing word order and semantic features, filtering action-driven segments, dividing structural sequences and identifying connection relationships, establishing semantic jump paths, reorganizing sentence order, and generating a data monitoring structure template for investment promotion in industrial parks. In this invention, by extracting verbs and resource nouns to construct sentence content, combining word order features and semantic tags to classify and filter segments, dividing structural sequences based on the position of resource nouns and merging content with consistent semantic direction, mapping connectors to jump paths to construct task behavior sequences, and realizing the logical organization and sequential reorganization of sentences, this generates a data expression method with traceability and structured features, improving the semantic recognition efficiency and data monitoring capabilities during the investment promotion process in industrial parks.

Description

Technical Field

[0001] This invention relates to the field of information retrieval structure technology, and in particular to a data monitoring method and system for the investment promotion process in industrial parks. Background Technology

[0002] The field of information retrieval structure technology mainly involves technical methods for data organization, storage, retrieval, and access. Core aspects include the design of storage formats for structured and unstructured data, the construction and optimization of index structures, query processing mechanisms, metadata management, and data access control mechanisms. This field aims to improve the accuracy and response speed of data access by building efficient data structures and retrieval algorithms, and is widely used in information-intensive scenarios such as database systems, information retrieval systems, document management systems, and data monitoring systems. Traditional data monitoring methods for industrial park investment promotion refer to the collection, analysis, and supervision of a large amount of data related to enterprise information, negotiation progress, resource matching, and project implementation during the investment promotion process. Data management typically relies on manual recording and report summarization. Registration information, contact persons, industry types, and preliminary cooperation intentions of relevant enterprises are entered into static tables or basic information systems. Management personnel then periodically compile and summarize the progress of investment projects and track project development through periodic reports. Data statistics and feedback largely depend on manual classification and updates, lacking real-time monitoring and dynamic analysis capabilities for process data.

[0003] Existing technologies lack the ability to deeply extract the semantic structure of text in the processing of investment promotion data in industrial parks. The content of sentences cannot be classified and summarized based on the combination of verbs and resource nouns, and the word order features are not effectively identified. As a result, the task behavior information in the text is difficult to decompose and organize, the data expression lacks a unified logical path, and the logical relationship between sentences cannot be effectively mapped. In the actual process, the information content is prone to problems such as fragmentation and disordered order. It is difficult to construct a data structure with a complete semantic chain, and there are obvious obstacles in task status tracking, stage result identification, and process reconstruction. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a data monitoring method for the investment promotion process in industrial parks, comprising the following steps: S1: Extract investment promotion text information from the park, segment sentences, identify combinations of verbs and resource-related nouns, analyze the positional distribution characteristics of verbs, assign semantic tags based on the results, and generate a list of task sentences; S2: Extract text from the park's investment promotion task record, identify whether the combination of verbs and resource-related nouns matches the task statement list, filter out segments with action-driven word order, determine completeness, and generate a set of semantically corresponding segments; S3: Analyze the position of resource-related nouns in the semantically corresponding fragment set, perform structural sequence division, establish arrangement rules, merge fragments with the same semantic direction, identify structural missing intervals between paragraphs, fill in content or connect sentences based on the semantic relationship between adjacent sentences or the transitional expressions in the original text, splice missing intervals, and generate a semantic path division map. S4: In the semantic path partitioning graph, a semantic sequence is established according to the text order, the structure of connecting words is extracted, the category is identified and mapped to the word order jump path, and a task behavior association graph is generated. S5: Based on the task behavior association graph, combined with the word order jump path and task content order, merge sentence fragments, reorganize paragraph order, construct a sentence organization structure with monitoring instruction arrangement characteristics, and build a data monitoring structure template for the park investment promotion process.

[0005] As a further embodiment of the present invention, the task statement list includes verb lexical units, resource-type nouns, and word order position tags; the semantic corresponding fragment set includes semantic fragment identifiers, action-dominant markers, and element integrity tags; the semantic path partitioning graph includes resource noun position nodes, semantic direction paragraphs, and path splicing rules; the task behavior association graph includes semantic order sequences, connector type classifications, and jump path nodes; and the park investment promotion process data monitoring structure template includes paragraph structure units, word order recombination patterns, and monitoring instruction sequences.

[0006] As a further aspect of the present invention, the phrase with action-driven characteristics refers to a sentence structure centered on a verb, in which the verb performs an action on resource-related nouns in the sentence, thus reflecting a behavior-oriented phrase structure.

