Supply chain settlement feedback big data analysis method and system based on artificial intelligence

Through the big data analysis method of supply chain settlement feedback based on artificial intelligence, the focus attention vector sequence and panoramic semantic vector representation are used to solve the problems of low efficiency and insufficient accuracy of supply chain settlement feedback analysis in the existing technology, and achieve high efficiency and accuracy The text understanding and intention mining provide deep and breadth data support for supply chain management.

CN118484480BActive Publication Date: 2024-11-22YUNJING (SHANGHAI) ELECTRONIC TECH CO LTD
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202410629914.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-21
Publication Date
2024-11-22
Estimated Expiration
2044-05-21

AI Technical Summary

Technical Problem

The existing supply chain settlement feedback analysis methods are inefficient and error-prone, unable to effectively understand the deep semantics of text, and are not very accurate when dealing with complex text relationships, which cannot meet the needs of modern enterprises for efficient and accurate data processing.

Method used

Adopt the big data analysis method of supply chain settlement feedback based on artificial intelligence. By obtaining the focus vector sequence and panoramic semantic vector representation, the supply chain text mining network is used to understand and intention mining, and the network is trained to minimize feature distance goals and improve analysis. Efficiency and accuracy.

Benefits of technology

It significantly improves the efficiency and accuracy of supply chain settlement feedback text analysis, provides more scientific and comprehensive data support, and provides in-depth and breadth of semantic information for supply chain management decisions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118484480B_ABST
    Figure CN118484480B_ABST
Patent Text Reader

Abstract

The present application provides a supply chain settlement feedback big data analysis method and system based on artificial intelligence, which aims to use deep learning technology to perform panoramic semantic vector representation of supply chain settlement feedback text, so as to capture the deep semantic information of the text and improve the accuracy and depth of text understanding. At the same time, the text is refined through multiple focus vector sequences, which further improves the accuracy and comprehensiveness of text analysis. In addition, the supply chain text mining network is trained by minimizing the training goal of the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence, so as to generate a trained supply chain text mining network, which provides more accurate and efficient support for subsequent supply chain settlement feedback text analysis. As a result, the accuracy and efficiency of supply chain settlement feedback text analysis are improved, providing more scientific and comprehensive data support for the supply chain management decision-making of enterprises.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of computer technology, and more specifically, to an artificial intelligence-based supply chain settlement feedback big data analysis method and system. Background Art

[0002] With the development of globalization and networking, supply chain management has become an indispensable part of corporate operations. Among them, supply chain settlement feedback is an important indicator for evaluating supply chain efficiency and customer satisfaction. However, traditional supply chain settlement feedback analysis methods are often based on manual processing, which is inefficient and error-prone, and cannot meet the needs of modern enterprises for efficient and accurate data processing.

[0003] In recent years, with the rapid development of artificial intelligence technology, the use of artificial intelligence technology for big data analysis has become an important trend in the development of the industry. Especially in the fields of natural language processing and deep learning, artificial intelligence technology has demonstrated powerful text analysis and understanding capabilities. However, in the analysis of supply chain settlement feedback text, existing technical methods still have certain limitations, such as the inability to understand the deep semantics of the text and low accuracy in processing complex text relationships. Summary of the invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiment of the present application provides a supply chain settlement feedback big data analysis method based on artificial intelligence, the method comprising:

[0005] Acquire a sample supply chain settlement feedback text, and acquire a plurality of focus attention vector sequences of the sample supply chain settlement feedback text, each focus attention vector sequence representing a conversation text understanding result expressed by each text attention unit in the sample supply chain settlement feedback text;

[0006] Performing panoramic semantic vector representation on the example supply chain settlement feedback text through a supply chain text mining network to generate a target panoramic semantic vector sequence of the example supply chain settlement feedback text;

[0007] Through the supply chain text mining network, a panoramic semantic vector is estimated for the example supply chain settlement feedback text according to the multiple focused attention vector sequences, and an estimated panoramic semantic vector sequence of the example supply chain settlement feedback text is generated, wherein the estimated panoramic semantic vector sequence at least represents a conversation text understanding result expressed by multiple text attention units in the example supply chain settlement feedback text;

[0008] Based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence, the supply chain text mining network is trained to generate a trained supply chain text mining network.

