A plm research material processing method and device fusing semantic understanding

Abstract

The application is suitable for the technical field of text processing, and provides a PLM research and development material processing method and device fusing semantic understanding, the PLM research and development material processing method fusing semantic understanding comprises the following steps: obtaining research and development material text to be recognized, and performing sentence processing on the research and development material text to obtain a plurality of to-be-recognized sentences; extracting the number of entity words and the number of logical connection words in the to-be-recognized sentences; calculating the topic coherence index, the position influence index and the format influence index of the to-be-recognized sentences; extracting key sentences according to the number of entity words, the number of logical connection words, the topic coherence index, the position influence index and the format influence index, and performing labeling processing on the key sentences. This kind of fusing semantic understanding mode ensures that the system not only pays attention to the surface structure of the sentence when extracting information, but also can understand its internal meaning, thereby improving the accuracy and effectiveness of information extraction.

Description

Technical Field

[0001] This invention belongs to the technical field of text processing, and particularly relates to a method and apparatus for processing PLM R&D data that integrates semantic understanding. Background Technology

[0002] Against the backdrop of rapid technological advancements, Product Lifecycle Management (PLM), as a management concept and technical framework, has gradually become an important tool for enterprises to optimize resource allocation, improve efficiency, and reduce costs at various stages, including R&D, design, production, sales, and service. With the continuous increase in R&D data, its form and content have become increasingly complex. How to efficiently and accurately process this data and extract valuable information has become a key challenge in the development of current PLM systems.

[0003] Traditional methods for processing R&D data mainly rely on manual review and experience-based judgment, which suffers from low efficiency and high error rates. With advancements in Natural Language Processing (NLP) technology, more and more automated processing techniques are being introduced into this field. However, existing NLP technologies still have shortcomings in semantic understanding, particularly in the identification and extraction of key information from R&D data, where recognition accuracy remains low and further improvements are needed. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a PLM R&D data processing method and apparatus that integrates semantic understanding, in order to solve the technical problem of low recognition accuracy of traditional methods in identifying and extracting key information in R&D data.

[0005] A first aspect of this invention provides a PLM R&D data processing method incorporating semantic understanding, the method comprising: The research and development data text to be identified is obtained, and the research and development data text is segmented into sentences to obtain multiple sentences to be identified; Extract the number of entity words and logical connectors in the statement to be identified; wherein, entity words include at least one of personal names, place names, organizations, dates, and professional terms; Calculate the topic coherence index, positional influence index, and format influence index of the statement to be identified; wherein, the topic coherence index refers to the degree of correlation between the statement and the topic; Based on the number of entity words, the number of logical connectors, the topic coherence index, the positional influence index, and the format influence index, key sentences are extracted and annotated.

[0006] Furthermore, the steps of calculating the topic coherence index, positional influence index, and format influence index of the statement to be identified include: A semantic flow graph is constructed, and a semantic flow breakage index and a dynamic topic contribution are calculated for each statement to be identified based on the semantic flow graph; wherein, the semantic flow breakage index is used to characterize the degree of impact on semantic coherence when the statement to be identified is missing in the context; and the dynamic topic contribution is used to characterize the degree of contribution of the statement to be identified to the semantic context. Within the same paragraph, the positional influence index of the first statement to be identified is set as the preset baseline weight; Based on the decay function, the positional influence index of the subsequent sentences to be identified is calculated; where, the positional influence index of the subsequent sentences to be identified = B_position*(d (i-1) B_position represents the preset baseline weight, d represents the decay factor, and i represents the order of the subsequent sentences to be identified in the paragraph; Extract the format types present in the statement to be identified, and obtain the preset weights corresponding to the format types; Substitute the preset weights into the preset function to obtain the format influence index; where the format influence index = 1 + α * (1 - 1 / (1 + ΣS_feature)), α represents the format influence factor, and ΣS_feature represents the sum of preset weights corresponding to the format type.

[0007] Furthermore, the step of constructing a semantic flow graph and calculating the semantic flow breakage index and dynamic topic contribution of each statement to be identified based on the semantic flow graph includes: Within the same paragraph, extract the semantic vectors corresponding to each of the multiple sentences to be identified; Calculate the cosine similarity of semantic vectors between adjacent sentences to be identified; The sentences to be identified are treated as nodes in a semantic flow graph, and the cosine similarity between adjacent sentences to be identified is used as the initial weight of the edges between nodes. Arrange and combine the nodes to obtain multiple sets of first nodes; Based on multiple sets of first nodes, the semantic flow breakage index and dynamic topic contribution are calculated for each current statement to be identified.

[0008] Furthermore, the step of calculating the semantic flow breakage index and dynamic topic contribution of each currently identified statement based on multiple first node sets includes: Extract the current minimum initial weight from the first node set; Calculate the first harmonic mean corresponding to multiple current minimum initial weights, and use the first harmonic mean as the complete global coherence score; where, the first harmonic mean = the total number of the first node set / (Σ(1 / multiple current minimum initial weights)); Remove the node corresponding to the current statement to be identified from all nodes to obtain multiple current nodes; Based on the multiple current nodes, a missing global coherence score is calculated; wherein, the complete global coherence score refers to the degree of semantic coherence when all nodes exist, and the missing global coherence score refers to the degree of semantic coherence when the current statement to be identified is indeed present in all nodes. The semantic flow breakage index is obtained by subtracting the complete global coherence score from the missing global coherence score. Calculate the first similarity between two adjacent statements to be identified; Calculate the second similarity between the current statement to be identified and the two adjacent statements to be identified, and take the minimum value of the two second similarities; Subtracting the minimum value from the first similarity score yields the dynamic topic contribution.

[0009] Further, the step of calculating the missing global coherence score based on the plurality of current nodes includes: The current node is arranged and combined to obtain multiple sets of second nodes; Extract the subsequent minimum initial weights from the second node set; Calculate the second harmonic mean corresponding to multiple subsequent minimum initial weights, and use the second harmonic mean as the missing global coherence score; where the second harmonic mean = total number of second node sets / (Σ(1 / multiple subsequent minimum initial weights)).

