Supplier relationship management method and system combined with big data analysis

Abstract

The invention provides a supplier relation management method and system combined with big data analysis, and the method comprises the steps: firstly constructing a session text semantic association network based on historical session text big data, and then mining a supplier interaction intention set based on the session text semantic association network; calling a natural language processing model to jointly analyze the intention set and the network, predicting a supplier relation state, positioning an optimized and improved node according to a prediction result, and finally generating a supplier relation dynamic management and control mechanism based on the optimized and improved node and the supplier interaction intention set and applying the mechanism to a cooperation process. Therefore, the continuous maintenance and adjustment of the intelligent manufacturing production line robot related field supplier relationship can be realized.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and in particular to a supplier relationship management method and system combined with big data analysis. Background Art

[0002] The cooperative relationship between enterprises and suppliers is crucial for the stable operation and efficient output of production lines. Currently, enterprises primarily rely on manual review and analysis of communication records with suppliers to manage supplier relationships. These records contain a large amount of conversational text information, but traditional processing methods only perform superficial classification and statistics on the text, making it difficult to deeply explore the rich information hidden behind the text. For example, manual analysis often fails to fully and accurately capture and interpret the cooperation needs, potential problems, and risk warnings expressed by suppliers in conversations. Furthermore, existing methods lack effective utilization of semantic connections in conversational text, failing to fully grasp the interaction logic and relationship context between suppliers and enterprises. This means that enterprises lack a scientific and systematic basis for evaluating the status of supplier relationships, making it difficult to predict potential relationship changes in advance. When problems arise in supplier relationships, they are unable to quickly identify key factors and implement targeted optimization and improvement measures. This hinders the smooth development of intelligent manufacturing production line robot-related businesses and the long-term stable development of enterprises. Summary of the Invention

[0003] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a supplier relationship management method combined with big data analysis, the method comprising:

[0004] Constructing a conversation text semantic association network based on historical conversation text big data between suppliers and enterprises, comprising multiple conversation text units and semantic association edges between conversation text units, wherein the association strength of the semantic association edges is determined according to the closeness of the semantic association between the conversation text units;

[0005] Based on the conversation text semantic association network, mining the supplier interaction intention in the conversation text unit to obtain the supplier interaction intention set;

[0006] Invoking a natural language processing model to jointly analyze the supplier interaction intention set and the conversation text semantic association network to predict the relationship status between the supplier and the enterprise, thereby obtaining a supplier relationship status prediction result;

[0007] Locating optimization and improvement nodes in the supplier relationship based on the supplier relationship status prediction results, wherein the optimization and improvement nodes are conversation text-related nodes that affect the stability of the relationship or lead to potential relationship changes;

[0008] Based on the optimization and improvement nodes and the supplier interaction intention set, a supplier relationship dynamic management and control mechanism is generated, and the supplier relationship dynamic management and control mechanism is applied to the supplier cooperation process.

[0009] On the other hand, an embodiment of the present invention also provides a supplier relationship management system combined with big data analysis, including a processor and a machine-readable storage medium, the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0010] Based on the above aspects, the embodiment of the present invention constructs a conversation text semantic association network, based on historical conversation text big data, describes the semantic association of conversation text units between suppliers and enterprises, mines supplier interaction intentions based on the conversation text semantic association network, and obtains a set containing multiple key intentions. The natural language processing model is called to jointly analyze the interaction intention set and the semantic association network to predict the relationship status between suppliers and enterprises. The optimization and improvement nodes are located according to the relationship status prediction results, which can quickly lock in the key factors affecting the relationship and avoid blind decision-making. Finally, a dynamic supplier relationship management and control mechanism is generated based on the optimization and improvement nodes and the interaction intention set, and it is applied to the supplier cooperation process, so that the supplier relationship can be continuously maintained and adjusted as the actual situation changes, effectively improving the stability of the supplier relationship and the cooperation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a schematic diagram of the execution flow of the supplier relationship management method combined with big data analysis provided by an embodiment of the present invention.

[0012] Figure 2 Schematic diagram of exemplary hardware and software components of a supplier relationship management system combined with big data analysis provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0013] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a supplier relationship management method combined with big data analysis provided by an embodiment of the present invention. The supplier relationship management method combined with big data analysis is introduced in detail below.

[0014] Step S110: Construct a conversation text semantic association network, which is based on the historical conversation text big data between suppliers and enterprises, and includes multiple conversation text units and semantic association edges between conversation text units. The association strength of the semantic association edges is determined according to the closeness of the semantic association between the conversation text units.

[0015] This embodiment uses the relationship management of robot suppliers in intelligent manufacturing production lines as an application scenario. Enterprises have long-term cooperative relationships with robot core component suppliers, complete machine assembly suppliers, and after-sales maintenance suppliers. The historical conversation text big data covers procurement negotiation records, technical parameter confirmation documents, quality issue communication emails, delivery progress coordination minutes, etc. When constructing a conversation text semantic association network, these historical texts need to be sorted and processed first. For example, all conversation texts with a certain robot core component supplier in year X are split according to chronological order and subject content to form multiple independent conversation text units, each of which revolves around a specific business theme, such as "Servo motor purchase price negotiation", "Encoder technical parameter confirmation", "Quarterly delivery volume adjustment negotiation", etc.

[0016] After identifying the conversational text units, the thematic content of each unit is analyzed, and core vocabulary representing the theme, such as "servo motor," "purchase price," "technical parameters," and "delivery volume," is extracted to form a list of topic-representing vocabulary. For any two conversational text units, the semantic connection strength is calculated by comparing the topic-representing vocabulary lists. When the semantic connection strength exceeds a preset threshold, a semantic connection edge is established, and the connection strength value is used as the connection strength. For example, the "Servo Motor Purchase Price Negotiation" unit and the "Quarterly Delivery Volume Adjustment Negotiation" unit both involve procurement-related topics and thus have a semantic connection edge. The connection strength is determined based on the overlap and semantic relevance of the two topical vocabulary.

[0017] Step S111: Divide the historical conversation text big data between suppliers and enterprises into multiple independent conversation text units according to chronological order and topic coherence, and each conversation text unit corresponds to a complete topic interaction content.

[0018] When segmenting historical conversation text big data, first sort by chronological order and then split based on topic coherence. For example, in the conversation transcripts from March 1st to 5th of Year X, the content centered around "Confirmation of the Assembly Process for New Collaborative Robots," thus forming one conversation text unit. The content from March 10th to 15th, focused on "Coordinating the Delivery Time for the First Prototypes," also constitutes another conversation text unit. Each conversation text unit includes all interactions under that topic, including meeting minutes, emails, and instant messaging logs, ensuring complete and coherent content within the unit.

[0019] Step S112: Filter out topic representation words that can uniquely identify the business topic of each conversation text unit, and form a topic representation word list corresponding to each conversation text unit.

[0020] For each conversational text unit, semantic analysis tools are used to extract core vocabulary. For example, in the unit "New Collaborative Robot Assembly Process Confirmation," terms such as "collaborative robot," "assembly process," "component tolerance," "assembly process," and "quality inspection standards" are extracted. These terms accurately reflect the unit's business theme. These terms are sorted by importance to form a list of vocabulary representing the topic of the conversational text unit. The length of the list is determined by the complexity of the topic to ensure comprehensive and unique identification of the topic.

[0021] Step S113: Compare the topic representation vocabulary lists of any two conversation text units, count the number of overlapping words and the number of semantically related words in the two topic representation vocabulary lists, and calculate the degree of semantic association between the two conversation text units.

[0022] Select any two conversation text units, such as the unit "Confirmation of the new collaborative robot assembly process" and the unit "Revision of the collaborative robot quality inspection standard," and obtain a list of topic-representing vocabulary for each. Compare the vocabulary in the lists and count identical words, such as "collaborative robot" and "quality inspection standard," as the number of overlapping words. For non-overlapping words, such as "assembly process" and "inspection process," the semantic association between the two is determined using a semantic association dictionary. If the association exceeds a preset threshold, it is counted as part of the number of semantically associated words.

[0023] Step S1131: Select any two conversation text units in the conversation text semantic association network, and obtain the topic representation vocabulary lists corresponding to the two conversation text units respectively, which are recorded as the first topic vocabulary list and the second topic vocabulary list.

[0024] Two units are randomly selected from the semantic association network of the conversation text. For example, the first unit is "signing of the robot servo motor purchase contract", and its first theme vocabulary list includes "servo motor", "purchase contract", "payment method", "warranty period", and "liability for breach of contract"; the second unit is "servo motor quality problem claim negotiation", and its second theme vocabulary list includes "servo motor", "quality problem", "claim amount", "warranty period", and "maintenance plan".