[0009] In a possible implementation of the first aspect, performing panoramic semantic vector representation on the example supply chain settlement feedback text by using a supply chain text mining network to generate a target panoramic semantic vector sequence of the example supply chain settlement feedback text includes:

[0010] Performing attention processing on each of the multiple focused attention vector sequences to generate an attention vector sequence corresponding to the corresponding focused attention vector sequence;

[0011] The attention vector sequences corresponding to the multiple focused attention vector sequences are integrated to generate a target panoramic semantic vector sequence of the example supply chain settlement feedback text.

[0012] In a possible implementation of the first aspect, the step of estimating a panoramic semantic vector of the example supply chain settlement feedback text according to the multiple focused attention vector sequences through the supply chain text mining network to generate an estimated panoramic semantic vector sequence of the example supply chain settlement feedback text includes:

[0013] The multiple focused attention vector sequences are fused to generate an estimated panoramic semantic vector sequence of the example supply chain settlement feedback text.

[0014] In a possible implementation of the first aspect, the supply chain text mining network includes a teaching text mining network and a taught text mining network, and the teaching text mining network is generated by optimizing parameter configuration information in the taught text mining network;

[0015] The step of performing panoramic semantic vector representation on the example supply chain settlement feedback text through a supply chain text mining network to generate a target panoramic semantic vector sequence of the example supply chain settlement feedback text includes:

[0016] Through the teaching text mining network, the example supply chain settlement feedback text is represented by a panoramic semantic vector to generate a teaching panoramic semantic vector sequence of the example supply chain settlement feedback text;

[0017] Performing a panoramic semantic vector representation on the example supply chain settlement feedback text through the taught text mining network to generate a taught panoramic semantic vector sequence of the example supply chain settlement feedback text;

[0018] The step of training the supply chain text mining network based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence to generate a trained supply chain text mining network includes:

[0019] Based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence, and the feature distance between the teaching panoramic semantic vector sequence and the taught panoramic semantic vector sequence, the taught text mining network is trained to generate the trained supply chain text mining network.

[0020] In a possible implementation of the first aspect, the step of estimating a panoramic semantic vector of the example supply chain settlement feedback text according to the multiple focused attention vector sequences through the supply chain text mining network to generate an estimated panoramic semantic vector sequence of the example supply chain settlement feedback text includes:

[0021] Through the teaching text mining network, a panoramic semantic vector is estimated for the example supply chain settlement feedback text according to the multiple focus attention vector sequences to generate a teaching estimated panoramic semantic vector sequence of the example supply chain settlement feedback text;

[0022] By means of the taught text mining network, a panoramic semantic vector is estimated for the example supply chain settlement feedback text according to the multiple focused attention vector sequences, so as to generate a taught estimated panoramic semantic vector sequence of the example supply chain settlement feedback text;

[0023] The step of training the taught text mining network based on a training objective of minimizing a feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence comprises:

[0024] Based on the training objectives of minimizing the feature distance between the teaching panoramic semantic vector sequence and the teaching estimated panoramic semantic vector sequence, and minimizing the feature distance between the taught panoramic semantic vector sequence and the taught estimated panoramic semantic vector sequence, the taught text mining network is trained to generate the trained supply chain text mining network.

[0025] In a possible implementation of the first aspect, the supply chain text mining network includes a teaching text mining network and a taught text mining network, and the teaching text mining network is generated by optimizing parameter configuration information in the taught text mining network;

[0026] The step of obtaining a plurality of focus vector sequences of the example supply chain settlement feedback text includes:

[0027] Performing a focused attention vector representation on the example supply chain settlement feedback text through the teaching text mining network to generate X teaching focused attention vector sequences, and performing a focused attention vector representation on the example supply chain settlement feedback text through the taught text mining network to generate X taught focused attention vector sequences;

[0028] Walking the X taught focus attention vector sequences, and for the currently walked target taught focus attention vector sequence, extracting a target taught focus attention vector sequence having the smallest feature distance with the target taught focus attention vector sequence from the X taught focus attention vector sequences;

[0029] configuring a focus attention vector sequence combination including the target teaching focus attention vector sequence and the target taught focus attention vector sequence until the walk is completed, generating X focus attention vector sequence combinations;

[0030] Based on the semantic relevance corresponding to each of the X focused attention vector sequence combinations, determining a text mining distinction parameter between the taught text mining network and the teaching text mining network; the text mining distinction parameter is negatively correlated with the semantic relevance corresponding to each of the focused attention vector sequence combinations;

[0031] The step of training the supply chain text mining network based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence to generate a trained supply chain text mining network includes:

[0032] Based on the training objectives of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence, and minimizing the text mining distinction parameters between the taught text mining network and the teaching text mining network, the taught text mining network is trained to generate the trained supply chain text mining network.