[0010] Further, the step of extracting key sentences based on the number of entity words, the number of logical connectors, the topic coherence index, the positional influence index, and the format influence index, and then annotating the key sentences, includes: Obtain the document type corresponding to the R&D data text; the document type includes design documents, review documents, or test documents. Match multiple feature weights corresponding to the document type; For each sentence to be identified, the number of entity words, the number of logical connectors, the topic coherence index, the positional influence index, and the format influence index are multiplied by their respective feature weights, and then summed to obtain a comprehensive score. Based on the comprehensive score, key sentences are extracted and marked.

[0011] Furthermore, the step of extracting key sentences based on the comprehensive score and annotating the key sentences includes: The comprehensive score and the ranking corresponding to each sentence to be identified are fitted into a curve; Extract the current comprehensive score corresponding to the position with the maximum curvature in the curve, and use the current comprehensive score as the first threshold; Multiply the current comprehensive score by 0.7 to obtain the second threshold; The statements to be identified that have a comprehensive score greater than the first threshold are designated as core statements. Statements whose overall score falls between the first threshold and the second threshold are designated as supporting key statements. The core statements and the supporting key statements are annotated.

[0012] A second aspect of the present invention provides a PLM R&D data processing device that integrates semantic understanding, comprising: The acquisition unit is used to acquire the R&D data text to be identified, and to process the R&D data text into sentences to obtain multiple sentences to be identified; The extraction unit is used to extract the number of entity words and the number of logical connectors in the statement to be identified; wherein, entity words include at least one of personal names, place names, organizations, dates, and professional terms; The calculation unit is used to calculate the topic coherence index, positional influence index, and format influence index of the statement to be identified; wherein, the topic coherence index refers to the degree of correlation between the statement and the topic; The annotation unit is used to extract key sentences based on the number of entity words, the number of logical connectors, the topic coherence index, the positional influence index, and the format influence index, and to annotate the key sentences.

[0013] A third aspect of the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the PLM R&D data processing method with fused semantic understanding described in the first aspect.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the PLM R&D data processing method with fused semantic understanding described in the first aspect.

[0015] The beneficial effects of this invention compared to existing technologies are as follows: By segmenting the R&D data text to be identified into sentences and extracting entity words and logical connectors from the sentences, this method can systematically analyze and process large amounts of R&D data. This automated processing method significantly improves the efficiency of information processing, enabling the analysis and understanding of large amounts of text data in a shorter time compared to traditional manual review methods. By calculating multiple indicators such as thematic coherence index, positional influence index, and format influence index, the semantic information of sentences can be deeply mined. This fusion of semantic understanding ensures that the system, when extracting information, not only focuses on the surface structure of sentences but also understands their inherent meaning, thereby improving the accuracy and effectiveness of information extraction. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the 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.

[0017] Figure 1 A schematic flowchart of a PLM R&D data processing method integrating semantic understanding provided by the present invention is shown; Figure 2 This diagram illustrates a PLM R&D data processing device that integrates semantic understanding, according to an embodiment of the present invention. Figure 3 A schematic diagram of a terminal device provided in an embodiment of the present invention is shown. Detailed Implementation

[0018] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0019] This invention provides a PLM R&D data processing method and apparatus that integrates semantic understanding to solve the technical problem of low recognition accuracy in identifying and extracting key information in R&D data using traditional methods.

[0020] First, this invention provides a PLM R&D data processing method that integrates semantic understanding. Please refer to... Figure 1 , Figure 1A schematic flowchart illustrating a PLM R&D data processing method incorporating semantic understanding, provided by this invention, is shown. Figure 1 As shown, this PLM R&D data processing method integrating semantic understanding may include the following steps: Step 101: Obtain the R&D document text to be identified, and process the R&D document text into sentences to obtain multiple sentences to be identified; The R&D document text to be identified can be any format of R&D document, such as Word document, PDF, TXT text, etc., and the content may include, but is not limited to, requirements documents, design reports, test plans, meeting minutes and papers.

[0021] Sentence segmentation is a fundamental step in natural language processing. By recognizing punctuation marks such as periods, question marks, exclamation marks, and semicolons, a complete document is divided into individual sentences. Each sentence, as a statement to be identified, becomes the basic object of all subsequent analysis.

[0022] Step 102: Extract the number of entity words and logical connectors in the statement to be identified; wherein, entity words include at least one of personal names, place names, organizations, dates, and professional terms; By counting the number of entities contained in a sentence, we can quantify its information density. Sentences with higher information density are more likely to be important. Entities are objectively existing objects in text that have specific meaning. In the specific field of research and development materials, these entities are particularly important.

[0023] Names and organizations, including project leaders, inventors, and collaborating teams, are crucial for tracing accountability and contributions. Dates mark key milestones and deadlines. Technical jargon is central to the research and development field. Sentences containing numerous technical terms (such as convolutional neural networks, radio frequency amplifiers, and gene editing) are highly likely to explain core technologies, principles, or methods, and are therefore of greater importance.

[0024] Logical connectors express the logical relationships between sentences, such as: however, but, therefore, in conclusion, secondly, or indicating a result. Sentences containing logical connectors often serve to connect paragraphs, summarize, contrast, or reveal causal relationships. For example, "Therefore, we adopted solution A" is a concluding sentence. "However, this solution has a performance bottleneck" points out a key problem. By counting these words, we can identify those sentences that play a crucial role in the logical structure.

[0025] Step 103: Calculate the topic coherence index, positional influence index, and format influence index of the statement to be identified; wherein, the topic coherence index refers to the degree of correlation between the statement and the topic; The topic coherence index measures how closely a sentence is related to the core theme of the entire document or paragraph. It understands the theme of the document or paragraph and then determines how closely each sentence relates to that theme. Sentences with higher relevance have higher index values, indicating that they are more likely to revolve around the core theme and are therefore more important.

[0026] The positional influence index assesses a sentence's importance based on its location within the document. Sentences located at the beginning, beginning, end of paragraphs, in chapter summaries, or in conclusions are typically introductory, summarizing, or concluding statements. Sentences in these advantageous positions are assigned higher index values.