[0007] As a further aspect of the present invention, the identification of categories and mapping to word order jump paths refers to identifying the functional categories of grammatical structures such as conjunctions, and transforming their functional relationships into jump logic paths between text word orders to construct semantic flow relationships.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Based on the park text information received by the regional input interface, the text is segmented into sentences by period, semicolon and comma, sentences containing both verbs and resource nouns are filtered out, sentence fragments are recorded, and a set of sentences including verbs and resource nouns is generated. S102: Call the set of statements including verbs and resource nouns, parse the position of the verbs in the statements, extract the position information of the verbs at the beginning, middle and end of the sentences, and statistically analyze the distribution features to generate a set of verb position distribution features for the sentences; S103: Based on the verb position distribution feature set of the statement, construct a mapping rule base for structure and semantic tags, match and label the statement structure, assign semantic tags to all statements and output them to obtain a list of task statements.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Obtain the original text content in the park's investment promotion task record, retrieve the verb and resource-related noun combinations in each statement in the text, and compare them one by one with the semantic combinations in the task statement list to establish a matching mapping structure and obtain a semantic combination matching result set. S202: Based on the semantic combination matching result set, filter the sentence fragments in the region that meet the preset position weight threshold, label them according to the correspondence between verb position distribution and semantic tags, and archive the sentences that meet the preset conditions to obtain a set of action-driven sentence fragments; S203: Invoke the action-driven statement fragment set, and sequentially determine whether the statement has three elements: resource noun, action verb and preset limiting condition. Perform a removal operation on statements with missing elements and establish a semantically corresponding fragment set.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Call the semantic corresponding fragment set, locate the position of resource-type nouns in the sentence fragment, extract the position index of the verb corresponding to the resource-type noun in each sentence, perform structural sequence partitioning of the sentence based on position nodes, and obtain the resource position structure sequence set; S302: Based on the resource location structure sequence set, call the semantic tags corresponding to the statements, construct the segment classification rules under the same semantic tag, perform position aggregation operation on the statement segments with the same tag, perform sequential arrangement and merging processing, and obtain the semantic classification structure paragraph set; S303: For the semantic classification structure paragraph set, identify the structural missing intervals between paragraphs, collect the end and start information of adjacent sentence fragments in the same type of paragraph, call the preset transition sentence template, perform content completion and connection operations, and establish a semantic path division map.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Call the semantic path partitioning graph, extract the position index of the sentence fragments in the structural paragraph in the original text, and perform a sequence reconstruction operation on the sentence fragments in the paragraph according to the original text arrangement order to generate the original semantic sequence set; S402: Based on the original semantic sequence set, extract the connection positions between segments, retrieve the conjunctions included at the sentence connection points, identify the categories based on the part-of-speech and functional features of the conjunctions, and record their positions to obtain a conjunction classification tag set; S403: Call the connector classification tag set, map the classified connectors to the position intervals between adjacent statements in the original semantic sequence, construct the jump path structure, and combine the semantic sequence and connection logic relationship to establish a task behavior association graph.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Obtain the word order jump path and task content order recorded in the task behavior association graph, locate the structural paragraph corresponding to the sentence fragment in the graph, perform order judgment on adjacent fragments and perform structural merging operation to generate a word order structure aggregation set. S502: Based on the aggregated set of word order structure, extract the defined jump logic constraints in each group of segments, identify the position indices of the start and end segments in the paragraphs, rearrange the structural paragraphs according to the path order, and establish a path constraint reorganization sequence set; S503: Call the path constraint recombination sequence set, extract the statement fragments with monitoring instruction arrangement characteristics in sequence, combine them in the order of instructions, and map them to the park investment promotion data flow framework to build a data monitoring structure template for the park investment promotion process.

[0013] A data monitoring system for the investment promotion process in a park includes: The system includes: The Task Statement List module is used to implement S1: extract investment promotion text information from the park, divide the statements, identify combinations of verbs and resource-related nouns, analyze the positional distribution characteristics of verbs, assign semantic tags based on the results, and generate a task statement list. The action-driven semantic recognition module is used to implement S2: extracting text from the park's investment promotion task record, identifying whether the combination of verbs and resource-related nouns matches the task statement list, filtering fragments with action-driven word order, judging completeness, and generating a set of semantically corresponding fragments; The semantic path construction module is used to implement S3: analyze the position of resource-type nouns in the set of semantically corresponding fragments, divide the structural sequence, establish arrangement rules, merge fragments with the same semantic direction, identify structural missing intervals between paragraphs, fill in content or connect sentences based on the semantic relationship between adjacent sentences or the transitional expressions in the original text, splice missing intervals, and generate a semantic path division map. The behavior jump graph module is used to implement S4: In the semantic path partitioning graph, a semantic sequence is established according to the text order, the structure of connecting words is extracted, the category is identified and mapped to the word order jump path, and a task behavior association graph is generated. The data monitoring template module is used to implement S5: based on the task behavior association graph, combined with the word order jump path and task content order, merge sentence fragments, reorganize paragraph order, construct a sentence organization structure with monitoring instruction arrangement characteristics, and build a data monitoring structure template for the park investment promotion process.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, verbs and resource-related nouns are extracted to construct sentence content. Segments are categorized and filtered by combining word order features and semantic tags. Structural sequences are divided according to the position of resource nouns and content with consistent semantic direction is merged. Connecting words are mapped to jump paths to construct task behavior sequences, thereby realizing the logical organization and sequential reorganization between sentences and generating a data expression method with traceability and structured features, which improves the semantic recognition efficiency and data monitoring capabilities in the park's investment promotion process. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides a method for monitoring data during the investment promotion process in industrial parks, comprising the following steps: S1: Extract text information content used for investment promotion in the park through the regional input interface, perform sentence segmentation operation on the text, extract the verbs and resource-related nouns included in the sentences to construct sentence content, perform sequence feature recognition on the sentence content based on the distribution characteristics of verbs at the beginning, middle and end of the sentence, and assign semantic labels based on the recognition results to construct a task sentence list; S2: Extract the original text content from the park's investment promotion task record, identify whether the combination of verbs and resource-related nouns in the text is consistent with the task statement list, combine semantic tags, filter statement fragments with action-driven features in the word order, and perform element integrity judgment on the statement fragments to generate a set of semantically corresponding fragments. S3: Call the semantic corresponding fragment set, analyze the occurrence position of resource-related nouns in the sentence fragments, divide the sentence fragments into structural sequences, establish arrangement rules based on semantic tags, merge sentence fragments belonging to the same semantic direction to form structural paragraphs, perform content splicing on missing intervals, and construct a semantic path partitioning graph; S4: Call the semantic path partitioning graph, establish a semantic sequence for the sentence fragments in the structural paragraph according to the original text order, extract the connecting word structure between the fragments, perform category recognition processing on the connecting words, and map them to the word order jump path to generate a task behavior association graph; S5: Invoke the task behavior association graph, and based on the word order jump path and task content order constructed in the graph, perform structural merging operation on the sentence fragments, combine the path jump logic to constrain and reorganize the paragraph arrangement order, construct a sentence organization structure with monitoring instruction arrangement characteristics, and build a data monitoring structure template for the park investment promotion process.