[0033] In a possible implementation of the first aspect, the method further includes:

[0034] Based on a text mining instruction for a target supply chain settlement feedback text in the supply chain settlement feedback big data, obtaining the target supply chain settlement feedback text;

[0035] Performing panoramic semantic vector representation on the target supply chain settlement feedback text through the trained supply chain text mining network to generate a target panoramic semantic vector sequence of the target supply chain settlement feedback text;

[0036] According to the target panoramic semantic vector sequence of the target supply chain settlement feedback text, supply chain feedback intention mining is performed on the target supply chain settlement feedback text to generate a supply chain feedback intention mining result;

[0037] Corresponding information push is performed according to the supply chain feedback intention mining results.

[0038] In a possible implementation of the first aspect, the step of performing supply chain feedback intention mining on the target supply chain settlement feedback text based on the target panoramic semantic vector sequence of the target supply chain settlement feedback text to generate a supply chain feedback intention mining result includes:

[0039] Pre-building an intention classification model based on deep learning, wherein the intention classification model is used to identify supply chain feedback intentions based on a target panoramic semantic vector sequence of the target supply chain settlement feedback text;

[0040] Using the existing labeled data set to train the intent classification model to obtain a trained intent classification model;

[0041] Inputting the target panoramic semantic vector sequence into the trained intent classification model, extracting key semantic features in the target panoramic semantic vector sequence through the intent classification model, and mapping them to predefined intent categories to output one or more intent labels and the confidence corresponding to each intent label, wherein the intent label represents the supply chain feedback intent expressed in the target supply chain settlement feedback text;

[0042] The intention tags with confidence less than the preset confidence are eliminated, and the similar intention tags in the remaining intention tags are merged, and the processed intention tags are output as the supply chain feedback intention mining results.

[0043] In a possible implementation of the first aspect, the step of executing corresponding information push according to the supply chain feedback intention mining result includes:

[0044] Establishing a mapping table, through which the supply chain feedback intention is associated with the corresponding information push content;

[0045] Analyze the historical behavior pattern data of the target user corresponding to the supply chain feedback intention mining result, and generate a personalized push strategy for the target user;

[0046] According to the supply chain feedback intention mining results, candidate push content information matching the current supply chain feedback intention of the target user is screened out from the mapping table, and target push content information is extracted from the candidate push content information based on the personalized push strategy for push.

[0047] On the other hand, an embodiment of the present application also provides an artificial intelligence-based supply chain settlement feedback big data analysis system, 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.

[0048] Based on the above aspects, by obtaining multiple focused attention vector sequences of the sample supply chain settlement feedback text, the key information and details in the text can be accurately captured. The supply chain text mining network is used to represent the sample text with panoramic semantic vectors, and the target panoramic semantic vector sequence is generated, so that the overall semantic information of the text can be fully and deeply displayed. Furthermore, by estimating the panoramic semantic vector of the sample text by focusing on the attention vector sequence and generating the estimated panoramic semantic vector sequence, it not only improves the depth and breadth of text understanding, but also provides a more refined reference standard for subsequent network training. Finally, based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence, the supply chain text mining network is trained, which significantly improves the accuracy and generalization ability of the network. As a result, while improving the efficiency of supply chain settlement feedback text analysis, it also greatly improves the accuracy and depth of analysis, providing more scientific and comprehensive data support for supply chain management decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram of the execution flow of the supply chain settlement feedback big data analysis method based on artificial intelligence provided in an embodiment of the present application.

[0050] Figure 2 It is a schematic diagram of the hardware architecture of the supply chain settlement feedback big data analysis system based on artificial intelligence provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] The present application will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of an artificial intelligence-based supply chain settlement feedback big data analysis method provided by an embodiment of the present application. The artificial intelligence-based supply chain settlement feedback big data analysis method is introduced in detail below.

[0052] Step S110, obtaining a sample supply chain settlement feedback text, and obtaining a plurality of focus attention vector sequences of the sample supply chain settlement feedback text, each focus attention vector sequence representing a conversation text understanding result expressed by each text attention unit in the sample supply chain settlement feedback text.