[0025] Step S1132: Compare each word in the first theme vocabulary list with each word in the second theme vocabulary list one by one, and record the number of identical words as the number of overlapping words.

[0026] Comparing "servo motor" in the first theme vocabulary list with "servo motor" in the second theme vocabulary list, we determined that they are identical. "Warranty period" appears in both lists and is also identified as an identical term. Other terms, such as "purchase contract" and "quality issue," and "payment method" and "claim amount," are different. Therefore, the number of overlapping terms is 2.

[0027] Step S1133: For the non-overlapping words in the first subject vocabulary list, search for words with similar semantics in the second subject vocabulary list, determine the degree of semantic association between the words through the semantic association dictionary, and record the number of words with a semantic association degree higher than the preset semantic threshold as the number of semantically associated words.

[0028] The non-overlapping terms in the first theme vocabulary list are "purchase contract," "payment method," and "liability for breach of contract." Searching for semantically similar terms in the second theme vocabulary list reveals that "purchase contract" and "claim amount" both involve contractual payments, and the semantic association dictionary indicates that their semantic association exceeds the preset threshold. "Liability for breach of contract" and "repair plan" both involve problem resolution, and their semantic association also exceeds the preset threshold. "Payment method" has no semantically similar terms in the second theme vocabulary list. Therefore, the number of semantically related terms is 2.

[0029] Step S1134: Calculate the sum of the number of overlapping words and the number of semantically associated words, and record it as the total number of associated words.

[0030] The total number of associated words is equal to the sum of the number of overlapping words and the number of semantically associated words, that is, 2 plus 2 equals 4.

[0031] Step S1135: Calculate the ratio of the total number of associated words to the total number of words in the two topic representation vocabulary lists to obtain an initial association coefficient.

[0032] The first theme vocabulary list has 5 words, the second theme vocabulary list has 5 words, and the total number of words is 5 plus 5, which equals 10. The initial association coefficient is the ratio of the total number of associated words, 4, to the total number of words, 10.

[0033] Step S1136: Analyze the time interval between two conversation text units in the historical conversation text big data. If the time interval is within the preset time range, perform a positive correction on the initial correlation coefficient; if it exceeds the preset time range, perform a negative correction.

[0034] Assume that the "Robot Servo Motor Purchase Contract Signing" unit occurs in January of Year X, and the "Servo Motor Quality Issue Claim Negotiation" unit occurs in March of Year X. The time interval is 2 months, and the preset time range is within 3 months. Therefore, the initial correlation coefficient is positively corrected, and the correction amplitude is determined by the ratio of the time interval to the preset range.

[0035] Step S1137: Determine the corrected correlation coefficient as the degree of semantic correlation between the two conversation text units.

[0036] The correlation coefficient obtained after positive correction is the degree of semantic correlation between the two conversation text units, which reflects the comprehensive correlation between the two in terms of subject content and time correlation.

[0037] Step S114: When the semantic association between two conversation text units is higher than a preset association threshold, a semantic association edge is established between the two conversation text units.

[0038] The preset association threshold is set based on the association of historical conversation text. If the calculated semantic association between two conversation text units exceeds the threshold, a semantic association edge is added to the network between the two units. For example, the semantic association between the "Robot Servo Motor Purchase Contract Signing" unit and the "Servo Motor Quality Issue Claim Negotiation" unit is higher than the preset threshold, so a semantic association edge is established between them.

[0039] Step S115: directly assigning the numerical value of the semantic association closeness as the association strength of the corresponding semantic association edge, ensuring a one-to-one correspondence between the association strength and the semantic association closeness.

[0040] The association strength of the semantic association edge directly adopts the calculated value of the semantic association closeness. For example, if the semantic association closeness value of the above two units is 0.6, then the association strength of the corresponding semantic association edge is 0.6, so that the association strength can accurately reflect the degree of semantic association between the two units.

[0041] Step S116: Integrate all conversation text units and established semantic association edges to form a conversation text semantic association network including unit attributes, edge connection relationships and association strengths.

[0042] All divided conversation text units are treated as network nodes, with each node containing attributes such as unit number, topic content, and time information. The established semantic association edges are organized according to their connection relationships, with each edge containing the starting node, ending node, and association strength information. These nodes and edges are integrated using network visualization tools to form a complete conversation text semantic association network, which intuitively displays the semantic associations between conversation text units.

[0043] Step S120: Based on the conversation text semantic association network, the supplier interaction intention in the conversation text unit is mined to obtain a supplier interaction intention set, where the supplier interaction intention set includes cooperation demand intention, problem feedback intention, risk suggestion intention and collaborative optimization intention.

[0044] Based on the constructed semantic association network of conversational text, intent analysis is performed on each conversational text unit. For example, in the "Quarterly Delivery Adjustment Negotiation" unit, the supplier's cooperation demand intent is identified by analyzing the text content, such as "Hoping to increase next month's delivery volume" and "Can the production plan be adjusted?"; in the "Servo Motor Quality Issue Communication" unit, the problem feedback intent is identified based on expressions such as "Some motors are operating with excessive noise" and "There is a stalling phenomenon."

[0045] The interaction intents of all conversation text units are categorized and aggregated to form a supplier interaction intent set. For related conversation text units, the relevance of their intents is comprehensively analyzed. For example, the "quality issue feedback" unit and the "maintenance solution negotiation" unit both belong to problem feedback-related intents and are therefore managed as a single category in the set.

[0046] Step S121: extracting all conversation text units from the conversation text semantic association network, and arranging them in chronological order to form an ordered conversation text unit sequence.

[0047] Traverse the conversation text semantic association network, extract all conversation text units, obtain the timestamp information of each unit, and arrange them in chronological order. For example, arrange the conversation text units with a supplier in year X in the order of January, February, March, etc., forming an orderly sequence from the beginning of the year to the end of the year, which facilitates subsequent analysis of changing trends in interaction intent along the timeline.

[0048] Step S122: Decompose each conversation text unit in the conversation text unit sequence into multiple independent semantic words, remove function words without actual semantics, and retain valid semantic words with business meanings.

[0049] Each conversational text unit is broken down into words. For example, the sentence "We hope to increase the delivery volume of collaborative robots next month to meet the production line expansion needs" is broken down into words like "hope," "increase," "next month," "collaborative robots," "delivery volume," "meet," "production line," "expansion," and "demand." Function words like "hope" and "with" are removed, while valid semantic words with business meaning, such as "increase," "next month," "collaborative robots," "delivery volume," "production line," "expansion," and "demand," are retained.

[0050] Step S123: semantically combine the valid semantic words of each conversation text unit to form a semantic combination fragment that can reflect the core meaning of the unit.

[0051] Combine valid semantic words according to semantic logic. For example, the valid semantic word combination of the above unit is "Increase the delivery volume of collaborative robots next month to meet the production line expansion needs." This fragment can accurately reflect the core meaning of the conversation text unit, facilitating subsequent matching with typical semantic expressions.

[0052] Step S124: Establish a supplier interaction intention comparison library, which contains typical semantic expressions corresponding to cooperation demand intention, problem feedback intention, risk suggestion intention and collaborative optimization intention.

[0053] In the supplier interaction intent comparison database, typical semantic expressions of collaborative demand intent include "adjusting delivery volume," "negotiating purchase prices," and "increasing order quantity." Typical semantic expressions of problem feedback intent include "abnormal product operation," "parameters not meeting standards," and "delivery delays." Typical semantic expressions of risk reporting intent include "raw material shortages may affect supply" and "inadequate production capacity makes it difficult to meet demand." Typical semantic expressions of collaborative optimization intent include "optimizing production processes," "improving product performance," and "cost reduction solutions." These typical semantic expressions are compiled from historical interaction cases and are updated regularly.

[0054] Step S125: matching the semantic combination fragment of each conversation text unit with the typical semantic expressions in the supplier interaction intention comparison library to determine the supplier interaction intention corresponding to each conversation text unit.

[0055] The semantic combination fragments are compared one by one with the typical semantic expressions in the reference library. For example, "Increase the delivery volume of collaborative robots next month to meet the production line expansion needs" matches the typical expression "adjust the delivery volume" in the cooperation demand intention. Therefore, the supplier interaction intention corresponding to this conversation text unit is the cooperation demand intention; "The operating noise of some servo motors exceeds the standard and there are jamming phenomena" matches the typical expression "abnormal product operation" in the problem feedback intention, and the corresponding intention is the problem feedback intention.

[0056] Step S1251: Extract typical semantic expressions corresponding to cooperation demand intention, problem feedback intention, risk suggestion intention and collaborative optimization intention from the supplier interaction intention comparison library to form four types of typical expression sets.