[0020] The task statement list includes verb morphemes, resource nouns, and word order position tags; the semantic corresponding fragment set includes semantic fragment identifiers, action-dominant markers, and element integrity tags; the semantic path partitioning diagram includes resource noun position nodes, semantic direction paragraphs, and path splicing rules; the task behavior association graph includes semantic sequence, connector type classification, and jump path nodes; and the park investment promotion process data monitoring structure template includes paragraph structure units, word order reorganization patterns, and monitoring instruction sequences.

[0021] Please see Figure 2 The specific steps of S1 are as follows: S101: Based on the park text information received by the regional input interface, the text is segmented into sentences by period, semicolon and comma, sentences containing both verbs and resource nouns are filtered out, sentence fragments are recorded, and a set of sentences including verbs and resource nouns is generated. The system receives raw text stream data from investment promotion activities in the park via a regional input interface. This text data originates from mobile recorders of investment promotion personnel, transcripts of meeting recordings, and log files from enterprise resource planning. The data formats include plain text (TXT) and rich text (RTF). Upon receiving the text, a text cleaning and preprocessing procedure is immediately initiated. A character matching algorithm based on regular expressions is used to sequentially identify the ASCII or Unicode encoding of three types of punctuation marks: periods, semicolons, and commas. When a corresponding encoding position is detected, a truncation operation is performed, physically dividing the long text stream into independent string units. Subsequently, for each segmented string unit, a pre-trained bidirectional long short-term memory (BiLSTM) network model is used for part-of-speech tagging. This BiLSTM model consists of an input layer, an embedding layer, two hidden layers, and an output layer. The input layer receives 300-dimensional word vectors generated by the Word2Vec algorithm. The embedding layer maps the word vectors to a high-dimensional feature space. The first hidden layer contains 256 neurons and uses the hyperbolic tangent Tanh activation function to process the forward features of the sequence. The second hidden layer also contains 256 neurons and is responsible for capturing the backward dependency features of the sequence. Fully connected layers are used, and a Dropout mechanism with a dropout rate of 0.2 is introduced at the connection points to prevent overfitting. The output layer uses the Softmax function to generate the probability distribution of the part-of-speech tag for each word. When training this model, a labeled dataset containing 50,000 entries of park investment promotion text was used. The dataset was manually cleaned to remove garbled characters and whitespace, and divided into training, validation, and test sets in a 7:2:1 ratio. During training, the cross-entropy loss function is used to measure the difference between the predicted and true part-of-speech tags. The calculation logic of the cross-entropy loss function is as follows: using the true part-of-speech tag distribution as the target distribution and the predicted distribution output by the model as the estimated distribution, the weighted negative sum of the log-likelihoods of the two for each category is calculated, i.e., the cross-entropy value, which measures the degree of probability deviation in the model's predictions. The Adam optimizer is used for optimization, with an initial learning rate set to 0.001. The network weight parameters are iteratively updated using the backpropagation algorithm. Using the part-of-speech tags output by the model, each string unit is scanned one by one, logically determining whether it simultaneously contains words marked as verbs and words marked as resource-related nouns. Resource-related nouns include terms from a pre-defined vocabulary such as "land," "factory," "funds," and "taxes." Only string units that simultaneously satisfy both conditions are retained, defined as valid sentence fragments, and stored in the sentence set database.For example, given the input text "The investment manager inspected the second phase of the land; and subsequently discussed the rent reduction policy," the program uses semicolons to segment the text. In the first sentence, "The investment manager inspected the second phase of the land," "inspected" is identified as a verb, and "land" is identified as a resource-related noun, thus satisfying the condition and being retained. In the second sentence, "and subsequently discussed the rent reduction policy," "discussed" is a verb, and "rent" is a resource-related noun, also being retained. The final generated set contains all sentence fragments that meet the conditions.

[0022] S102: Call the set of statements including verbs and resource nouns, parse the position of verbs in the statements, extract the position information of verbs at the beginning, middle and end of the sentences, and statistically analyze the distribution features to generate a set of verb position distribution features in the sentences; First, the total word count of the sentence is obtained using a word segmentation tool. Then, the index position of the verb within the sentence is located. The sentence-beginning interval is defined as an index value less than or equal to the total word count multiplied by 0.2; the sentence-end interval is defined as an index value greater than or equal to the total word count multiplied by 0.8; and the middle interval is the range between the sentence-beginning and sentence-end. Through numerical comparison logic, the verb position index is compared with the boundary values ​​of these three intervals to determine the specific position of the verb. The statistical distribution feature generation process for the position information uses the Gaussian kernel density estimation method. This method does not pre-determine the data distribution form but uses a Gaussian kernel function to smooth each verb position data point. Specifically, for each position data point, a Gaussian distribution curve with a standard deviation of 1.0 is constructed centered on that point. The curves corresponding to all data points are superimposed and summed in the same coordinate system to obtain the overall probability density function curve. The peak region of this curve represents the high-frequency position region where the verb appears. For example, when processing 1000 investment promotion records, if 700 sentences have verbs at the beginning of the sentence, 200 in the middle, and 100 at the end, the Gaussian kernel density estimation will show a significant density peak at the beginning of the sentence. Based on this density distribution, a sentence verb position distribution feature set is generated, containing position probability density values, peak coordinates, and skewness coefficients. This feature set explicitly records the positional preferences of different types of verbs in specific contexts. For example, action verbs such as "inspect" and "visit" tend to be distributed at the beginning of the sentence, while result verbs such as "agree" and "sign" tend to be distributed at the end. These feature data are stored in the form of structured key-value pairs, where the key name is the position interval identifier and the key value is the corresponding probability density integral value.