[0053] For example, the sample supply chain settlement feedback text refers to the feedback records collected from customers or suppliers during the actual supply chain transaction process, which usually contains various issues, opinions and suggestions related to supply chain settlement, and is an important data source for understanding and improving supply chain processes. For example, a text may be "In this transaction, the delivery time was delayed by two days, which resulted in our inability to complete the settlement on time. We hope that your company can improve the efficiency of logistics distribution." This is a sample supply chain settlement feedback text, which reflects the impact of delivery delays on settlement.

[0054] The focused attention vector sequence refers to the semantic features of each text attention unit (such as sentences, phrases or keywords) extracted from the example supply chain settlement feedback text through natural language processing and deep learning technology. These text attention units carry important semantic information in the text. For example, in the above example, "delivery time was delayed by two days" can be a text attention unit. Through the deep learning model, this text attention unit can be converted into a focused attention vector, which captures the semantics of this text attention unit, that is, the information of delivery delay.

[0055] Based on the above, in this embodiment, the server first accesses a database storing sample supply chain settlement feedback texts, which can be feedback records from customers or suppliers in previous supply chain transactions, including various settlement-related questions, opinions and suggestions. The server uses natural language processing technology to pre-process these sample supply chain settlement feedback texts, such as word segmentation, stop word removal, etc.

[0056] Next, the server runs a pre-trained deep learning model, such as BERT or Transformer, to identify the text attention units of these processed sample supply chain settlement feedback texts. These text attention units can be sentences, phrases, or keywords, which carry important semantic information in the sample supply chain settlement feedback texts. For each identified text attention unit, a focused attention vector can be generated, which captures the semantic features of the attention unit of the sample supply chain settlement feedback text.

[0057] For example, in a feedback text about supply chain settlement delays, a text attention unit can be "delivery time delayed by two days", and the corresponding focused attention vector will encode the semantics of this unit, including key information such as delivery and time delay. The server will combine the focused attention vectors of all identified text attention units into a sequence, which is the focused attention vector sequence of the sample supply chain settlement feedback text.

[0058] Step S120, performing panoramic semantic vector representation on the example supply chain settlement feedback text through a supply chain text mining network to generate a target panoramic semantic vector sequence of the example supply chain settlement feedback text.

[0059] Exemplarily, the panoramic semantic vector representation refers to converting text into a high-dimensional vector through a deep learning model, capturing the overall semantic information of the example supply chain settlement feedback text. Unlike a single focused attention vector, the panoramic semantic vector representation takes into account all key information and their interrelationships in the example supply chain settlement feedback text, thereby providing a more comprehensive and accurate text semantic description. For example, continuing with the previous example text "In this transaction, the delivery time was delayed by two days, which resulted in our inability to complete the settlement on time. We hope that your company can improve the efficiency of logistics distribution." As an example. The panoramic semantic vector representation will not only focus on the key information that "the delivery time was delayed by two days", but will also consider the impact of the delay on settlement, customer expectations, etc., to generate a vector representation that integrates all important information in the text.

[0060] The target panoramic semantic vector sequence is a series of panoramic semantic vectors obtained by processing the example supply chain settlement feedback text through the supply chain text mining network, which not only captures the semantic information of each key unit in the text, but also integrates the global information of the entire example supply chain settlement feedback text, thereby providing an accurate and comprehensive text semantic description. For example, the target panoramic semantic vector sequence will include panoramic semantic vectors reflecting multiple aspects such as delivery delays, settlement impacts, and customer expectations. These vectors are arranged in order to form a vector sequence that can fully describe the semantics of the text.

[0061] That is, in this embodiment, after obtaining the focused attention vector sequence of the example supply chain settlement feedback text, the server will call the supply chain text mining network, which can understand the professional terms and context in the supply chain field.

[0062] The server inputs the sample supply chain settlement feedback text into the supply chain text mining network, which first performs a deep semantic analysis on each text attention unit. By analyzing the grammatical structure, lexical meaning, and contextual relationship of the text, the supply chain text mining network can generate a panoramic semantic vector, which not only captures the semantics of a single text attention unit, but also integrates the global information of the entire sample supply chain settlement feedback text.

[0063] For example, for the feedback text about delivery delay mentioned above, the panoramic semantic vector will contain information such as the impact of delivery delay on the overall efficiency of the supply chain, possible causes, and potential solutions. The server will combine the panoramic semantic vectors of all text attention units into a sequence, namely the target panoramic semantic vector sequence, which comprehensively represents the overall semantics of the sample supply chain settlement feedback text.