[0057] Typical semantic expressions of four types of intentions are extracted from the reference library. The typical expression set of cooperative demand intention includes expressions such as "adjusting delivery volume", "negotiating purchase price", and "increasing order quantity"; the typical expression set of problem feedback intention includes expressions such as "abnormal product operation", "parameters not meeting standards", and "delivery delay"; the typical expression set of risk suggestion intention includes expressions such as "shortage of raw materials may affect supply" and "insufficient production capacity makes it difficult to meet demand"; the typical expression set of collaborative optimization intention includes expressions such as "optimizing production processes", "improving product performance", and "cost reduction plans".

[0058] Step S1252: perform similarity comparison between the semantic combination fragment of each conversation text unit and each typical semantic expression in the four typical expression sets, and calculate four similarity values ​​corresponding to the four types of intentions.

[0059] The semantic combination fragment "Increase the delivery volume of collaborative robots next month to meet the production line expansion needs" is compared with "adjust the delivery volume" in the typical expression set of cooperation demand intention to obtain a similarity value; compared with the typical expressions of problem feedback intention, risk suggestion intention, and collaborative optimization intention, three other similarity values ​​are obtained respectively. The four values ​​correspond to the matching degree of the four types of intentions.

[0060] Step S1253: Find the maximum value among the four similarity values, and determine that the intent type corresponding to the maximum value is the candidate intent.

[0061] Among the four similarity values, the value corresponding to the cooperation demand intention is the largest, so the candidate intention is the cooperation demand intention.

[0062] Step S1254: Check whether the similarity value corresponding to the candidate intent is higher than the preset matching threshold. If it is higher than the preset matching threshold, directly determine the candidate intent as the supplier interaction intent corresponding to the conversation text unit.

[0063] The preset matching threshold is set according to the historical matching accuracy. If the similarity value of the candidate intention is higher than the threshold, the supplier interaction intention corresponding to the conversation text unit is determined to be a cooperation demand intention.

[0064] Step S1255: If the similarity value corresponding to the candidate intent is lower than the preset matching threshold, the associated unit of the conversation text unit in the conversation text semantic association network is extracted to obtain the supplier interaction intent corresponding to the associated unit.

[0065] When the matching degree between the semantic combination fragment and the typical expression is low, for example, the similarity between a certain fragment "equipment operating parameters need further confirmation" and various typical expressions is lower than the threshold, then the associated unit of the conversation text unit in the network is searched. Assuming that the associated unit is "collaborative robot technical parameter confirmation", its corresponding interaction intention is the cooperation demand intention.

[0066] Step S1256: Count the intention type that appears most frequently in the associated unit and use the intention type as an auxiliary reference intention.

[0067] If the conversation text unit has multiple associated units, corresponding to cooperation demand intention, collaborative optimization intention, and cooperation demand intention, the intention type with the highest frequency of occurrence is cooperation demand intention, which is used as an auxiliary reference intention.

[0068] Step S1257: Combine the candidate intent and the auxiliary reference intent, and determine the semantic consistency between the two through the natural language processing model. If the consistency is higher than the preset consistency threshold, the auxiliary reference intent is determined to be the supplier interaction intention corresponding to the conversation text unit; if the consistency is lower than the preset consistency threshold, recheck the disassembly logic of the semantic combination fragment, correct it, and match it again until the corresponding supplier interaction intention is determined.

[0069] The semantic consistency between the candidate intent and the auxiliary reference intent is analyzed through the natural language processing model. If the consistency is high, for example, both involve cooperation-related content, the auxiliary reference intent is determined to be the interaction intention of the unit; if the consistency is low, the semantic combination fragments are re-disassembled, such as correcting "equipment operating parameters need further confirmation" to "collaborative robot operating parameters need further confirmation" and then matching again until the intent type is determined.

[0070] Step S126: Collect the supplier interaction intentions corresponding to all conversation text units, classify and summarize them by intention type, and form a supplier interaction intention set including cooperation demand intention, problem feedback intention, risk suggestion intention and collaborative optimization intention.

[0071] The interaction intents of all conversation text units were categorized by type, and the number and distribution of each intent were counted to form a set of supplier interaction intentions. For example, the set contained 15 intentions for collaboration needs, 8 for problem feedback, 5 for risk reporting, and 10 for collaborative optimization. This clearly demonstrates the distribution of suppliers' interaction intentions in different areas.

[0072] Step S130: Calling a natural language processing model to jointly analyze the supplier interaction intention set and the conversation text semantic association network, predicting the relationship status between the supplier and the enterprise, and obtaining a supplier relationship status prediction result, which includes a description of the relationship stability and a description of the potential relationship change tendency.

[0073] A natural language processing model, fine-tuned with specialized corpus from the supplier collaboration field, was used as input for a set of supplier interaction intentions and the structured data of the semantic association network of conversational text. The model analyzed the set of intentions, calculating the proportion of cooperation demand intentions and collaborative optimization intentions. A high proportion indicates a relatively stable relationship. The network structure data was then analyzed to observe the strength of the correlation between conversational text units corresponding to problem feedback intentions and risk suggestion diagrams. High and concentrated correlations indicate a potential for relationship change. For example, the analysis showed that cooperation demand intentions accounted for 60% and collaborative optimization intentions accounted for 20%, with the relationship stability described as "the current overall cooperation relationship is stable, with a strong willingness to collaborate on both sides." Problem feedback intentions were concentrated around product quality, with a high correlation strength and the potential for relationship change described as "attention needs to be paid to the impact of accumulated product quality issues on the cooperation relationship, as there is a risk of relationship fluctuation."

[0074] Step S131: Import professional corpus in the supplier cooperation field to fine-tune the natural language processing model so that the natural language processing model can recognize business semantic expressions in supplier interaction scenarios.

[0075] We collect professional corpus from supplier collaboration, including procurement contract texts, technical agreement documents, quality dispute resolution records, and delivery coordination emails. After removing duplicate and irrelevant content, we annotate the corpus with interaction scenario types (e.g., procurement negotiations, technical confirmations, quality issue resolution) and core intent labels (e.g., collaboration requirements, problem feedback). We divide the annotated corpus into training and validation corpora, load a pre-trained natural language processing model, set learning parameters, and then train it. We adjust the model weights through backpropagation until the model's accuracy in identifying interaction scenario types and labeling core intent labels reaches preset standards.

[0076] Step S1311: Collect professional corpus in the field of supplier cooperation, which covers text content of scenarios such as cooperation negotiation conversations, product delivery communication, quality issue handling, and contract terms negotiation.

[0077] In the context of managing relationships with suppliers for intelligent manufacturing production line robots, collaborative negotiation conversations include minutes of robot purchase price negotiation meetings and technical parameter confirmation meetings. Product delivery communications encompass emails confirming robot component delivery times, complete machine assembly progress notifications, and delivery acceptance reports. Quality issue handling includes robot operational failure feedback, repair plan communication records, and quality rectification notices. Contract terms negotiations involve draft supplementary agreements to the purchase contract, records of payment term changes, and confirmation texts of after-sales service terms. These specialized data are collected through internal enterprise document management systems, email servers, meeting record databases, and other channels to ensure coverage of key interaction scenarios throughout the supplier collaboration process.

[0078] Step S1312: Remove duplicate text segments, garbled text segments, and text segments irrelevant to supplier cooperation from the professional corpus, and retain valid professional corpus.

[0079] The collected professional corpus is pre-processed. Duplicate text fragments, such as repeated contract clauses and identical delivery notices sent multiple times, are identified and deleted using text duplication detection tools. Encoding verification tools are used to detect and remove garbled text fragments caused by formatting errors, such as emails with misplaced characters or unusual symbols. Manual screening removes text unrelated to supplier collaboration, such as daily work schedules for internal employees and communication records with other non-supplier companies. For example, when processing corpus related to quality issues, duplicate submissions of the same fault description are deleted, and garbled text in poorly formatted maintenance logs is removed. Clear and complete fault analysis reports and communication content regarding solutions are retained to form a collection of effective professional corpus.

[0080] Step S1313: Annotate each text segment in the valid professional corpus to obtain annotated valid professional corpus, where the annotation content includes the supplier interaction scenario type to which the text segment belongs and the corresponding core intent label.

[0081] For each valid professional corpus text segment, the interaction scenario type is clarified based on its content. For example, "Records of the Fault Analysis Meeting on Substandard Robot Welding Accuracy" is marked as a quality problem handling scenario; "Minutes of the Robot Annual Procurement Volume and Price Negotiation Meeting" is marked as a cooperative negotiation conversation scenario. At the same time, the core intent labels are marked according to the core meaning of the text segment. For example, "Email requesting to speed up the delivery of core robot components" is marked as a cooperative demand intent label; "Report reflecting the compatibility issues of the robot control system" is marked as a problem feedback intent label; "Letter reminding of the risk of price increases of parts and raw materials" is marked as a risk suggestion intent label; "Proposal to optimize the robot maintenance cycle" is marked as a collaborative optimization intent label. During the labeling process, for text segments with ambiguous expressions, the labeling content is jointly confirmed through cross-departmental collaborative meetings to ensure the accuracy of the labeling.