[0023] S103: Based on the verb position distribution feature set of the sentences, construct a mapping rule base between structure and semantic tags, match and label the sentence structure, assign semantic tags to all sentences and output them to obtain the task sentence list; Based on the verb position distribution feature set of sentences, a mapping rule base for structure and semantic labels is constructed. This construction process employs a decision tree classification algorithm, using verb position distribution features as the basis for splitting root nodes and internal nodes, and semantic labels as leaf nodes. For example, rule one is set: if the integral value of the verb position probability density in the first interval of the sentence is greater than 0.6, and the verb belongs to the "movement behavior" category, then the matched semantic label is "on-site investigation stage"; rule two is set: if the verb is located in the last interval of the sentence and is accompanied by resource nouns such as "contract" or "agreement," then the matched semantic label is "signing and implementation stage." During the construction of the decision tree, the optimal splitting attribute is selected using the principle of minimizing the Gini coefficient. The Gini coefficient is calculated by subtracting the sum of squared probabilities of each category from 1, used to measure the impurity of the dataset. Using this mapping rule base, all input sentence structures are matched one by one, mapping each sentence to a specific semantic category and assigning a corresponding semantic label. The output results are compiled into a task sentence list, which not only contains the original sentence content but also includes semantic labels, verb position feature values, and confidence scores. For example, for the statement "The delegation inspected the factory in Area A", based on the characteristic that the verb "inspected" is at the beginning of the sentence and the object is "factory", the tag "resource recommendation" is matched and assigned a confidence score of 0.95. Each data item in the list contains a unique task ID, timestamp, and processing status marker to ensure that subsequent steps can be accurately invoked.

[0024] Please see Figure 3 The specific steps of S2 are as follows: S201: Call the task statement list, obtain the original text content in the park's investment promotion task record, retrieve the verb and resource-related noun combinations in each statement in the text, and compare them one by one with the semantic combinations in the task statement list to establish a matching mapping structure and obtain the semantic combination matching result set. The system invokes the original text content from the task statement list and the park's investment promotion task records, initiating a dual retrieval mechanism. First, a full scan of the original text content is performed, using named entity recognition technology to extract verb entities and resource-related noun entities from each statement, constructing entity pair combinations from the original text. Next, this entity pair combination is compared with existing semantic combinations in the task statement list using vectorization. The comparison process employs cosine similarity calculation logic, mapping verbs and nouns in the original text to 512-dimensional word vectors, concatenating them into a 1024-dimensional combination vector A. Similarly, the semantic combinations in the list are processed to obtain vector B. The dot product of vector A and vector B is calculated and divided by the product of their magnitudes to obtain the similarity score. If the similarity score is greater than the preset matching threshold of 0.85, the two are considered a successful match. This threshold of 0.85 was determined after testing the similarity distribution of 10,000 manually annotated synonym pairs. Experimental data shows that at a threshold of 0.85, the false recognition rate is below 0.02 and the recall rate is above 0.96. When establishing the matching mapping structure, a hash table is created, using the unique identifier of the original text statement as the key and the semantic tags and detailed attributes from the list of matched task statements as values. The final semantic combination matching result set contains all successfully matched statement pairs and their corresponding similarity scores, matching times, and associated resource types.

[0025] S202: Based on the semantic combination matching result set, filter the sentence fragments in the region that meet the preset position weight threshold, label them according to the correspondence between the verb position distribution and semantic tags, and archive the sentences that meet the preset conditions to obtain the action-dominant sentence fragment set; First, a quantitative index for action dominance is defined, which is determined by the weighted sum of the verb's semantic weight and positional weight. The semantic weight of the verb is calculated based on the TF-IDF algorithm, where TF represents the word frequency of the verb in the current document, and IDF represents the logarithm of the ratio of the total number of documents in the corpus to the number of documents containing the verb, i.e., log(N / df), used to reflect the overall scarcity and semantic importance of the verb; the positional weight directly uses the probability density value of the positional distribution feature set generated in S102. The semantic weight coefficient is set to 0.6, and the positional weight coefficient is set to 0.4. The two are multiplied and summed to obtain the action dominance score. For example, if the TF-IDF value of a verb in a sentence is 0.05, which becomes 0.8 after normalization, and the positional probability density is 0.7, then the dominance score is 0.8 multiplied by 0.6 plus 0.7 multiplied by 0.4, resulting in 0.76. Sentence fragments with scores greater than the set benchmark value of 0.65 are then filtered out. The baseline value of 0.65 is set based on the statistical median after annotating the action intensity of 500 historical investment promotion documents. Experiments show that statements with a value higher than this accurately reflect actual investment promotion actions. Subsequently, the selected statements are annotated in a secondary manner based on the correspondence between verb position distribution and semantic tags. For example, statements with high dominant scores and verbs preceding the verb are marked as "actively initiated," while those with verbs following the verb are marked as "passively responded." Finally, all statement fragments that meet the criteria and have been annotated are archived to generate a set of action-dominant statement fragments. This set is stored in JSON format and includes the original statement text, dominant score, type tag, and source index.