[0064] Step S130, through the supply chain text mining network, a panoramic semantic vector is estimated for the example supply chain settlement feedback text based on the multiple focused attention vector sequences, and an estimated panoramic semantic vector sequence of the example supply chain settlement feedback text is generated, wherein the estimated panoramic semantic vector sequence at least represents the conversation text understanding results expressed by multiple text attention units in the example supply chain settlement feedback text.

[0065] Next, the server can use the previously acquired focused attention vector sequence and the supply chain text mining network to estimate the panoramic semantic vector of the example supply chain settlement feedback text, thereby verifying whether the supply chain text mining network can accurately reconstruct the overall semantics of the text by focusing on important text units.

[0066] The server takes the sequence of focused attention vectors as input and processes it through the supply chain text mining network. The supply chain text mining network estimates the panoramic semantics of the entire text based on the local semantic information contained in the sequence of focused attention vectors. In this process, the supply chain text mining network fuses the semantics of different text attention units to generate an estimated panoramic semantic vector sequence that can reflect the overall meaning of the text.

[0067] For example, in the feedback text about delayed delivery time, the supply chain text mining network estimates the panoramic semantics of the text based on the focused attention vectors of key units such as "the delivery time was delayed by two days". The estimation process takes into account the relationships between the various parts of the text to ensure that the estimated panoramic semantic vector can accurately reflect the overall meaning of the text.

[0068] Step S140, training the supply chain text mining network based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence to generate a trained supply chain text mining network.

[0069] Finally, the server uses the target panoramic semantic vector sequence and the estimated panoramic semantic vector sequence to train the supply chain text mining network. The training goal is to minimize the feature distance between the target panoramic semantic vector sequence and the estimated panoramic semantic vector sequence, that is, to allow the supply chain text mining network to better reconstruct the overall semantics of the text from the focused attention vector sequence through learning.

[0070] The server can use the back-propagation algorithm to adjust the parameters of the supply chain text mining network by calculating the gap between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence. The process is iterative, and the server will repeat the above steps many times until the supply chain text mining network can accurately estimate the panoramic semantics of the text from the focused attention vector sequence.

[0071] For example, during the training of feedback text on delayed delivery time, if the supply chain text mining network cannot accurately estimate the panoramic semantics of the text from the focused attention vector at first, the server will adjust the parameters of the supply chain text mining network through the back propagation algorithm. As the training progresses, the supply chain text mining network will gradually learn how to better integrate the semantic information of different text attention units, so as to more accurately estimate the overall meaning of the text. In the end, the server will obtain a trained supply chain text mining network that can effectively process similar supply chain settlement feedback text and provide valuable semantic analysis results.

[0072] Based on the above steps, by obtaining multiple focused attention vector sequences of the sample supply chain settlement feedback text, the key information and details in the text can be accurately captured. The supply chain text mining network is used to represent the sample text with panoramic semantic vectors, and the target panoramic semantic vector sequence is generated, so that the overall semantic information of the text can be fully and deeply displayed. Furthermore, by focusing on the attention vector sequence to estimate the panoramic semantic vector of the sample text and generate an estimated panoramic semantic vector sequence, it not only improves the depth and breadth of text understanding, but also provides a more refined reference standard for subsequent network training. Finally, based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence, the supply chain text mining network is trained, which significantly improves the accuracy and generalization ability of the network. As a result, while improving the efficiency of supply chain settlement feedback text analysis, it also greatly improves the accuracy and depth of analysis, providing more scientific and comprehensive data support for supply chain management decisions.

[0073] In a possible implementation, step S120 may include:

[0074] Step S121, performing attention processing on each of the multiple focused attention vector sequences to generate an attention vector sequence corresponding to the corresponding focused attention vector sequence.

[0075] Step S122, integrating the attention vector sequences corresponding to the multiple focused attention vector sequences to generate a target panoramic semantic vector sequence of the example supply chain settlement feedback text.

[0076] In this embodiment, the server can perform attention processing on each focus attention vector sequence in order to emphasize the key information in the sample supply chain settlement feedback text while weakening or ignoring the secondary information. Specifically, the server calculates the weight of each focus attention vector, which reflects the importance of the focus attention vector in the overall semantics. The calculation of the weight can be based on the statistical characteristics of the vector, the degree of correlation with other vectors, or predefined rules.