[0082] Step S1314: Divide the annotated valid professional corpus into fine-tuning training corpus and fine-tuning verification corpus according to a preset ratio.

[0083] The annotated valid professional corpus is divided according to a preset ratio, such as 70% and 30%. From the corpus corresponding to each interaction scenario type and core intent label, a corresponding proportion of text fragments are randomly selected to form the fine-tuning training corpus and fine-tuning verification corpus. For example, in the corpus of the quality problem handling scenario, 70% of the fault feedback sheets and maintenance plan communication records are selected as the fine-tuning training corpus, and the remaining 30% is used as the fine-tuning verification corpus; in the corpus of the cooperative negotiation conversation scenario, the price negotiation records, technical parameter confirmation minutes and other text fragments are also divided in proportion to ensure that the fine-tuning training corpus and the fine-tuning verification corpus are consistent in the distribution of scenario types and intent labels, so as to avoid the uneven distribution of data affecting the model fine-tuning effect.

[0084] Step S1315: Load the pre-trained natural language processing model and set the learning parameters for fine-tuning the natural language processing model, including the learning step size, training batch size, and iteration termination condition.

[0085] Load a natural language processing model pre-trained based on general corpus, which already has basic text semantic understanding capabilities. Set learning parameters based on the characteristics of the cooperative corpus of the intelligent manufacturing production line robot supplier: the learning step size is set to a value that adapts to the professional field corpus to ensure the stability of the model parameter adjustment; the training batch size is determined based on the amount of corpus data and computing resources to balance the training efficiency and model convergence effect; the iteration termination condition is set to stop training when the recognition accuracy of the fine-tuning verification corpus remains stable for multiple consecutive rounds and is not lower than the preset standard, or terminate when the maximum number of iterations is reached. For example, the maximum number of iterations is set to the preset rounds. If, during the training process, the scene recognition accuracy and intent labeling accuracy of the verification corpus exceed the preset threshold for three consecutive rounds, the iteration is terminated early.

[0086] Step S1316: Input the fine-tuning training corpus into the pre-trained natural language processing model, use the annotated interaction scenario types and core intent labels as training targets, and adjust the internal weight parameters of the natural language processing model through back propagation.

[0087] The divided fine-tuning training corpus is input into the pre-trained natural language processing model in batches. The model extracts features and performs semantic analysis on each text segment, and outputs the predicted interaction scenario type and core intent label. The prediction result is compared with the labeled target value, and the loss value is calculated. The internal weight parameters of the model are adjusted layer by layer from the output layer to the input layer through the backpropagation algorithm to enhance the recognition ability of professional terminology and business logic in supplier cooperation scenarios. For example, for the training corpus of "Communication Records on the Rectification of Excessive Noise Problems in Robot Servo Motor Operation", the model continuously optimizes the recognition weights of the "Quality Problem Handling" scenario type and the "Problem Feedback" intent label through learning, improving the accuracy of judgment on similar texts.

[0088] Step S1317: After each round of training, the fine-tuning verification corpus is input into the natural language processing model to test the natural language processing model's recognition accuracy of the interaction scenario type and the accuracy of the core intent labeling.

[0089] After each round of training, the fine-tuned verification corpus is input into the currently trained natural language processing model, and the model predicts the scene type and intent label for the text fragments in the verification corpus. The accuracy of interaction scene type recognition is calculated by statistically analyzing the ratio of the number of scene types correctly predicted by the model to the total number of scene types in the verification corpus. The accuracy of core intent labeling is calculated by statistically analyzing the ratio of the number of core intent labels correctly predicted to the total number of intent labels in the verification corpus. For example, after a round of training, 100 pieces of verification corpus related to quality problem handling are input, and the model correctly identifies 92 scene types and correctly labels 90 core intents. In this case, the scene recognition accuracy and intent labeling accuracy for that round are the corresponding ratios.

[0090] Step S1318: If the recognition accuracy and labeling accuracy of the natural language processing model both meet the preset performance standards, stop fine-tuning; if they do not meet the preset performance standards, adjust the learning parameters and continue training until the performance of the natural language processing model meets the preset standards.

[0091] Preset performance standards are set based on actual business needs. For example, the accuracy of interaction scenario type recognition must be no less than a preset value, and the accuracy of core intent labeling must be no less than a preset value. When the verification results after a round of training show that both accuracy rates meet or exceed the preset standards, fine-tuning is stopped and the current model parameters are saved as the fine-tuned natural language processing model. If the standards are not met, the loss value trend is analyzed and learning parameters such as the learning step size or training batch size are adjusted. For example, the learning step size is reduced when the loss value decreases slowly, and the batch size is adjusted when the model fluctuates significantly. Then, the next round of training is continued. This process is repeated until the model performance meets the preset standards, ensuring that the model can accurately identify the business semantic expressions in the interaction scenarios of robots and suppliers in intelligent manufacturing production lines.

[0092] Step S132: Integrate each supplier interaction intention in the supplier interaction intention set with the corresponding conversation text unit association information to form intention association data that can be read by a natural language processing model.

[0093] Extract the identification information, timestamp, and topic representation vocabulary list of the conversation text unit corresponding to each intent in the supplier interaction intention set and integrate it with the intent type. For example, associate the "collaboration demand intention - increase in robot procurement volume" with the corresponding conversation text unit ID, interaction time, and topic vocabulary such as "procurement volume, production capacity, and delivery cycle." Structured processing is performed according to the format required by the natural language processing model, converting it into intent-related data containing text sequences, intent labels, and contextual features. The text sequence is the core content of the conversation text unit, the intent label is the corresponding interaction intent type, and the contextual features include time information and topic vocabulary features, ensuring that the model can read and understand the relationship between intent and text units.

[0094] Step S133: extracting the number of connections of conversation text units, the association strength distribution of semantic association edges, and the unit topic association density from the conversation text semantic association network to form network structure data that can be read by a natural language processing model.

[0095] The number of connections within a conversational text unit refers to the total number of semantically related edges established between each unit and other units in the semantic association network. For example, if a conversational text unit about robot technology upgrades is associated with five other units, the number of connections is five. The association strength distribution of semantic association edges refers to the distribution of association strength values ​​for all associated edges within that unit, such as the proportion of strength values ​​in different intervals. The unit topic association density is calculated by calculating the overlap between the topic representation vocabulary of the unit and the topic representation vocabulary of the associated units. This data is converted into vector form, with the number of connections serving as a one-dimensional feature, the association strength distribution divided into multiple dimensional features by interval, and the topic association density serving as a one-dimensional feature. Together, these form network structure data, adapting to the input format of natural language processing models.

[0096] Step S134: The intention association data and the network structure data are simultaneously input into the fine-tuned natural language processing model, and the intention association data is deeply semantically analyzed through the natural language processing model to identify the association logic between the interaction intentions of different suppliers.

[0097] After receiving intent-related data, the fine-tuned natural language processing model converts the text sequence into a semantic vector through an embedding layer. This vector then passes through multiple Transformer encoder layers for feature extraction, capturing the contextual semantic relationships within the text. For example, by analyzing the intent-related data for "collaboration demand intention - increase procurement volume" and "collaborative optimization intention - increase production capacity," the model identifies a causal relationship between the two: the demand for increased procurement volume prompts suppliers to propose collaborative optimization suggestions to increase production capacity. Similarly, by analyzing the associated data for "problem feedback intention - high robot failure rate" and "risk suggestion intention - unstable component quality," the model identifies the progressive relationship between problem feedback and risk warnings.

[0098] Step S135: topologically analyzing the network structure data using the natural language processing model to determine the semantic association stability and association change trend between conversation text units.

[0099] The model analyzes features such as the number of connections, association strength distribution, and topic association density in the network structure data. If the number of connections in a conversation text unit is stable, the association strength distribution is concentrated, and the topic association density is high, then its semantic association stability is judged to be strong. If the number of connections fluctuates greatly, the association strength distribution is dispersed, and the topic association density decreases, then the association change trend is identified as weakening. For example, when analyzing the network structure data of a conversation text unit about robot after-sales service, if the number of connections with other units remains at a high level, the association strength is concentrated in the high value range, and the topic association density is stable, then the semantic association stability of the unit is judged to be strong. If the number of connections gradually decreases and the association strength shifts to the low value range, then the association change trend is identified as weakening.