[0026] S203: Invoke the action-driven statement fragment set, and sequentially determine whether the statement has three elements: resource noun, action verb and preset limiting condition. Perform a removal operation on statements with missing elements and establish a semantically corresponding fragment set. First, each statement fragment is parsed to generate a syntactic dependency tree, with the root node typically being the core verb (action verb). The dependency path is searched; if a child node marked "nsubj" or "agent" exists, it is determined to possess the "resource subject" element; if a child node marked "dobj" or "pobj" exists with a resource noun part of speech and is modified by a qualifier or quantifier, it is determined to possess the "preset constraint" element. The action verb itself is the root node. The program logic is set so that the statement is marked as "complete" only if the "resource subject" node, the "action verb" node, and the "preset constraint" node are all detected. For statements lacking any element, a removal operation is performed, removing the statement from the current processing queue and recording it in the exception log. For example, the statement "The enterprise (subject) signed (verb) a lease agreement (constraint)" has all three elements and is retained; while the statement "Considering (verb)" is deemed invalid and removed due to the lack of subject and resource constraint. The sentences that are ultimately retained form a semantically corresponding fragment set. Each piece of data in this set strictly conforms to an SVO (subject-verb-object) or similar subject-action-object structure, ensuring the data quality for subsequent processing.

[0027] Please see Figure 4 The specific steps of S3 are as follows: S301: Call the semantic corresponding fragment set, locate the position of resource-type nouns in the sentence fragment, extract the position index of the verb corresponding to the resource-type noun in each sentence, perform structural sequence partitioning of the sentence based on position nodes, and obtain the resource position structure sequence set; The semantically corresponding fragment set is invoked to initiate the position index extraction program. Each statement in the set is traversed, and string lookup functions in the programming language are used to locate the start and end character indices of resource-type nouns within the statement string. The start index is used as the position index value of the resource-type noun. Simultaneously, the position index of the verb within the statement is obtained. Based on these two index values, a structure sequence based on position nodes is constructed. This sequence is a two-dimensional array; the first dimension represents the statement's sequence number within the text, and the second dimension contains the verb index and the resource noun index. For example, in statement 5, if the verb is at index 10 and the resource noun is at index 25, the generated sequence item would be [5, 10, 25]. This process is performed on all statements to obtain a resource position structure sequence set. This set accurately quantifies the physical distribution of resources and actions within the text stream, providing a data foundation for subsequent spatiotemporal aggregation.

[0028] S302: Based on the resource location structure sequence set, call the semantic tags corresponding to the statements, construct the fragment classification rules under the same semantic tag, perform location aggregation operation on statement fragments with the same tag, perform sequential arrangement and merging processing, and obtain the semantic classification structure paragraph set; First, sentence fragments are grouped based on semantic tags; for example, all sentences tagged "price negotiation" are grouped together. Within each group, a distance-based aggregation algorithm is applied. The difference in resource location indices between adjacent sentence fragments is calculated. If the absolute value of the difference is less than a preset aggregation radius of 50 characters, the two fragments are considered to belong to the same semantic scenario, and a merging operation is performed. This 50-character aggregation radius is based on the fact that the continuous attention span for the same topic in human reading habits typically does not exceed 50 characters. Statistical analysis of eye-tracking data from 200 test subjects showed that the probability of topic switching within a 50-character range is less than 0.05. The merging process includes concatenating the text content in the order of appearance and updating the location indices to the merged range. After performing sequential sorting and merging on all groups, a semantically categorized paragraph set is generated. This set reorganizes scattered sentence fragments into paragraph blocks with coherent semantics; for example, it aggregates scattered question-and-answer statements about "water and electricity facilities" into a complete "infrastructure confirmation" paragraph.

[0029] S303: For a semantically categorized paragraph set, identify structural gaps between paragraphs, collect the end and beginning information of adjacent sentence fragments in the same type of paragraph, call the preset transition sentence template, perform content completion and connection operations, and establish a semantic path division map. The process identifies structural gaps between paragraphs, either logically or temporally. Specifically, it checks for uncovered text regions between the end of paragraph A and the beginning of paragraph B. If found, it collects the semantic feature vectors of the last sentence of paragraph A and the first sentence of paragraph B, and retrieves the text content located between these two positions in the original text. Pre-defined transitional sentence templates are used to complete and connect the missing text. A natural language generation model (such as a lightweight version of GPT-2) is employed to semantically smooth the missing regions, or transitional sentences extracted from the original text are directly used to fill them in. A semantic path partitioning graph is constructed, with paragraphs as nodes and the concatenated content or logical relationships between paragraphs as edges. The graph's data structure uses an adjacency matrix, with matrix elements storing the connection weights between paragraphs. For example, if paragraphs A and B are adjacent and semantically coherent in the original text, the connection weight is set to 1.0; if a long irrelevant interpolation occurs in between, the weight decays according to distance. The final generated semantic path partitioning graph comprehensively depicts the narrative flow of the investment promotion task from beginning to end.