[0077] For example, in a text describing supply chain settlement, words such as “delayed payment” and “liquidated penalty” can be assigned higher attention weights because they are crucial to understanding the feedback situation of supply chain settlement.

[0078] As a result, the server generates a sequence of attention vectors corresponding to each sequence of focused attention vectors, which not only contain the original semantic information but also have weights attached to represent their importance.

[0079] After obtaining multiple attention vector sequences, the server will perform an integration operation to generate a target panoramic semantic vector sequence for the sample supply chain settlement feedback text. During the integration process, the server will first perform a weighted summation of all attention vectors. Attention vectors with higher weights will have a greater impact on the result during the summation process, thereby ensuring that important semantic information is fully reflected in the panoramic semantic vector.

[0080] For example, if key words such as “delayed payment” and “penalty for breach of contract” are assigned higher weights in the attention processing stage, the semantic information of these key words will be strengthened in the integrated panoramic semantic vector.

[0081] The final generated panoramic semantic vector sequence is a high-dimensional vector representation that captures the overall semantic information of the example supply chain settlement feedback text, in which key information is appropriately emphasized. The panoramic semantic vector sequence can be used for subsequent text classification, clustering, information retrieval and other tasks to provide data support for supply chain management and decision-making.

[0082] In a possible implementation, step S130 includes: fusing the multiple focused attention vector sequences to generate an estimated panoramic semantic vector sequence of the example supply chain settlement feedback text.

[0083] In this embodiment, the focus attention vector sequence captures different aspects of semantic information in the text. For example, some focus attention vectors may represent logistics information in the supply chain, some may represent payment status, and some may reflect customer satisfaction.

[0084] In order to generate an estimated panoramic semantic vector sequence that can fully reflect the overall semantics of the text, the server uses a specific fusion algorithm, such as weighted averaging, concatenation, or feature fusion through a deep learning model. In this process, the server can consider the importance of different focus vectors in the text and assign corresponding weights to them.

[0085] Taking weighted averaging as an example, the server will assign a weight value to each focus vector based on its importance and relevance. These weight values ​​can be learned from training data or set based on domain knowledge or expert experience. Then, the server will sum all weighted focus vectors to obtain a comprehensive vector representation, which integrates the information of each focus vector.

[0086] Through the above fusion processing, the server obtains a comprehensive vector that integrates the information of multiple focused attention vectors. The comprehensive vector is the estimated panoramic semantic vector of the example supply chain settlement feedback text, which captures the overall semantic information of the text, including logistics status, payment status, customer satisfaction and other aspects. The estimated panoramic semantic vector sequence can be regarded as a highly condensed representation of the text, which can not only be used for tasks such as text classification and clustering, but also as a basis for further analysis and mining of supply chain settlement feedback text.

[0087] For example, in supply chain risk management, the server can use this estimated panoramic semantic vector sequence to identify potential supply chain problems, such as delayed delivery, quality issues, etc. By comparing the estimated panoramic semantic vector sequences of different time periods, the server can also discover changing trends and abnormal situations in the supply chain, so as to take corresponding countermeasures in a timely manner.

[0088] In a possible implementation, the supply chain text mining network includes a teaching text mining network and a taught text mining network, and the teaching text mining network is generated by optimizing parameter configuration information in the taught text mining network.

[0089] Step S120 may further include:

[0090] Step S123, performing panoramic semantic vector representation on the example supply chain settlement feedback text through the teaching text mining network, and generating a teaching panoramic semantic vector sequence of the example supply chain settlement feedback text.

[0091] Step S124, performing panoramic semantic vector representation on the example supply chain settlement feedback text through the taught text mining network, and generating a taught panoramic semantic vector sequence of the example supply chain settlement feedback text.

[0092] Step S140 may include: training the taught text mining network based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence, and the feature distance between the teaching panoramic semantic vector sequence and the taught panoramic semantic vector sequence, to generate the trained supply chain text mining network.

[0093] In this embodiment, in the supply chain text mining task, the server adopts a special network structure, including a teaching text mining network and a taught text mining network. The design purpose of this structure is to improve the accuracy and efficiency of text mining.

[0094] In detail, the server first constructs two text mining networks: a teaching text mining network and a taught text mining network, which are similar in structure but different in parameter configuration and initial state. The teaching text mining network is generated by optimizing the parameter configuration information in the taught text mining network, so the teaching text mining network can understand how to better perform the text mining task to some extent.