[0100] Step S136: Combine the semantic deep analysis results and the topological analysis results, calculate the proportion of cooperation demand intention and collaborative optimization intention in all intentions, and generate a description of the relationship stability.

[0101] The sum of the number of collaborative demand intentions and collaborative optimization intentions in the supplier interaction intention set is counted, and the percentage of these intentions relative to the total number of intentions is calculated. If the percentage is high, and deep semantic analysis shows positive correlations between these two intentions and other intentions, and topological analysis shows strong semantic correlation stability within the corresponding conversational text units, the relationship stability can be described as "the supplier relationship is generally stable, the willingness to cooperate is positive, and the atmosphere for collaborative optimization is positive." If the percentage is moderate, but some correlations fluctuate and the correlation stability is average, the relationship can be described as "the supplier relationship is generally stable, with some room for collaborative optimization." If the percentage is low, the correlations are loose, and the correlation stability is weak, the relationship can be described as "the supplier relationship is insufficiently stable, and the willingness to cooperate needs to be improved."

[0102] Step S137: Analyze the concentrated occurrence scenarios and associated unit distribution of problem feedback intentions and risk suggestion diagrams, and generate a description of potential relationship change tendencies.

[0103] Identify interaction scenarios where problem feedback intentions and risk suggestion diagrams are concentrated, such as quality issue handling scenarios and delivery delay communication scenarios. Analyze the distribution of associated units within the semantic association network for the conversational text units corresponding to these intentions. If they are concentrated in a specific type of topic unit (such as units related to component quality), identify potential relationship change trends. For example, if problem feedback intentions and risk suggestion diagrams are concentrated in quality issue handling scenarios, and the associated units are mostly units related to the quality of core robot components, the potential relationship change trend can be described as "due to core component quality issues, there is a potential tendency for supplier relationships to become tense." If they are concentrated in delivery communication scenarios, and the associated units are mostly units related to logistics delays, the potential relationship change trend can be described as "due to delivery progress, there is a potential tendency for supplier cooperation efficiency to decline."

[0104] Step S138: Integrate the relationship stability description and the potential relationship change tendency description to obtain a supplier relationship status prediction result.

[0105] The descriptions of relationship stability and potential changes are logically integrated to form a comprehensive supplier relationship status forecast. For example, the integrated result might read, "The supplier relationship is generally stable, with a positive willingness to cooperate and a positive atmosphere for collaborative optimization; however, due to quality issues with core components, there is a potential for tension, requiring close attention to progress in quality rectification." This supplier relationship status forecast reflects both the current stability of the supplier relationship and its potential trends.

[0106] Step S140: locating optimization and improvement nodes in the supplier relationship based on the supplier relationship status prediction result, where the optimization and improvement nodes are conversation text-related nodes that affect the stability of the relationship or lead to potential relationship change tendencies.

[0107] Based on the relationship stability and potential change profiles in the supplier relationship status prediction results, the corresponding conversation text units and associated units are found in the conversation text semantic association network. These units are key nodes that may influence the supplier relationship status. For example, based on the stability profile of "positive willingness to cooperate," the corresponding cooperative negotiation conversation unit is found; based on the change profile of "potential tension caused by core component quality issues," the corresponding quality problem feedback unit and associated component inspection unit are found. These units are then used as candidate optimization and improvement nodes.

[0108] Step S141: extracting key semantic expressions related to relationship stability from the relationship stability description of the supplier relationship status prediction result, wherein the key semantic expressions are conversation text segments that directly reflect the cooperation stability.

[0109] We screened out text snippets from the relationship stability descriptions that clearly reflect the state of cooperation, such as key semantic expressions like "high proportion of cooperation demand intention," "consensus reached on collaborative optimization measures," and "delivery timeliness meets agreement." These expressions, derived from the core content of conversational text units, directly reflect the stability of supplier relationships, such as "both parties reached an agreement on the annual robot procurement plan" and "supplier completed customized parts delivery on time."

[0110] Step S142: extracting a triggering semantic expression that triggers a relationship change from the potential relationship change tendency description of the supplier relationship status prediction result, wherein the triggering semantic expression is a conversation text segment that may cause a relationship fluctuation.

[0111] From the description of potential relationship change trends, we extract text snippets related to relationship change, such as triggering semantic expressions like "core component failure rate exceeds standard," "delivery delays increase," and "unresolved payment terms disputes." These expressions correspond to question feedback within conversational text units, such as "robot servo motors experience frequent operational failures" and "parts delivery was delayed three times in the third quarter." These conversational text snippets are key factors that may trigger fluctuations in supplier relationships.

[0112] Step S143: searching for a conversation text unit containing the key semantic expression in the conversation text semantic association network, and marking the unit as a stable influence unit.

[0113] Using text search tools, we match conversational text units containing key semantic expressions within the semantic association network of conversational text. For example, we search for collaborative negotiation conversation units containing "procurement plan agreement" and technical communication units containing "collaborative optimization measures implemented." These units are then labeled as stabilizing influence units. These units have positive semantic associations with other units in the semantic association network, supporting the stability of supplier relationships.

[0114] Step S144: searching for a conversation text unit containing the trigger semantic expression in the conversation text semantic association network, and marking the unit as a change trigger unit.

[0115] Similarly, text retrieval is used to locate conversational text units containing triggering semantic expressions within the conversational text semantic association network. For example, quality feedback units containing "frequent failures" and logistics communication units containing "delivery delays" are labeled as change triggering units. These units may be connected to multiple problem-related units in the network and are the source nodes that trigger relationship changes.

[0116] Step S145: Find all conversation text units that have semantically associated edges with the stable influence unit and mark them as stable associated units.

[0117] Traverse all semantically related edges of the stable influence unit in the conversation text semantic association network and find other conversation text units connected to it, such as the "Capacity Planning Confirmation" unit and the "Pricing Terms Refinement" unit associated with the "Procurement Plan Agreed" unit. Mark these units as stable association units. Stable association units and stable influence units together form an association network that supports stable relationships.

[0118] Step S146: Find all conversation text units that have semantically associated edges with the change triggering unit and mark them as change associated units.

[0119] Traverse the semantic association edges of the change trigger unit to find the associated conversation text units, such as the "Parts Inspection Report," "Maintenance Plan Negotiation," and "Quality Improvement Request" units associated with the "Frequent Failures" unit. These units are marked as change-associated units. The change-associated units and the change trigger unit form an association chain that may cause relationship fluctuations.

[0120] Step S147: Counting the sum of the association strengths of the stable influence units, stable association units, change trigger units, and change association units in the conversation text semantic association network, and filtering out units whose sum of association strengths is higher than a preset strength threshold.

[0121] Obtain the association strength values ​​for all semantically associated edges for each labeled unit and sum all the association strength values ​​for the same unit to obtain the total association strength. A preset strength threshold is determined based on the average sum of the association strengths of all units in the network, for example, 1.5 times the average sum of the association strengths. Compare the sum of the association strengths of each unit with the threshold to select units with higher sums of association strengths, indicating that these units have greater influence in the network.

[0122] Step S1471: Obtain all semantic association edges corresponding to each stable influence unit, stable association unit, change trigger unit, and change association unit in the conversation text semantic association network.

[0123] Using a network traversal tool, we extract a list of semantically associated edges for each marked unit, identifying the target units connected by each edge and the corresponding association strength values. For example, the semantically associated edges for the stability-influencing unit "Agreeing on Procurement Plan" include edges connecting to "Confirming Capacity Planning" and "Detailing Price Terms," ​​and each edge has a corresponding association strength value.

[0124] Step S1472: extract the association strength value of each semantic association edge, sum the association strength values ​​of all semantic association edges of each unit, and obtain the total association strength of each unit.

[0125] For each marking unit, the association strength values ​​of all its semantic association edges are accumulated. For example, the association strength values ​​of the three association edges of the stability influence unit are corresponding values, and the sum is the result of adding the three together; the association strength values ​​of the two association edges of the change trigger unit are corresponding values, and the sum is the result of adding the two together, so as to obtain the total association strength of each unit.

[0126] Step S1473: setting a preset strength threshold of the sum of association strengths, wherein the preset strength threshold is determined according to the sum of the average association strengths of all units in the conversation text semantic association network.

[0127] Calculate the average sum of the association strengths of all units in the conversation text semantic association network and set the preset strength threshold to a certain multiple of this average, such as 1.2 or 1.5 times. The specific multiple is adjusted based on the network size and unit association density. For example, if the average sum of the association strengths of all units in the network is a corresponding value, then the preset strength threshold is set to 1.5 times that value.

[0128] Step S1474: Compare the sum of the association strengths of each unit with a preset strength threshold, and filter out units whose sum of association strengths is higher than the preset strength threshold to form a target unit set.