[0027] The format impact index assesses the importance of a sentence based on its text formatting. The system identifies formatting features such as whether the sentence is bold, italicized, underlined, a heading, or an item in a bulleted list.

[0028] Specifically, step 103 includes steps 1031 to 1035: Step 1031: Construct a semantic flow graph, and calculate the semantic flow breakage index and dynamic topic contribution of each statement to be identified based on the semantic flow graph; wherein, the semantic flow breakage index is used to characterize the degree of impact on semantic coherence when the statement to be identified is missing in the context; the dynamic topic contribution is used to characterize the degree of contribution of the statement to be identified in the context semantics; Graph theory and dynamic analysis are used to quantify the importance of sentences in a document's semantic network. The entire document is modeled as a graph network. Each sentence to be identified is a node in this graph. The weight of the edge connecting two nodes represents the semantic similarity between the two sentences. The higher the similarity, the greater the weight of the edge. The result is an interconnected, weighted semantic network that visually demonstrates how ideas flow from one sentence to another in the document.

[0029] The Semantic Flow Disruption Index accurately identifies key sentences that connect different parts of the text and serve as crucial points of convergence. Even if a sentence itself lacks technical jargon or specific formatting, it can still be vital. The importance of a node can be measured by the disruption its removal would cause to the network. In the complete semantic flow graph, the global coherence score is first calculated. A sentence node (and its connected edges) is temporarily removed from the graph. The global coherence score is then recalculated after removing the node. The greater the decrease in this score, the greater the sentence's role in maintaining the semantic coherence of the entire text, and the higher its Semantic Flow Disruption Index.

[0030] Dynamic topic contribution (RTC) identifies sentences that introduce new ideas, propose new methods, draw conclusions, or indicate shifts in thought—these sentences are the engines driving the document's content. The document can be viewed as a sequence (sentence 1 -> sentence 2 -> ... -> sentence N). For sentence i, we can analyze its thematic connections and changes with the local context formed by the preceding k sentences. A sentence has a high RTC if it introduces a new, context-dependent key concept, or provides an important summary or shift in the preceding topic.

[0031] Specifically, step 1031 includes steps A1 to A5: Step A1: Extract the semantic vectors corresponding to each of the multiple sentences to be identified within the same paragraph; The traditional BERT model is used to convert the sentence to be recognized into a semantic vector.

[0032] Step A2: Calculate the cosine similarity of semantic vectors between adjacent sentences to be identified; Adjacent sentences refer to two sentences that are sequentially next to each other in the original text (such as sentence 1 and sentence 2, sentence 2 and sentence 3). The result of this step is to provide edge weights for subsequent graph construction. The more similar the two sentences are, the smoother the semantic flow between them, and the stronger the connection.

[0033] Step A3: Treat the sentences to be identified as nodes in the semantic flow graph, and use the cosine similarity between adjacent sentences to be identified as the initial weights of the edges between nodes; The linear text sequence is abstracted into a non-linear network graph that reflects the internal semantic relationships.

[0034] Step A4: Arrange and combine the nodes to obtain multiple sets of first nodes; Different subsets of sentences (nodes) are systematically generated for subsequent simulation of the impact of node removal on the network.

[0035] Suppose there is a very short document consisting of 5 sentences: [S1, S2, S3, S4, S5].

[0036] Arrange and combine [S1, S2, S3, S4, S5] to obtain the following sets of first nodes: S1 -> S2: Path [S1-S2]; S1 -> S3: Path [S1-S2-S3]; S1 -> S4: Path [S1-S2-S3-S4]; S1 -> S5: Path [S1-S2-S3-S4-S5]; S2 -> S3: Path [S2-S3]; S2 -> S4: Path [S2-S3-S4]; S2 -> S5: Path [S2-S3-S4-S5]; S3 -> S4: Path [S3-S4]; S3 -> S5: Path [S3-S4-S5]; S4 -> S5: Path [S4-S5]. (The reverse path is similar, since it's an undirected graph.) Step A5: Based on multiple sets of first nodes, calculate the semantic flow breakage index and dynamic topic contribution for each current statement to be identified.

[0037] By utilizing the principles of graph theory, we can extract the deeper importance of sentences from the network structure.

[0038] In the embodiments corresponding to steps A1 to A5, a systematic method is provided for calculating the semantic flow breakage index and dynamic topic contribution of the sentence to be identified by extracting semantic vectors, calculating similarity, constructing a semantic flow graph, and analyzing the node set. This method not only enhances the understanding of text but also lays the foundation for information processing and analysis.

[0039] Specifically, step A5 includes steps A51 to A58: Step A51: Extract the current minimum initial weight from the first node set; Find the edge with the lowest weight in the complete graph; this is the current minimum initial weight. This minimum value represents the weakest link in the entire semantic network.

[0040] Suppose we have a very short document consisting of 5 sentences: [S1, S2, S3, S4, S5]. First, we construct their semantic flow graph, where the weights of the edges are the cosine similarity between the sentences (an assumed value is used here for demonstration purposes). This graph can be intuitively represented as: S1--0.7-->S2--0.9-->S3--0.2-->S4--0.8-->S5.

[0041] S1 -> S2: Path [S1-S2], current minimum initial weight = min(0.7) = 0.7; S1 -> S3: Path [S1-S2-S3], current minimum initial weight = min(0.7, 0.9) = 0.7; S1 -> S4: Path [S1-S2-S3-S4], current minimum initial weight = min(0.7, 0.9, 0.2) = 0.2; S1 -> S5: Path [S1-S2-S3-S4-S5], current minimum initial weight = min(0.7, 0.9, 0.2, 0.8) = 0.2; S2 -> S3: Path [S2-S3], current minimum initial weight = min(0.9) = 0.9; S2 -> S4: Path [S2-S3-S4], current minimum initial weight = min(0.9, 0.2) = 0.2; S2 -> S5: Path [S2-S3-S4-S5], current minimum initial weight = min(0.9, 0.2, 0.8) = 0.2; S3 -> S4: Path [S3-S4], current minimum initial weight = min(0.2) = 0.2; S3 -> S5: Path [S3-S4-S5], current minimum initial weight = min(0.2, 0.8) = 0.2; S4 -> S5: Path [S4-S5], current minimum initial weight = min(0.8) = 0.8.