[0030] Please see Figure 5 The specific steps of S4 are as follows: S401: Call the semantic path partitioning graph, extract the position index of the sentence fragments in the structural paragraph in the original text, perform the order reconstruction operation on the sentence fragments in the paragraph according to the original text arrangement, and generate the original semantic sequence set; The semantic path partitioning graph is invoked to extract the global position index of sentence fragments within each structural paragraph in the original text. Following the linear arrangement of the original text, the sentence fragments within each paragraph are reconstructed using a bubble sort algorithm, ordered in ascending order based on their starting position index to ensure consistency of the semantic sequence over time. During sorting, if overlapping position indices are found between two sentence fragments, they are processed according to semantic priority rules, prioritizing fragments containing core resource quantification data (such as amount or area). After reconstruction, an original semantic sequence set is generated. This sequence set recreates the actual order of events in investment promotion conversations or records, for example, first "introducing the park overview," then "inquiring about enterprise needs," and finally "matching land resources." Each element in the sequence contains a precise timestamp or text offset, ensuring a rigorous temporal order of the data flow.

[0031] S402: Based on the original semantic sequence set, extract the connection positions between segments, retrieve the conjunctions included at the sentence connection points, identify the categories based on the part-of-speech and functional features of the conjunctions, and record their positions to obtain a conjunction classification tag set; Based on the original semantic sequence set, the connection positions between adjacent segments are scanned, and the text content at those positions is extracted. Part-of-speech tagging tools are used to identify the connecting words at these positions, and the categories are determined based on the part-of-speech features (such as conjunctions and adverbs) and functional features (such as contrast, causality, and sequence). A connecting word classification table is established, as shown in Table 1. For each identified connecting word, its specific position coordinates in the semantic sequence (i.e., between which two statement IDs it lies) are recorded, resulting in a connecting word classification tag set. For example, "but" is identified as a contrast category, and its position is recorded between statement ID_101 and statement ID_102; "therefore" is identified as a causal category, and its position is recorded between statement ID_205 and statement ID_206.

[0032] Table 1. Classification and Functional Characteristics of Connectives As shown in Table 1, basic logical weights are assigned to different categories of connectors for edge weight calculation in subsequent graph construction. This classification tag set not only stores the words themselves, but also their corresponding logical attributes and weight parameters.

[0033] S403: Call the connector classification tag set, map the classified connectors to the position intervals between adjacent statements in the original semantic sequence, construct the jump path structure, and combine the semantic sequence and the connection logic relationship to establish a task behavior association graph; The system invokes a set of conjunction classification tags to map the classified conjunctions back to the positional intervals between adjacent statements in the original semantic sequence. A jump path structure is constructed, defining the transition probabilities between nodes. The transition probabilities are calculated based on the logical relationships between conjunctions: if the conjunction is "causal," the transition probability from the previous statement to the next statement is set to 0.9; if it is "contrast," the transition probability is set to 0.5. Combining the natural order of the semantic sequence with the above logical relationships, a task behavior association graph is established. This graph uses a directed weighted graph data structure, where nodes represent semantic statement fragments, edges represent connection relationships, and the weight of the edges is determined by the logical weight of the conjunctions. For example, node A (proposing a quote) is connected to node B (expressing objection) via "but" (contrast, weight 0.8), and then connected to node C (adjusting the plan) via "therefore" (causal, weight 0.9). During graph construction, the NetworkX library is used for graph data instantiation and storage to ensure the graph has queryable and traversable topological characteristics.

[0034] Please see Figure 6 The specific steps of S5 are as follows: S501: Call the task behavior association graph, obtain the word order jump path and task content order recorded in the graph, locate the structural paragraph corresponding to the sentence fragment in the graph, perform order judgment on adjacent fragments and perform structural merging operation to generate a word order structure aggregation set. The task behavior association graph is invoked to obtain the sequence jump paths and task content order recorded in the graph. The corresponding structural paragraphs of the statement fragments in the graph are located, and the order of adjacent fragments in the graph is determined. If two fragments have a directly connected edge in the graph, and the edge direction is consistent with the time axis, they are determined to be a sequential structure; if there is a bidirectional edge or a loop, they are determined to be a cyclic structure. A structure merging operation is performed. For adjacent fragments determined to be strongly related sequential structures (i.e., edge weight greater than 0.85), they are merged into a single supernode. This weight threshold of 0.85 is determined based on the modularity maximization algorithm in graph theory. By calculating the community division quality under different thresholds, it was found that the modularity reaches a peak of 0.62 at 0.85, at which point the generated structure aggregation set best represents the independent business stage. A sequence structure aggregation set is generated, which simplifies the complexity of the graph by merging fine-grained statements into coarse-grained business blocks, such as aggregating the three nodes "ask unit price," "ask property fee," and "ask water and electricity fee" into the "inquiry stage" supernode.

[0035] S502: Based on the sentence structure aggregation set, extract the defined jump logic constraints in each group of segments, identify the position indices of the start and end segments in the paragraph, rearrange the structural paragraphs according to the path order, and establish a path constraint reorganization sequence set; Based on the aggregated set of sentence structure, the defined jump logic constraints in each group of segments are extracted. Logical constraints include mandatory preconditions and mutually exclusive conditions. The starting and ending segment positions in each segment are identified. Starting segments are typically nodes with an in-degree of 0 or whose in-degree originates solely from the previous stage, while ending segments are nodes with an out-degree of 0 or that point to the next stage. The structural segments are rearranged according to the path order. If a conflict is found between the actual sentence order and the logical constraints (e.g., "signing the contract" appears before "checking the site"), it is corrected or marked as an abnormal process according to a preset mandatory logic template. A path constraint reorganization sequence set is established. This sequence set is constructed using a topological sorting algorithm to ensure that all directed edge dependencies are satisfied. If a circular dependency exists, a depth-first search (DFS) algorithm is used to break the cycle, selecting the edge with the lowest weight for pruning. The final sequence set is a strictly linear process description that conforms to the business logic of investment promotion.