[0129] For each stable influence unit, stable association unit, change trigger unit and change association unit, the sum of their association strength values ​​is extracted one by one and compared with the preset strength threshold. If the sum of the association strength of a unit exceeds the preset strength threshold, the unit is included in the target unit set. For example, in the conversation text semantic association network of a robot core component supplier, the sum of the association strength of a stable influence unit is the corresponding value, and the preset strength threshold is 1.5 times the value. Because the sum of the association strength of the unit is higher than the threshold, it is included in the target unit set; the sum of the association strength of another change trigger unit does not reach the threshold, so it is temporarily not included.

[0130] Step S1475: Analyze the subject representation vocabulary of each unit in the target unit set to determine whether the subject of the unit is directly related to supplier relationship maintenance.

[0131] Extract the topic representation vocabulary list of each unit in the target unit set, and judge the topic relevance through the topic matching rules. The topic matching rules include core topic vocabulary related to supplier relationship maintenance, such as "quality assurance", "delivery cycle", "price negotiation", "technical support", "contract performance", etc. If the topic representation vocabulary list of a unit contains at least one core topic vocabulary, then the unit topic is judged to be directly related to supplier relationship maintenance; if it does not contain any core topic vocabulary, then it is judged to be irrelevant to the topic. For example, the topic representation vocabulary of a unit is "robot reducer life test" and "quality inspection standard". Because it contains the core topic vocabulary "quality inspection standard", it is judged to be topic related; the topic representation vocabulary of another unit is "workshop equipment layout planning", which does not contain core topic vocabulary and is judged to be irrelevant to the topic.

[0132] Step S1476: retain the units whose subject is directly related to supplier relationship maintenance, and eliminate the units that are irrelevant to the subject, to obtain a preliminary screening unit set.

[0133] Based on the results of the topic relevance judgment, the target unit set is screened. Units judged to be topic-related are retained to form a preliminary screening unit set; units judged to be topic-irrelevant are removed from the target unit set. For example, in the target unit set, after topic analysis, units containing topic terms such as "quality assurance agreement," "delivery delay processing," and "technical parameter adjustment" are retained, while units involving irrelevant topics such as "workshop environment temperature control" and "logistics transportation route planning" are removed to form a preliminary screening unit set.

[0134] Step S1477: Check the supplier interaction intention corresponding to each unit in the preliminary screening unit set. If the intention type is a problem feedback intention or a risk suggestion intention, further verify whether the sum of the association strength of the unit is continuously higher than the preset strength threshold.

[0135] Traverse each unit in the preliminary screening unit set and query its corresponding intent type in the supplier interaction intention set. For units whose intent type is problem feedback intention or risk suggestion intention, retrieve the total record of its association strength in different time periods to check whether it is higher than the preset strength threshold for multiple consecutive cycles. For example, the corresponding intent type of a unit is problem feedback intention, and the feedback content is "the failure rate of the robot servo motor exceeds the standard". Retrieve the total record of its association strength for the past three months. If the total association strength of each month is higher than the preset threshold, the verification is passed; if the total association strength of a certain month is lower than the threshold, the stability of the unit needs to be re-evaluated.

[0136] Step S1478: Final confirmation is performed on the units that have passed the verification, and they are retained in the preliminary screening unit set to form the final optimized and improved node set.

[0137] After continuous verification of the sum of the association strengths, the units that passed the verification were identified as the final optimization and improvement nodes and retained in the preliminary screening unit set. For the units that failed the verification, their topic relevance and association strength data were checked again. If they still did not meet the requirements, they were eliminated. The final optimization and improvement node set formed includes all units that are topic-related, have a satisfactory sum of association strengths, and are continuously stable. Each node is labeled with the corresponding impact type (stable impact type or change trigger type) and the associated intention type. For example, the "Quality Inspection Standard Negotiation" unit was finally determined to be a stable impact type optimization and improvement node, and the "Servo Motor Failure Rate Feedback" unit was determined to be a change trigger type optimization and improvement node, together forming the optimization and improvement node set.

[0138] Step S150: Based on the optimization and improvement nodes and the supplier interaction intention set, a supplier relationship dynamic management and control mechanism is generated, and the supplier relationship dynamic management and control mechanism is applied to the supplier cooperation process to achieve continuous maintenance and adjustment of supplier relationships.

[0139] Based on the set of optimization and improvement nodes and the set of supplier interaction intentions, corresponding control measures are formulated for different types of optimization and improvement nodes. The implementing entities, execution cycles, and effect evaluation criteria are clearly defined. These control measures are then embedded into the supplier collaboration process and dynamically optimized, forming a complete dynamic supplier relationship management and control mechanism. For example, relationship strengthening measures are formulated for nodes with stabilizing influences, and risk prevention and control measures are formulated for nodes with changing triggers. By periodically implementing these measures during the collaboration process, stable and risk factors in supplier relationships can be promptly addressed, achieving continuous maintenance and adjustment of the relationship.

[0140] For example, step S151: for each optimization and improvement node, query the supplier interaction intention corresponding to the optimization and improvement node in the supplier interaction intention set to determine the impact property of the optimization and improvement node.

[0141] Analyze each node in the optimization and improvement node set one by one, and find the supplier interaction intention type corresponding to each node through the relationship between the node and the supplier interaction intention set. If the intention type corresponding to the node is mainly a cooperative demand intention or a collaborative optimization intention, then the impact nature of the node is determined to be a stabilizing and promoting nature. If the intention type corresponding to the node is mainly a problem feedback intention or a risk suggestion intention, then the impact nature of the node is determined to be a risk-inducing nature. For example, the "Technical Parameter Collaborative Optimization" node corresponds to the collaborative optimization intention, and the impact nature is a stabilizing and promoting nature; the "Raw Material Quality Fluctuation Feedback" node corresponds to the problem feedback intention, and the impact nature is a risk-inducing nature.

[0142] Step S152: If the impact property of the optimization and improvement node is a stabilizing and promoting property, then relationship strengthening management and control measures are constructed, and the relationship strengthening management and control measures include increasing the conversation communication frequency of the topic corresponding to the optimization and improvement node, determining the response process for the cooperation needs under the topic, and establishing a tracking mechanism for collaborative optimization.

[0143] For optimization and improvement nodes with a stable and promoting nature, specific relationship strengthening measures should be designed around their corresponding themes. Increasing the frequency of conversational communication can be set as adding a special video conference every month or adding an on-site technical exchange every quarter; the collaborative demand response process must clearly define the responsible departments and time limits for each link of demand reception, evaluation, feedback and execution; the collaborative optimization tracking mechanism must establish a tracking ledger to record the implementation progress of the optimization measures, interim results and content that needs to be adjusted. For example, for the "collaborative optimization of robot assembly process" node, a special process optimization communication will be added every month, clarifying that the demand response must complete the preliminary evaluation within two working days, establish an optimization measure execution tracking table, and update the implementation status every week.

[0144] Step S153: If the impact of the optimization and improvement node is risk-inducing, risk prevention and control measures are constructed. The risk prevention and control measures include establishing a rapid feedback channel for the problem corresponding to the optimization and improvement node, building a response plan for risk prompts, and setting a progress monitoring node for problem solving.

[0145] Develop risk prevention and control measures for nodes that are optimized and improved due to the nature of risk induction to reduce the impact of potential risks. Rapid feedback channels can be equipped with dedicated contact personnel, emergency contact mailboxes, and real-time communication groups to ensure real-time transmission of problem information. Response plans must clearly define risk level classification standards, the handling process for different levels of risk, and responsible personnel. Progress monitoring nodes must set key checkpoints in the problem-solving process, such as problem confirmation, solution formulation, implementation of measures, and verification of results, with a time limit set for completion for each node. For example, for the "Risk of Delayed Delivery of Robot Core Components" node, a full-time delivery coordinator will be appointed, a delivery risk emergency communication group will be established, and a response plan including delay warnings, activation of alternative solutions, and loss assessment will be developed. Three progress monitoring nodes will be set up when the delivery cycle is halfway through, before the expected delay occurs, and after the problem is resolved.

[0146] Step S154: Designate an execution entity for each relationship strengthening management and control measure and risk prevention and control measure. The execution entity is the specific department or position within the enterprise responsible for maintaining supplier relationships.

[0147] Determine the implementing entity for each control measure based on its business attributes and division of responsibilities. Relationship strengthening control measures involving technical exchanges can be implemented by the technology research and development department, while measures involving response to cooperative needs can be implemented by the procurement department. Risk prevention and control measures involving problem feedback can be implemented by the quality management department, while measures involving plan implementation can be implemented by the supply chain management department. For example, the measure of "increasing the frequency of communication for collaborative process optimization" will be implemented by the process engineer position in the technology research and development department, while the measure of "implementing the plan for responding to delivery delay risks" will be implemented by the supplier management specialist position in the supply chain management department.