[0042] Step A52: Calculate the first harmonic mean corresponding to multiple current minimum initial weights, and use the first harmonic mean as the complete global coherence score; where, the first harmonic mean = total number of the first node set / (Σ(1 / multiple current minimum initial weights)); The harmonic mean is characterized by a strong bias towards smaller values. In a set of values, even a single very small value can significantly lower the overall mean. The coherence of the entire semantic flow is not determined by the average connection strength, but rather by the weakest connections.

[0043] In the example above, there are N=5 nodes, and a total of N(N-1) / 2 = 10 node pairs.

[0044] List all current minimum initial weights: [0.7, 0.7, 0.2, 0.2, 0.9, 0.2, 0.2, 0.2, 0.2, 0.8].

[0045] First harmonic mean = n / (Σ(1 / x) i )). x i This represents the i-th current minimum initial weight.

[0046] Where, Σ(1 / x i ) = 1 / 0.7 + 1 / 0.7 + 1 / 0.2 + 1 / 0.2 + 1 / 0.9 + 1 / 0.2 + 1 / 0.2 + 1 / 0.2 + 1 / 0.2 + 1 / 0.8 = 1.428 + 1.428 + 5 + 5 + 1.111 + 5 + 5 + 5 + 5 + 1.25 = 35.225 The first harmonic mean is approximately 10 / 35.225 ≈ 0.284.

[0047] Step A53: Remove the node corresponding to the current statement to be identified from all nodes to obtain multiple current nodes; Remove the sentence node we are evaluating from the complete graph (denoted as node K). Since node K is removed, the edges that were originally connected to K also disappear. This forms a new, incomplete semantic flow graph.

[0048] Step A54: Calculate the missing global coherence score based on the multiple current nodes; wherein, the complete global coherence score refers to the degree of semantic coherence when all nodes exist, and the missing global coherence score refers to the degree of semantic coherence when the current statement to be identified is indeed present in all nodes. In this new incomplete graph, the edges with the lowest weights are found again, and then the harmonic mean of these minimum weights is calculated. This yields the missing global coherence score.

[0049] Specifically, step A54 includes steps A541 to A543: Step A541: Arrange and combine the current node to obtain multiple sets of second nodes; The current node refers to all the nodes remaining after removing the sentence node to be evaluated from the complete graph. These nodes form a new, smaller semantic flow graph (called the incomplete graph). The same method is used again for this incomplete graph composed of the remaining nodes. This step ensures that the graph is deconstructed and analyzed in the same way regardless of whether it is complete or incomplete, laying the foundation for fair comparisons later.

[0050] Step A542: Extract the subsequent minimum initial weights from the second node set respectively; In each second set of nodes (i.e., a local part of the incomplete graph), find the edge with the lowest weight. This weight is called the subsequent minimum initial weight. This step corresponds one-to-one with extracting the current minimum initial weight from the first set of nodes when calculating the complete global coherence score.

[0051] Step A543: Calculate the second harmonic mean corresponding to multiple subsequent minimum initial weights, and use the second harmonic mean as the missing global coherence score; where, the second harmonic mean = the total number of the second node set / (Σ(1 / multiple subsequent minimum initial weights)).

[0052] The second harmonic mean quantifies the minimum level of coherence that the entire document's semantic flow can maintain in the absence of a key sentence. This "weakest link" effect is used to evaluate both complete and incomplete networks. This ensures that the scores for complete global coherence and missing global coherence are comparable.

[0053] For example, suppose we have a very simple document consisting of 5 sentences: [S1, S2, S3, S4, S5]. Removing the target sentence S3 from the graph and sequence gives us the new sentence sequence: [S1, S2, S4, S5]. Similarly, we calculate the strongest path weights between all node pairs (now there are only 4 nodes and 6 node pairs).

[0054] S1 -> S2: Subsequent minimum initial weight = 0.7; S1 -> S4: Path [S1-S2-S4], subsequent minimum initial weight = min(0.7, 0.3) = 0.3; S1 -> S5: Path [S1-S2-S4-S5], subsequent minimum initial weight = min(0.7, 0.3, 0.8) = 0.3; S2 -> S4: Subsequent minimum initial weight = 0.3; S2 -> S5: Path [S2-S4-S5], subsequent minimum initial weight = min(0.3, 0.8) = 0.3; S4 -> S5: Subsequent minimum initial weight = 0.8; (The same applies to the reverse path, since it is an undirected graph) All subsequent minimum initial weights: [0.7, 0.3, 0.3, 0.3, 0.3, 0.8]; Σ(1 / x i = 1 / 0.7 + 1 / 0.3 + 1 / 0.3 + 1 / 0.3 + 1 / 0.3 + 1 / 0.8 = 1.428 + 3.333 + 3.333 + 3.333 + 3.333 + 1.25 = 16.010 The second harmonic mean = 6 / 16.010 ≈ 0.375.

[0055] In the embodiments corresponding to steps A541 to A543, the semantic coherence of the remaining nodes after removing a certain statement is systematically analyzed by permuting and combining the current nodes. The impact of this missing statement on global coherence can be quantified by extracting the subsequent minimum initial weight and calculating its harmonic mean. The harmonic mean is used instead of the arithmetic mean because it is more sensitive to smaller values ​​and can better capture weak points in semantic coherence. The calculation result of the missing global coherence score can be compared with the complete global coherence score to clarify the importance of the currently identified statement in the overall semantics.

[0056] Step A55: Subtract the complete global coherence score from the missing global coherence score to obtain the semantic flow breakage index; If node K is unimportant, removing it may not significantly change the weakest link in the entire network. Therefore, the two scores will differ very little, resulting in a low fracture index.