[0036] S503: Call the path constraint to reorganize the sequence set, extract the statement fragments with monitoring instruction arrangement characteristics in sequence, combine them in the order of instructions, and map them to the park investment promotion data flow framework to build a data monitoring structure template for the park investment promotion process; The system invokes path constraints to reorganize the sequence set, extracting sentence fragments with monitoring instruction arrangement characteristics in sequence. Monitoring instruction characteristics refer to fragments containing control verbs such as "monitor," "record," "upload," and "review." Instructions are combined according to their order of appearance in the sequence and mapped to the park's investment promotion data flow framework. This framework predefines four standard stages: "lead acquisition," "intent communication," "solution negotiation," and "signing and entry." When constructing the mapping relationship, the Euclidean distance between the feature vector of a sentence fragment and the standard description vector of each stage is calculated, assigning the sentence to the stage with the smallest distance. The Euclidean distance calculation logic is to calculate the square root of the sum of the squares of the differences between corresponding elements of two vectors. For example, if a sentence's feature vector is (1, 2) and the standard stage vector is (1, 3), the distance is 1. A data monitoring structure template for the park's investment promotion process is constructed. This template is presented in tabular or XML format, specifying the key data items to be monitored, the responsible persons, and the completion deadlines for each business stage. For example, in the "solution negotiation" stage, the template mandates monitoring data fields such as "rent quotation" and "number of days of rent-free period," and sets data integrity verification rules. The final output template serves as the configuration baseline for operation, guiding subsequent automated monitoring tasks.

[0037] Please see Figure 7 A data monitoring system for the investment promotion process in a park, comprising: The Task Statement List module is used to implement S1: extract investment promotion text information from the park, divide the statements, identify combinations of verbs and resource-related nouns, analyze the positional distribution characteristics of verbs, assign semantic tags based on the results, and generate a task statement list. The action-driven semantic recognition module is used to implement S2: extracting text from the park's investment promotion task record, identifying whether the combination of verbs and resource-related nouns matches the task statement list, filtering fragments with action-driven word order, judging completeness, and generating a set of semantically corresponding fragments; The semantic path construction module is used to implement S3: analyze the position of resource-type nouns in the semantic corresponding fragment set, divide the structural sequence, establish arrangement rules, merge fragments with the same semantic direction, identify structural missing intervals between paragraphs, fill in content or connect sentences based on the semantic relationship between adjacent sentences or the transitional expressions in the original text, splice missing intervals, and generate a semantic path division map. The behavior jump graph module is used to implement S4: In the semantic path partitioning graph, a semantic sequence is built according to the text order, the structure of connecting words is extracted, the category is identified and mapped to the word order jump path, and a task behavior association graph is generated. The data monitoring template module is used to implement S5: based on the task behavior association graph, combined with the word order jump path and task content order, merge sentence fragments, reorganize paragraph order, construct a sentence organization structure with monitoring instruction arrangement characteristics, and build a data monitoring structure template for the park investment promotion process.

[0038] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the technical solution.

Claims

1. A method for monitoring data during the investment promotion process in a park, characterized in that, Includes the following steps: S1: Extract investment promotion text information from the park, segment sentences, identify combinations of verbs and resource-related nouns, analyze the positional distribution characteristics of verbs, assign semantic tags based on the results, and generate a list of task sentences; S2: Extract text from the park's investment promotion task record, identify whether the combination of verbs and resource-related nouns matches the task statement list, filter out segments with action-driven word order, determine completeness, and generate a set of semantically corresponding segments; S3: Analyze the position of resource-related nouns in the semantically corresponding fragment set, perform structural sequence division, establish arrangement rules, merge fragments with the same semantic direction, identify structural missing intervals between paragraphs, fill in content or connect sentences based on the semantic relationship between adjacent sentences or the transitional expressions in the original text, splice missing intervals, and generate a semantic path division map. S4: In the semantic path partitioning graph, a semantic sequence is established according to the text order, the structure of connecting words is extracted, the category is identified and mapped to the word order jump path, and a task behavior association graph is generated. S5: Based on the task behavior association graph, combined with the word order jump path and task content order, merge sentence fragments, reorganize paragraph order, construct a sentence organization structure with monitoring instruction arrangement characteristics, and build a data monitoring structure template for the park investment promotion process.

2. The data monitoring method for the investment promotion process in industrial parks according to claim 1, characterized in that, The task statement list includes verb lexical units, resource-type nouns, and word order position tags. The semantic corresponding fragment set includes semantic fragment identifiers, action-dominant markers, and element integrity tags. The semantic path partitioning graph includes resource noun position nodes, semantic direction paragraphs, and path splicing rules. The task behavior association graph includes semantic sequence, connector type classification, and jump path nodes. The park investment promotion process data monitoring structure template includes paragraph structure units, word order recombination patterns, and monitoring instruction sequences.

3. The data monitoring method for the investment promotion process in industrial parks according to claim 1, characterized in that, The word order characterized by action-oriented features refers to a sentence structure centered on a verb, in which the verb performs an action on resource-related nouns in the sentence, reflecting a behavior-oriented word order expression.