[0148] Step S155: setting an execution cycle for each control measure, wherein the execution cycle is determined according to the interaction intention type corresponding to the measure and the association strength of the optimization and improvement node.

[0149] For control measures that promote stability, if the corresponding intention is long-term collaborative optimization and the node correlation strength is high, the execution cycle can be set to monthly or quarterly; if the correlation strength is low, the execution cycle can be set to semi-annual. For control measures that trigger risks, if the corresponding intention is emergency problem feedback and the node correlation strength is high, the execution cycle can be set to weekly or monthly; if it is a general risk suggestion, the execution cycle can be set to quarterly. For example, the "technical parameter collaborative optimization tracking" measure has a high correlation strength and is a long-term collaborative intention, so the execution cycle is set to once a month; the "raw material quality risk investigation" measure has a medium correlation strength, so the execution cycle is set to once a quarter.

[0150] Step S156: Divide all control measures, execution entities and execution cycles according to the links of the supplier cooperation process to form a control execution list for each link.

[0151] The supplier cooperation process includes preliminary inspections, contract signing, production delivery, quality inspection, after-sales service, and contract renewal evaluation. Each control measure is organized into corresponding process steps based on the business scenario it corresponds to. For example, the "Cooperation Demand Response Process Execution" measure is grouped into the contract signing phase; the "Delivery Progress Monitoring" measure is grouped into the production delivery phase; and the "Quick Feedback on Quality Issues" measure is grouped into the quality inspection phase. The control execution checklist for each step includes all control measures within that step, the corresponding execution entities, and the execution cycle, organized in sequential order to form a structured checklist.

[0152] Step S157: Constructing an effectiveness evaluation standard for the control measures, wherein the effectiveness evaluation standard includes the change in the proportion of positive intentions and the reduction ratio of risk intentions in the corresponding conversation text unit after the measures are implemented.

[0153] The effectiveness evaluation criteria must quantify the effectiveness of the implementation of control measures. The change in the proportion of positive intentions refers to the difference between the proportion of cooperation demand intentions and collaborative optimization intentions in the corresponding conversation text unit to the total number of intentions after the implementation of the measure and the difference before implementation; the reduction in the proportion of risk intentions refers to the reduction in the number of problem feedback intentions and risk suggestion intentions after the implementation of the measure compared to the number before implementation. For example, the effectiveness evaluation criteria for a relationship strengthening measure is set as an increase in the proportion of positive intentions by more than a certain percentage, and the criteria for a risk prevention and control measure is set as a reduction in risk intentions by more than a certain percentage.

[0154] Step S158: Embed the management and control execution list for each stage into the corresponding stage of the supplier cooperation process, and trigger the execution of management and control measures according to the execution cycle during the process execution.

[0155] Through the process management system, the control execution checklist is linked to the supplier collaboration process. When the process advances to a certain stage, the system automatically prompts the corresponding control measures and execution requirements. After receiving the prompt, the implementing entity will carry out the control measures according to the execution cycle and record the execution status in the system. For example, when the collaboration process enters the production delivery stage, the system automatically triggers the execution prompt of the "delivery progress monitoring" measure. The supplier management specialist will perform monitoring on a monthly basis and record the monitoring results.

[0156] Step S159: Regularly collect conversation text data after the execution of control measures, analyze the changes in the intent type in the conversation text data, judge the control effect against the effect evaluation standard, and adjust the execution subject, execution cycle or specific content of the control measures according to the control effect, update the control execution list for each link, and realize the iterative optimization of the dynamic control mechanism of supplier relationships.

[0157] Collect conversation text data after the implementation of control measures on a fixed cycle (e.g., quarterly), identify and count the intent types of the data, and calculate the change in the proportion of positive intent and the reduction in risk intent. Compare the statistical results with the effectiveness evaluation standards. If the standards are met, the control effect is determined to be good and the existing measures are maintained. If the standards are not met, analyze the reasons, adjust the implementation entity (e.g., replace it with a more professional department), shorten or extend the implementation cycle (e.g., change quarterly implementation to monthly implementation), or modify the content of the measures (e.g., increase the frequency of communication). The updated control implementation checklist is re-embedded in the cooperation process to complete the iterative optimization of the control mechanism. For example, after the implementation of a certain risk prevention and control measure, the risk intent reduction ratio did not meet the standard. Analysis found that the implementation cycle was too long, so the implementation cycle was adjusted from quarterly to monthly, and the control implementation checklist was updated.

[0158] Figure 2A schematic diagram illustrates exemplary hardware and software components of a supplier relationship management system 100 combined with big data analytics, which can implement the concepts of the present application, according to some embodiments of the present application. For example, a processor 120 can be used in the supplier relationship management system 100 combined with big data analytics to perform the functions described in the present application.

[0159] For example, the supplier relationship management system 100 combined with big data analysis may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and storage media 140 in different forms, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the supplier relationship management system 100 combined with big data analysis may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application can be implemented according to the program instructions. The supplier relationship management system 100 combined with big data analysis also includes an I / O interface 150 between the computer and other input and output devices.

[0160] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When a processor executes the computer-executable instructions, the supplier relationship management method combined with big data analysis as described above is implemented.

[0161] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A supplier relationship management method combined with big data analysis, characterized in that: The method comprises: Constructing a conversation text semantic association network based on historical conversation text big data between suppliers and enterprises, comprising multiple conversation text units and semantic association edges between conversation text units, wherein the association strength of the semantic association edges is determined according to the closeness of the semantic association between the conversation text units; Based on the conversation text semantic association network, mining the supplier interaction intention in the conversation text unit to obtain the supplier interaction intention set; Invoking a natural language processing model to jointly analyze the supplier interaction intention set and the conversation text semantic association network to predict the relationship status between the supplier and the enterprise, thereby obtaining a supplier relationship status prediction result; Locating optimization and improvement nodes in the supplier relationship based on the supplier relationship status prediction results, wherein the optimization and improvement nodes are conversation text-related nodes that affect the stability of the relationship or lead to potential relationship changes; Based on the optimization and improvement nodes and the supplier interaction intention set, a supplier relationship dynamic management and control mechanism is generated, and the supplier relationship dynamic management and control mechanism is applied to the supplier cooperation process.

2. The supplier relationship management method combined with big data analysis according to claim 1, characterized in that: The step of constructing a conversation text semantic association network includes: Divide the historical conversation text big data between suppliers and enterprises into multiple independent conversation text units based on chronological order and topic coherence. Each conversation text unit corresponds to a complete topic interaction content. Filtering out topic representation words that can uniquely identify the business topic of each conversation text unit to form a topic representation word list corresponding to each conversation text unit; Compare the topic representation vocabulary lists of any two conversation text units, count the number of overlapping words and the number of semantically related words in the two topic representation vocabulary lists, and calculate the degree of semantic connection between the two conversation text units; When the semantic association between two conversation text units is higher than a preset association threshold, a semantic association edge is established between the two conversation text units; The value of the semantic association closeness is directly assigned to the association strength of the corresponding semantic association edge, ensuring a one-to-one correspondence between the association strength and the semantic association closeness; All conversation text units and established semantic association edges are integrated to form a conversation text semantic association network that includes unit attributes, edge connection relationships and association strength.

3. The supplier relationship management method combined with big data analysis according to claim 2, characterized in that: The method of comparing the topic representation vocabulary lists of any two conversation text units, counting the number of overlapping words and the number of semantically related words in the two topic representation vocabulary lists, and calculating the degree of semantic association between the two conversation text units includes: Select any two conversation text units in the conversation text semantic association network, and obtain the topic representation vocabulary lists corresponding to the two conversation text units respectively, which are recorded as the first topic vocabulary list and the second topic vocabulary list; Compare each word in the first theme vocabulary list with each word in the second theme vocabulary list one by one, and record the number of identical words as the number of overlapping words; For the non-overlapping words in the first theme vocabulary list, search for words with similar semantics in the second theme vocabulary list, determine the degree of semantic association between the words using the semantic association dictionary, and record the number of words with a semantic association degree higher than a preset semantic threshold as the number of semantically associated words; Calculate the sum of the number of overlapping words and the number of semantically related words, and record it as the total number of related words; Calculate the ratio of the total number of associated words to the total number of words in the two topic representation word lists to obtain the initial association coefficient; Analyze the time interval between two conversation text units in the historical conversation text big data. If the time interval is within the preset time range, the initial correlation coefficient is corrected positively; if it exceeds the preset time range, a negative correction is made. The corrected correlation coefficient is determined as the closeness of the semantic correlation between two conversation text units.