[0057] If node K is very important, it may be a key node connecting two important subnetworks. Removing it may create a new connection in the incomplete graph that is much weaker than the original weakest link, resulting in a significant decrease in the missing global coherence score.

[0058] Semantic Flow Discontinuity Index = Complete Global Coherence Score - Missing Global Coherence Score. A sentence with a high semantic flow discontinuity index is not necessarily the one with the highest information density, but it is certainly structurally indispensable. Its absence will lead to a precipitous drop in semantic understanding.

[0059] Step A56: Calculate the first similarity between two adjacent statements to be identified; In the semantic flow graph, the cosine similarity between the preceding (K-1) and following (K+1) nodes of the current sentence (node ​​K), which are not directly connected, is directly calculated. This measures the fluency of the context if node K were skipped. If the preceding and following texts are already very similar, it indicates that node K may only provide supplementary information and has a small contribution.

[0060] Step A57: Calculate the second similarity between the current statement to be identified and the two adjacent statements to be identified, and take the minimum value of the two second similarities; Calculate the similarity between node K and node K-1, and the similarity between node K and node K+1, and then take the smaller of the two values.

[0061] This measures how strongly the current sentence maintains its connection with the context, at least on one side. Taking the minimum value is a conservative strategy that focuses on the weakest link between the sentence and its context.

[0062] Step A58: Subtract the minimum value from the first similarity to obtain the dynamic topic contribution.

[0063] If the contribution is negative or a very small positive number, it means that the integration of the current sentence with the context is roughly the same as the coherence of the context itself. This suggests that the current sentence may not introduce new, key information, but simply follows the existing semantic flow.

[0064] If the contribution is a large positive number, it means that the current sentence significantly alters the semantic flow. It may introduce a new concept, a transition, or a conclusion, making it essential for connecting the sentences before and after it. It is this sentence that causes a valuable leap or deepening of semantics before and after it.

[0065] In the embodiments corresponding to steps A51 to A58, the semantic flow breakage index and dynamic topic contribution are systematically calculated by comprehensively considering the similarity between statements and their impact on global semantic coherence. This method effectively captures the semantic structure of the text, helping to understand the importance and role of each statement in the overall semantics.

[0066] Step 1032: Within the same paragraph, set the positional influence index of the first sentence to be identified as the preset baseline weight; Set the weight of the first sentence of the paragraph to B_position (for example, set it to 1.0).

[0067] Step 1033: Based on the decay function, calculate the positional influence index of the subsequent sentences to be identified; where, the positional influence index of the subsequent sentences to be identified = B_position*(d (i-1) B_position represents the preset baseline weight, d represents the decay factor, and i represents the order of the subsequent sentences to be identified in the paragraph; An exponential decay function is used. Starting from the second sentence (i=2), its position weight = B_position * (d ^(i-1)).

[0068] d is the decay factor, a constant between 0 and 1 (e.g., 0.8). i is the sentence's position number within the paragraph.

[0069] Calculation example: Assume B_position=1.0, d=0.8; Sentence 1 (paragraph beginning): Weight = 1.0; Sentence 2: Weight = 1.0 * (0.8) 1 = 0.8; Sentence 3: Weight = 1.0 * (0.8) 2 = 0.64; Sentence 4: Weight = 1.0 * (0.8) 3 = 0.512; This model transforms a qualitative experience (precedence = importance) into a precise, repeatable quantitative indicator.

[0070] As an optional embodiment of this application, the last sentence is a conclusion or a transitional sentence, and its weight is multiplied by a last sentence compensation coefficient e (e.g., e=1.2) based on the attenuation calculation. Weight_last sentence = (B_position * (d^(n-1))) * e (n is the total number of sentences in the block).

[0071] Step 1034: Extract the format types present in the statement to be identified, and obtain the preset weights corresponding to the format types; Assign a preset weight S_feature to each format type, for example: Step 1035: Substitute the preset weights into the preset function to obtain the format influence index; wherein, the format influence index = 1 + α * (1 - 1 / (1 + ΣS_feature)), α represents the format influence factor, and ΣS_feature represents the sum of preset weights corresponding to the format type.

[0072] ΣS_feature is a simple sum of the weights of all formatting in a sentence. When there is no formatting, ΣS_feature = 0, then the formatting influence index = 1 + α * (1 - 1 / (1+0)) = 1 + α * (1-1) = 1. This means that the baseline value for no formatting is 1.

[0073] When there is a format, as ΣS_feature increases from 0, 1 / (1 +ΣS_feature) will decrease, causing (1 - 1 / (1 +ΣS_feature)) to increase, thus the exponent is greater than 1.

[0074] This is a saturation function. When ΣS_feature is small, the exponential growth is rapid. When ΣS_feature is very large, 1 / (1 +ΣS_feature) approaches 0, and the entire exponent approaches 1 + α. This means that no matter how many formats a sentence has, its format influence exponent has an upper limit (1 + α), preventing a sentence from receiving an unreasonably high score due to too many formats.

[0075] α is an adjustable coefficient used to control the overall influence of format features in the final comprehensive score. If format is considered less important, α can be set smaller (e.g., 0.1); if format is considered very important, α can be set larger (e.g., 0.5).

[0076] For a plain statement without any formatting, ΣS_feature=0, then the format weight=1, which serves as the baseline.

[0077] As the number of formats increases, the overall weight increases smoothly and gradually, avoiding score explosion. Example: A statement in bold only: ΣS_feature=1.0 -> Format weight ≈ 1 + 1.5 * (1 - 1 / 2) = 1.75 A statement that is simultaneously bold, highlighted, and red: ΣS_feature=1.0+0.7+0.8=2.5 -> Format weight≈1+1.5*(1-1 / 3.5)≈2.07.

[0078] In the embodiments corresponding to steps 1031 to 1035, a comprehensive framework is provided for calculating the topic coherence, positional influence, and format influence of the sentence to be identified by constructing a semantic flow graph and introducing various influencing factors (such as semantic flow breaks, position, and format). This method not only enhances the depth of text analysis but also lays the foundation for subsequent semantic understanding and information extraction.