4. The data monitoring method for the investment promotion process in industrial parks according to claim 1, characterized in that, The identification of categories and mapping to word order jump paths refers to identifying the functional categories of lexical structures that connect words, and transforming their functional relationships into jump logic paths between text word orders to construct semantic flow relationships.

5. The data monitoring method for the investment promotion process in industrial parks according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Based on the park text information received by the regional input interface, the text is segmented into sentences by period, semicolon and comma, sentences containing both verbs and resource nouns are filtered out, sentence fragments are recorded, and a set of sentences including verbs and resource nouns is generated. S102: Call the set of statements including verbs and resource nouns, parse the position of the verbs in the statements, extract the position information of the verbs at the beginning, middle and end of the sentences, and statistically analyze the distribution features to generate a set of verb position distribution features for the sentences; S103: Based on the verb position distribution feature set of the statement, construct a mapping rule base for structure and semantic tags, match and label the statement structure, assign semantic tags to all statements and output them to obtain a list of task statements.

6. The data monitoring method for the investment promotion process in industrial parks according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Obtain the original text content in the park's investment promotion task record, retrieve the verb and resource-related noun combinations in each statement in the text, and compare them one by one with the semantic combinations in the task statement list to establish a matching mapping structure and obtain a semantic combination matching result set. S202: Based on the semantic combination matching result set, filter the sentence fragments in the region that meet the preset position weight threshold, label them according to the correspondence between verb position distribution and semantic tags, and archive the sentences that meet the preset conditions to obtain a set of action-driven sentence fragments; S203: Invoke the action-driven statement fragment set, and sequentially determine whether the statement has three elements: resource noun, action verb and preset limiting condition. Perform a removal operation on statements with missing elements and establish a semantically corresponding fragment set.

7. The data monitoring method for the investment promotion process in industrial parks according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Call the semantic corresponding fragment set, locate the position of resource-type nouns in the sentence fragment, extract the position index of the verb corresponding to the resource-type noun in each sentence, perform structural sequence partitioning of the sentence based on position nodes, and obtain the resource position structure sequence set; S302: Based on the resource location structure sequence set, call the semantic tags corresponding to the statements, construct the segment classification rules under the same semantic tag, perform position aggregation operation on the statement segments with the same tag, perform sequential arrangement and merging processing, and obtain the semantic classification structure paragraph set; S303: For the semantic classification structure paragraph set, identify the structural missing intervals between paragraphs, collect the end and start information of adjacent sentence fragments in the same type of paragraph, call the preset transition sentence template, perform content completion and connection operations, and establish a semantic path division map.

8. The data monitoring method for the investment promotion process in industrial parks according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Call the semantic path partitioning graph, extract the position index of the sentence fragments in the structural paragraph in the original text, and perform a sequence reconstruction operation on the sentence fragments in the paragraph according to the original text arrangement order to generate the original semantic sequence set; S402: Based on the original semantic sequence set, extract the connection positions between segments, retrieve the conjunctions included at the sentence connection points, identify the categories based on the part-of-speech and functional features of the conjunctions, and record their positions to obtain a conjunction classification tag set; S403: Call the connector classification tag set, map the classified connectors to the position intervals between adjacent statements in the original semantic sequence, construct the jump path structure, and combine the semantic sequence and connection logic relationship to establish a task behavior association graph.

9. The data monitoring method for the investment promotion process in industrial parks according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Obtain the word order jump path and task content order recorded in the task behavior association graph, locate the structural paragraph corresponding to the sentence fragment in the graph, perform order judgment on adjacent fragments and perform structural merging operation to generate a word order structure aggregation set. S502: Based on the aggregated set of word order structure, extract the defined jump logic constraints in each group of segments, identify the position indices of the start and end segments in the paragraphs, rearrange the structural paragraphs according to the path order, and establish a path constraint reorganization sequence set; S503: Call the path constraint recombination sequence set, extract the statement fragments with monitoring instruction arrangement characteristics in sequence, combine them in the order of instructions, and map them to the park investment promotion data flow framework to build a data monitoring structure template for the park investment promotion process.

10. A data monitoring system for the investment promotion process in a park, characterized in that, The system is used to implement the data monitoring method for the investment promotion process in a park as described in any one of claims 1-9, and the system includes: The Task Statement List module is used to implement S1: extract investment promotion text information from the park, divide the statements, identify combinations of verbs and resource-related nouns, analyze the positional distribution characteristics of verbs, assign semantic tags based on the results, and generate a task statement list. The action-driven semantic recognition module is used to implement S2: extracting text from the park's investment promotion task record, identifying whether the combination of verbs and resource-related nouns matches the task statement list, filtering fragments with action-driven word order, judging completeness, and generating a set of semantically corresponding fragments; The semantic path construction module is used to implement S3: analyze the position of resource-type nouns in the set of semantically corresponding fragments, divide the structural sequence, establish arrangement rules, merge fragments with the same semantic direction, identify structural missing intervals between paragraphs, fill in content or connect sentences based on the semantic relationship between adjacent sentences or the transitional expressions in the original text, splice missing intervals, and generate a semantic path division map. The behavior jump graph module is used to implement S4: In the semantic path partitioning graph, a semantic sequence is established according to the text order, the structure of connecting words is extracted, the category is identified and mapped to the word order jump path, and a task behavior association graph is generated. The data monitoring template module is used to implement S5: based on the task behavior association graph, combined with the word order jump path and task content order, merge sentence fragments, reorganize paragraph order, construct a sentence organization structure with monitoring instruction arrangement characteristics, and build a data monitoring structure template for the park investment promotion process.

Images