4. The supplier relationship management method combined with big data analysis according to claim 1, characterized in that: The step of mining the supplier interaction intention in the conversation text unit based on the conversation text semantic association network to obtain a supplier interaction intention set includes: Extracting all conversation text units from the conversation text semantic association network and arranging them in chronological order to form an ordered conversation text unit sequence; Decompose each conversation text unit in the conversation text unit sequence into multiple independent semantic words, remove function words without actual semantic meaning, and retain valid semantic words with business meaning; Semantically combine the effective semantic words of each conversation text unit to form a semantic combination fragment that can reflect the core meaning of the unit; Establish a supplier interaction intention comparison library, which contains typical semantic expressions corresponding to cooperation demand intention, problem feedback intention, risk suggestion intention, and collaborative optimization intention; Matching the semantic combination fragments of each conversation text unit with typical semantic expressions in the supplier interaction intention comparison library to determine the supplier interaction intention corresponding to each conversation text unit; The supplier interaction intentions corresponding to all conversation text units are collected and classified and summarized by intent type to form a supplier interaction intention set including cooperation demand intention, problem feedback intention, risk suggestion intention and collaborative optimization intention.

5. The supplier relationship management method combined with big data analysis according to claim 4 is characterized in that: The matching of the semantic combination fragment of each conversation text unit with the typical semantic expressions in the supplier interaction intention comparison library to determine the supplier interaction intention corresponding to each conversation text unit includes: Extract typical semantic expressions corresponding to cooperation demand intention, problem feedback intention, risk suggestion intention, and collaborative optimization intention from the supplier interaction intention comparison database to form four types of typical expression sets; The semantic combination fragment of each conversation text unit is compared with each typical semantic expression in the four typical expression sets to obtain four similarity values ​​corresponding to the four types of intentions. Find the maximum value among the four similarity values ​​and determine the intent type corresponding to the maximum value as the candidate intent; Check whether the similarity value corresponding to the candidate intent is higher than a preset matching threshold. If higher than the preset matching threshold, directly determine the candidate intent as the supplier interaction intent corresponding to the conversation text unit; If the similarity value corresponding to the candidate intent is lower than the preset matching threshold, extract the associated unit of the conversation text unit in the conversation text semantic association network and obtain the supplier interaction intent corresponding to the associated unit; Count the intention types that appear most frequently in the associated units and use them as auxiliary reference intentions; Combining the candidate intent and the auxiliary reference intent, the natural language processing model is used to determine the semantic consistency between the two. If the consistency is higher than the preset consistency threshold, the auxiliary reference intent is determined to be the supplier interaction intention corresponding to the conversation text unit; if the consistency is lower than the preset consistency threshold, the disassembly logic of the semantic combination fragment is rechecked, corrected, and matched again until the corresponding supplier interaction intention is determined.

6. The supplier relationship management method combined with big data analysis according to claim 1, characterized in that: The calling of the natural language processing model to jointly analyze the supplier interaction intention set and the conversation text semantic association network to predict the relationship status between the supplier and the enterprise, and obtain the supplier relationship status prediction result, including: Import professional corpus from the supplier cooperation field to fine-tune the natural language processing model so that it can recognize business semantic expressions in supplier interaction scenarios; Integrate each supplier interaction intention in the supplier interaction intention set with the corresponding conversation text unit association information to form intention association data that can be read by a natural language processing model; Extracting the number of connections of conversation text units, the distribution of association strength of semantic association edges, and the unit topic association density from the conversation text semantic association network to form network structure data that can be read by a natural language processing model; Inputting the intention association data and the network structure data into a fine-tuned natural language processing model simultaneously, performing deep semantic analysis on the intention association data through the natural language processing model to identify the association logic between the interaction intentions of different suppliers; Performing topological analysis on network structure data using the natural language processing model to determine the stability and trend of semantic associations between conversation text units; Combining the results of deep semantic analysis with topological analysis, we calculate the proportion of cooperation demand intentions and collaborative optimization intentions in all intentions and generate a description of the relationship stability. Analyze the concentrated occurrence scenarios and associated unit distribution of problem feedback intentions and risk suggestion diagrams to generate a description of potential relationship change trends; The description of the relationship stability and the description of the potential relationship change tendency are integrated to obtain the supplier relationship status prediction result.

7. The supplier relationship management method combined with big data analysis according to claim 6, characterized in that: The imported professional corpus in the supplier cooperation field is used to fine-tune the natural language processing model so that the natural language processing model can recognize business semantic expressions in supplier interaction scenarios, including: Collect professional corpus in the field of supplier cooperation, including text content from cooperation negotiation conversations, product delivery communication, quality issue handling, and contract terms negotiation scenarios; Remove duplicate text fragments, garbled text fragments, and text fragments irrelevant to supplier cooperation from the professional corpus to retain valid professional corpus; Annotating each text segment in the valid professional corpus to obtain annotated valid professional corpus, wherein the annotation content includes the supplier interaction scenario type to which the text segment belongs and the corresponding core intent label; Divide the annotated valid professional corpus into fine-tuning training corpus and fine-tuning verification corpus according to the preset ratio; Load the pre-trained natural language processing model and set the learning parameters for fine-tuning the natural language processing model, including the learning step size, training batch size, and iteration termination conditions; The fine-tuning training corpus is input into the pre-trained natural language processing model, with the annotated interaction scenario types and core intent labels as training targets, and the internal weight parameters of the natural language processing model are adjusted through backpropagation; After each round of training, the fine-tuning verification corpus is input into the natural language processing model to test the natural language processing model's recognition accuracy of interaction scenario types and the accuracy of labeling core intent labels; If the recognition accuracy and labeling accuracy of the natural language processing model both meet the preset performance standards, fine-tuning is stopped; if the preset performance standards are not met, the learning parameters are adjusted and training is continued until the performance of the natural language processing model meets the preset standards.

8. The supplier relationship management method combined with big data analysis according to claim 1, characterized in that: The step of locating optimization and improvement nodes in the supplier relationship according to the supplier relationship status prediction result includes: Extracting key semantic expressions related to relationship stability from the relationship stability description of the supplier relationship status prediction result, wherein the key semantic expressions are conversation text segments that directly reflect the cooperation stability; Extracting a trigger semantic expression that triggers relationship change from the potential relationship change tendency description in the supplier relationship status prediction result, wherein the trigger semantic expression is a conversation text segment that may cause relationship fluctuation; Searching for conversation text units containing the key semantic expressions in the conversation text semantic association network and marking them as stable influence units; Searching for a conversation text unit containing the trigger semantic expression in the conversation text semantic association network, and marking the unit as a change trigger unit; Find all conversation text units that have semantic association edges with the stable influence unit and mark them as stable association units; Find all conversation text units that have semantic association edges with the change trigger unit and mark them as change association units; Counting the sum of the association strengths of the stable influence units, stable association units, change trigger units, and change association units in the conversation text semantic association network, and screening out units whose sum of association strengths is higher than a preset strength threshold; The screened units are determined as optimization and improvement nodes in the supplier relationship, and each optimization and improvement node corresponds to a determined impact type, which is a stable impact type or a change trigger type.

9. The supplier relationship management method combined with big data analysis according to claim 8, characterized in that: The counting of the sum of the association strengths of the stable influence units, stable association units, change trigger units, and change association units in the conversation text semantic association network, and screening out units whose sum of association strengths is higher than a preset strength threshold, includes: Obtain all semantic association edges corresponding to each stable influence unit, stable association unit, change trigger unit, and change association unit in the conversation text semantic association network; Extract the association strength value of each semantic association edge, sum the association strength values ​​of all semantic association edges of each unit, and obtain the total association strength of each unit; Setting a preset strength threshold of the sum of association strengths, wherein the preset strength threshold is determined according to the sum of the average association strengths of all units in the conversation text semantic association network; Compare the sum of the association strengths of each unit with a preset strength threshold, and select units whose sum of association strengths is higher than the preset strength threshold to form a target unit set; Analyze the topic representation vocabulary of each unit in the target unit set to determine whether the subject of the unit is directly related to supplier relationship maintenance; The units with themes directly related to supplier relationship maintenance are retained, and the units with no relevance to themes are eliminated to obtain a preliminary screening unit set; Check the supplier interaction intention corresponding to each unit in the preliminary screening unit set. If the intention type is problem feedback intention or risk suggestion intention, further verify whether the sum of the association strength of the unit is continuously higher than the preset strength threshold; The units that have passed the verification are finally confirmed and retained in the preliminary screening unit set to form the final optimization and improvement node set.

10. A supplier relationship management system combined with big data analysis, characterized in that: The supplier relationship management system combined with big data analysis includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the supplier relationship management method combined with big data analysis as described in any one of claims 1 to 9 above.

Images