[0079] Step 104: Extract key sentences based on the number of entity words, the number of logical connectors, the topic coherence index, the positional influence index, and the format influence index, and annotate the key sentences.

[0080] The system calculates a comprehensive score based on the number of entity words, logical words, topic index, position index, and format index. Based on this comprehensive score, the system sets a threshold, classifying sentences with scores above the threshold as key sentences.

[0081] Key sentences are highlighted in the original text, generating a separate summary or list of key points, and tagged with tags (such as core technology, project risks, experimental conclusions, etc.).

[0082] Specifically, step 104 includes steps 1041 to 1044: Step 1041: Obtain the document type corresponding to the R&D data text; the document type includes design document, review document, or test document; The focus of information differs significantly at different stages of development. This is achieved through document metadata (such as filenames and templates), user specifications, or by analyzing document content (e.g., test documents containing test cases). This provides context for subsequent decisions. The system needs to know what type of document it is processing in order to evaluate it using the correct criteria.

[0083] Step 1042: Match multiple feature weights corresponding to the document type; The system has a pre-defined weight configuration library. For example, design documents place greater emphasis on technical terms (core technologies) and logical connectors (design logic), while the weight of positional influence (first sentence of paragraphs) is relatively low because design descriptions may be distributed throughout the document. Review documents prioritize names of people and organizations (clearly identifying responsible parties) and thematic coherence (whether the document revolves around the topic). Test documents highly value formatting influences (such as test case numbers and bolded steps) and logical connectors.

[0084] This step reflects static configuration; the system calls the corresponding weight combination based on the identified type.

[0085] This system predefines several document types (such as technical design reports, project progress summaries, and defect analysis reports) and configures an optimal feature weight vector W_type for each type.

[0086] The document type can be automatically identified using a lightweight document classifier (based on keywords or a text classification model), or it can be manually specified by the user.

[0087] Example: For technical design reports, W_technical = [0.25, 0.15, 0.35, 0.15, 0.10] (emphasizing thematic coherence and entities). For project progress minutes, W_progress = [0.15, 0.25, 0.20, 0.25, 0.15] (emphasizing logical connectors and positions to capture timelines and decision points).

[0088] Step 1043: For each sentence to be identified, multiply the number of entity words, the number of logical connectors, the topic coherence index, the positional influence index, and the format influence index by their respective feature weights, and add them together to obtain a comprehensive score; Overall score = (Number of entity words * W_entity) + (Number of logical words * W_logic) + (Topic index * W_topic) + (Position index * W_position) + (Format index * W_format).

[0089] Step 1044: Based on the comprehensive score, extract key sentences and mark them.

[0090] In the embodiments corresponding to steps 1041 to 1044, this step systematically evaluates the importance of each sentence to be identified by combining different features and weights. This method can effectively extract sentences with real informational value from a large amount of text. Matching feature weights is a crucial step, ensuring that the context of different document types is fully considered, thereby improving the accuracy and relevance of the extraction. A data-driven approach is provided through comprehensive scoring to identify key sentences, avoiding the bias of subjective judgment. The annotation process not only enhances the identifiability of key sentences but also lays the foundation for subsequent document processing (such as summary generation and information retrieval).

[0091] Specifically, step 1044 includes steps B1 to B6: Step B1: Fit a curve to the comprehensive score and the ranking corresponding to each sentence to be identified; The discrete sentence rating data is transformed into a continuous distribution graph whose shape can be analyzed to discover patterns. The horizontal axis (X) represents the sentence ranking (from 1st to 1st). The vertical axis (Y) represents the overall score of the corresponding sentence. This curve typically shows a sharp drop followed by a flattening pattern. A few sentences at the top of the ranking have very high scores, then the scores drop rapidly, and in the middle and later stages, many sentences have very low scores that are close to each other.

[0092] Example: The first-place score is 90, the second is 85, the third is 80, then the fourth suddenly drops to 50, and from the fifth to the last place, the scores slowly decrease between 20 and 50. This curve clearly illustrates the gap between important and unimportant sentences.

[0093] Step B2: Extract the current comprehensive score corresponding to the position with the maximum curvature in the curve, and use the current comprehensive score as the first threshold; In mathematics, curvature measures the degree of bending of a curve. A greater curvature indicates a more drastic change in the curve's direction at that point. The system calculates the curvature at every point on this fitted curve. The location of maximum curvature corresponds to the steepest point of descent in the score distribution. This point represents the point where the score drops precipitously. The score corresponding to this point (the current overall score) is defined as the first threshold. After this point, the importance of sentences decreases significantly. This threshold is not set manually but is determined by the document's own quality distribution. This allows the method to adapt to documents of varying lengths and qualities.

[0094] Step B3: Multiply the current comprehensive score by 0.7 to obtain the second threshold; It scales down the first threshold (representing the highest standard) by 0.7 to capture supporting statements that, while not as important as the core sentence, are still quite significant.

[0095] Step B4: Select the statements to be identified with a comprehensive score greater than the first threshold as core statements; Step B5: Identify statements whose overall scores fall between the first threshold and the second threshold as supporting key statements; The core statements are located in a very small number of sentences to the left of the point of maximum curvature. These are the soul and foundation of the document, containing the most essential technologies, conclusions, or decisions.

[0096] Supporting key statements provide explanations, arguments, examples, or details to the core statements. They are also key statements in themselves, but belong to the second tier.

[0097] Step B6: Annotate the core statements and the supporting key statements.

[0098] Key statements should be highlighted more prominently, for example, with red highlighting and the label "[Key]". Supporting statements should be highlighted in a secondary manner, for example, with yellow highlighting and the label "[Support]".

[0099] In the embodiments corresponding to steps B1 to B6, this series of steps combines comprehensive scoring with curve fitting, using a mathematical model to determine the thresholds for core and supporting statements, ensuring the rationality and accuracy of the extraction. By setting two different thresholds, the system can more effectively capture information hierarchy, making the classification of core and supporting statements clearer and meeting users' multi-level information needs. This method not only improves the efficiency of extracting key statements but also makes the extraction process transparent, allowing users to understand and trust the logical basis of the extracted information.

[0100] In the embodiments corresponding to steps 101 to 104, by segmenting the R&D data text to be identified into sentences and extracting entity words and logical connectors from the sentences, this method can systematically analyze and process large amounts of R&D data. This automated processing method significantly improves the efficiency of information processing, and compared with traditional manual review methods, it can complete the analysis and understanding of large amounts of text data in a shorter time. By calculating multiple indicators such as topic coherence index, positional influence index, and format influence index, the semantic information of sentences can be deeply mined. This fusion of semantic understanding ensures that when extracting information, the system not only focuses on the surface structure of sentences but also understands their inherent meaning, thereby improving the accuracy and effectiveness of information extraction.

[0101] like Figure 2 This invention provides a PLM R&D data processing device that integrates semantic understanding. Please refer to [link / reference]. Figure 2 , Figure 2 This diagram illustrates a PLM R&D data processing device integrating semantic understanding provided by the present invention, as shown below. Figure 2 The PLM R&D data processing device integrating semantic understanding shown includes: The acquisition unit 21 is used to acquire the R&D data text to be identified, and to perform sentence segmentation on the R&D data text to obtain multiple sentences to be identified; Extraction unit 22 is used to extract the number of entity words and the number of logical connectors in the statement to be identified; wherein, entity words include at least one of personal names, place names, organizations, dates and professional terms; The calculation unit 23 is used to calculate the topic coherence index, positional influence index, and format influence index of the statement to be identified; wherein, the topic coherence index refers to the degree of correlation between the statement and the topic; The annotation unit 24 is used to extract key sentences based on the number of entity words, the number of logical connectors, the topic coherence index, the positional influence index, and the format influence index, and to annotate the key sentences.

[0102] This invention provides a PLM (Professional Management Model) R&D data processing device that integrates semantic understanding. By segmenting the R&D data text to be identified into sentences and extracting entity words and logical connectors, this method can systematically analyze and process large amounts of R&D data. This automated processing significantly improves information processing efficiency, enabling the analysis and understanding of large amounts of text data in a shorter time compared to traditional manual review methods. By calculating multiple indicators such as topic coherence index, positional influence index, and format influence index, the semantic information of sentences can be deeply mined. This integrated semantic understanding approach ensures that the system not only focuses on the surface structure of sentences but also understands their inherent meaning when extracting information, thereby improving the accuracy and effectiveness of information extraction.

[0103] Figure 3 This is a schematic diagram of a terminal device provided in an embodiment of the present invention. Figure 3 As shown, a terminal device 3 in this embodiment includes: a processor 30, a memory 31, and a computer program 32 stored in the memory 31 and executable on the processor 30, such as a PLM R&D data processing program that integrates semantic understanding. When the processor 30 executes the computer program 32, it implements the steps described in the above embodiments of the PLM R&D data processing method that integrates semantic understanding, for example... Figure 1 Steps 101 to 104 are shown. Alternatively, when the processor 30 executes the computer program 32, it implements the functions of each unit in the above-described device embodiments, for example... Figure 2The function of the unit shown.

[0104] For example, the computer program 32 can be divided into one or more units, which are stored in the memory 31 and executed by the processor 30 to complete the present invention. The one or more units can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program 32 in the terminal device 3. For example, the specific functions of each unit of the computer program 32 can be divided as follows: The acquisition unit is used to acquire the R&D data text to be identified, and to process the R&D data text into sentences to obtain multiple sentences to be identified; The extraction unit is used to extract the number of entity words and the number of logical connectors in the statement to be identified; wherein, entity words include at least one of personal names, place names, organizations, dates, and professional terms; The calculation unit is used to calculate the topic coherence index, positional influence index, and format influence index of the statement to be identified; wherein, the topic coherence index refers to the degree of correlation between the statement and the topic; The annotation unit is used to extract key sentences based on the number of entity words, the number of logical connectors, the topic coherence index, the positional influence index, and the format influence index, and to annotate the key sentences.

[0105] The terminal device includes, but is not limited to, a processor 30 and a memory 31. Those skilled in the art will understand that... Figure 3 This is merely an example of a terminal device 3 and does not constitute a limitation on a terminal device 3. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal device may also include input / output devices, network access devices, buses, etc.

[0106] The processor 30 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0107] The memory 31 can be an internal storage unit of the terminal device 3, such as a hard disk or memory of the terminal device 3. The memory 31 can also be an external storage device of the terminal device 3, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device 3. Furthermore, the memory 31 can include both internal and external storage units of the terminal device 3. The memory 31 is used to store the computer program and other programs and data required by the roaming control device. The memory 31 can also be used to temporarily store data that has been output or will be output.

[0108] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0109] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0110] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this invention. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0111] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0112] This invention provides a computer program product that, when run on a mobile terminal, enables the mobile terminal to implement the steps described in the above-described method embodiments.

[0113] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0114] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0115] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0116] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / network devices and methods can be implemented in other ways. For example, the apparatus / network device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0117] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units.

[0118] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0119] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0120] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [the described condition or event] is detected" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once [the described condition or event] is detected," or "in response to detection of [the described condition or event]."

[0121] Furthermore, in the description of this invention and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0122] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including, but not limited to," unless otherwise specifically emphasized.

[0123] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A PLM R&D data processing method integrating semantic understanding, characterized in that, The PLM R&D data processing method that integrates semantic understanding includes: The research and development data text to be identified is obtained, and the research and development data text is segmented into sentences to obtain multiple sentences to be identified; Extract the number of entity words and logical connectors in the statement to be identified; wherein, entity words include at least one of personal names, place names, organizations, dates, and professional terms; Calculate the topic coherence index, positional influence index, and format influence index of the statement to be identified; wherein, the topic coherence index refers to the degree of correlation between the statement and the topic; Based on the number of entity words, the number of logical connectors, the topic coherence index, the positional influence index, and the format influence index, key sentences are extracted and annotated.

2. The PLM R&D data processing method integrating semantic understanding as described in claim 1, characterized in that, The steps for calculating the topic coherence index, positional influence index, and format influence index of the sentence to be identified include: A semantic flow graph is constructed, and a semantic flow breakage index and a dynamic topic contribution are calculated for each statement to be identified based on the semantic flow graph; wherein, the semantic flow breakage index is used to characterize the degree of impact on semantic coherence when the statement to be identified is missing in the context; and the dynamic topic contribution is used to characterize the degree of contribution of the statement to be identified to the semantic context. Within the same paragraph, the positional influence index of the first statement to be identified is set as the preset baseline weight; Based on the decay function, the positional influence index of the subsequent sentences to be identified is calculated; where, the positional influence index of the subsequent sentences to be identified = B_position*(d (i-1) B_position represents the preset baseline weight, d represents the decay factor, and i represents the order of the subsequent sentences to be identified in the paragraph; Extract the format types present in the statement to be identified, and obtain the preset weights corresponding to the format types; Substitute the preset weights into the preset function to obtain the format influence index; where the format influence index = 1 + α * (1 - 1 / (1 + ΣS_feature)), α represents the format influence factor, and ΣS_feature represents the sum of preset weights corresponding to the format type.

3. The PLM R&D data processing method integrating semantic understanding as described in claim 2, characterized in that, The steps of constructing a semantic flow graph and calculating the semantic flow breakage index and dynamic topic contribution of each statement to be identified based on the semantic flow graph include: Within the same paragraph, extract the semantic vectors corresponding to each of the multiple sentences to be identified; Calculate the cosine similarity of semantic vectors between adjacent sentences to be identified; The sentences to be identified are treated as nodes in a semantic flow graph, and the cosine similarity between adjacent sentences to be identified is used as the initial weight of the edges between nodes. Arrange and combine the nodes to obtain multiple sets of first nodes; Based on multiple sets of first nodes, the semantic flow breakage index and dynamic topic contribution are calculated for each current statement to be identified.

4. The PLM R&D data processing method integrating semantic understanding as described in claim 3, characterized in that, The step of calculating the semantic flow breakage index and dynamic topic contribution of each currently identified statement based on multiple first node sets includes: Extract the current minimum initial weight from the first node set; Calculate the first harmonic mean corresponding to multiple current minimum initial weights, and use the first harmonic mean as the complete global coherence score; where, the first harmonic mean = the total number of the first node set / (Σ(1 / multiple current minimum initial weights)); Remove the node corresponding to the current statement to be identified from all nodes to obtain multiple current nodes; Based on the multiple current nodes, a missing global coherence score is calculated; wherein, the complete global coherence score refers to the degree of semantic coherence when all nodes exist, and the missing global coherence score refers to the degree of semantic coherence when the current statement to be identified is indeed present in all nodes. The semantic flow breakage index is obtained by subtracting the complete global coherence score from the missing global coherence score. Calculate the first similarity between two adjacent statements to be identified; Calculate the second similarity between the current statement to be identified and the two adjacent statements to be identified, and take the minimum value of the two second similarities; Subtracting the minimum value from the first similarity score yields the dynamic topic contribution.

5. The PLM R&D data processing method integrating semantic understanding as described in claim 3, characterized in that, The step of calculating the missing global coherence score based on the plurality of current nodes includes: The current node is arranged and combined to obtain multiple sets of second nodes; Extract the subsequent minimum initial weights from the second node set; Calculate the second harmonic mean corresponding to multiple subsequent minimum initial weights, and use the second harmonic mean as the missing global coherence score; where the second harmonic mean = total number of second node sets / (Σ(1 / multiple subsequent minimum initial weights)).

6. The PLM R&D data processing method integrating semantic understanding as described in claim 1, characterized in that, The steps of extracting key sentences based on the number of entity words, the number of logical connectors, the topic coherence index, the positional influence index, and the format influence index, and then annotating the key sentences, include: Obtain the document type corresponding to the R&D data text; the document type includes design documents, review documents, or test documents. Match multiple feature weights corresponding to the document type; For each sentence to be identified, the number of entity words, the number of logical connectors, the topic coherence index, the positional influence index, and the format influence index are multiplied by their respective feature weights, and then summed to obtain a comprehensive score. Based on the comprehensive score, key sentences are extracted and marked.

7. The PLM R&D data processing method integrating semantic understanding as described in claim 6, characterized in that, The steps of extracting key sentences based on the comprehensive score and annotating the key sentences include: The comprehensive score and the ranking corresponding to each sentence to be identified are fitted into a curve; Extract the current comprehensive score corresponding to the position with the maximum curvature in the curve, and use the current comprehensive score as the first threshold; Multiply the current comprehensive score by 0.7 to obtain the second threshold; The statements to be identified that have a comprehensive score greater than the first threshold are designated as core statements. Statements whose overall score falls between the first threshold and the second threshold are designated as supporting key statements. The core statements and the supporting key statements are annotated.

8. A PLM R&D data processing device integrating semantic understanding, characterized in that, The PLM R&D data processing device that integrates semantic understanding includes: The acquisition unit is used to acquire the R&D data text to be identified, and to process the R&D data text into sentences to obtain multiple sentences to be identified; The extraction unit is used to extract the number of entity words and the number of logical connectors in the statement to be identified; wherein, entity words include at least one of personal names, place names, organizations, dates, and professional terms; The calculation unit is used to calculate the topic coherence index, positional influence index, and format influence index of the statement to be identified; wherein, the topic coherence index refers to the degree of correlation between the statement and the topic; The annotation unit is used to extract key sentences based on the number of entity words, the number of logical connectors, the topic coherence index, the positional influence index, and the format influence index, and to annotate the key sentences.

9. A terminal device, characterized in that, The terminal device includes: a memory, a processor, and a PLM R&D data processing program with fusion semantic understanding stored in the memory and executable on the processor, wherein the PLM R&D data processing program with fusion semantic understanding is configured to implement the steps in the PLM R&D data processing method with fusion semantic understanding as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps in the PLM R&D data processing method with fused semantic understanding as described in any one of claims 1 to 7.

Images