[0095] When the server receives the example supply chain settlement feedback text, it first uses the teaching text mining network to perform a panoramic semantic vector representation on the example supply chain settlement feedback text. This process involves inputting the text into the network, and after a series of complex calculations and processing, it finally outputs a panoramic semantic vector sequence, which is also called the teaching panoramic semantic vector sequence because it is generated by a more experienced and optimized teaching text mining network.

[0096] Next, the server will use the taught text mining network to perform panoramic semantic vector representation on the same example supply chain settlement feedback text. Similar to the taught text mining network, this process will also output a panoramic semantic vector sequence, but due to differences in network parameters and states, this panoramic semantic vector sequence may be different from the taught panoramic semantic vector sequence. The sequence generated by the taught text mining network is called the taught panoramic semantic vector sequence.

[0097] After generating the teaching and taught panoramic semantic vector sequences, the next goal of the server is to train the taught text mining network to make its performance closer to the teaching text mining network. To this end, the server adopts a training objective based on feature distance. Specifically, it aims to minimize the feature distance between the estimated panoramic semantic vector sequence (generated by the taught text mining network) and the target panoramic semantic vector sequence (generated by the teaching text mining network or based on other reliable sources). At the same time, the server also considers minimizing the feature distance between the teaching panoramic semantic vector sequence and the taught panoramic semantic vector sequence.

[0098] By continuously adjusting the parameters of the taught text mining network to minimize the above-mentioned feature distance, the server is eventually able to generate a trained supply chain text mining network that is closer in performance to the taught text mining network and is therefore able to perform supply chain text mining tasks more accurately. This training method not only improves the accuracy of text mining, but also accelerates the convergence speed of the network, enabling the server to process large amounts of supply chain text data more efficiently.

[0099] In a possible implementation, step S130 may further include:

[0100] Step S131, using the teaching text mining network, performing panoramic semantic vector estimation on the example supply chain settlement feedback text according to the multiple focus vector sequences, and generating a teaching estimated panoramic semantic vector sequence of the example supply chain settlement feedback text.

[0101] Step S132, using the taught text mining network, performing panoramic semantic vector estimation on the example supply chain settlement feedback text according to the multiple focused attention vector sequences, and generating a taught estimated panoramic semantic vector sequence of the example supply chain settlement feedback text.

[0102] The step of training the taught text mining network based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence includes: training the taught text mining network based on the training goals of minimizing the feature distance between the teaching panoramic semantic vector sequence and the teaching estimated panoramic semantic vector sequence, and minimizing the feature distance between the taught panoramic semantic vector sequence and the taught estimated panoramic semantic vector sequence, to generate the trained supply chain text mining network.

[0103] In this embodiment, during the supply chain text mining process, the server uses the teaching text mining network and the taught text mining network to estimate the panoramic semantic vector of the example supply chain settlement feedback text.

[0104] In detail, the server first uses the teaching text mining network to estimate the panoramic semantic vector of the example supply chain settlement feedback text based on the multiple focused attention vector sequences extracted previously. In this process, the teaching text mining network comprehensively considers various focused attention points in the text, such as product delivery time, payment status, quality feedback, etc., and integrates this information into a comprehensive semantic representation. After being processed by the teaching text mining network, the server generates a teaching estimated panoramic semantic vector sequence, which captures the overall semantic information of the example supply chain settlement feedback text from the perspective of the teaching text mining network.

[0105] Next, the server uses the taught text mining network to estimate the panoramic semantic vector for the same example supply chain settlement feedback text. The taught text mining network also refers to multiple focused attention vector sequences during processing, but because its initial parameters and structure are different from those of the teaching text mining network, the generated taught estimated panoramic semantic vector sequence may be different from the estimated result of the teaching text mining network.

[0106] After generating the teaching and taught estimated panoramic semantic vector sequences, the server aims to improve the performance of the taught text mining network so that it can produce results closer to the teaching text mining network when processing the same task. To this end, the server sets a training goal, which is to minimize the feature distance between the teaching panoramic semantic vector sequence and the teaching estimated panoramic semantic vector sequence, and to minimize the feature distance between the taught panoramic semantic vector sequence and the taught estimated panoramic semantic vector sequence.

[0107] In order to achieve the above training objectives, the server will iteratively train the taught text mining network. In each iteration, the server will calculate the feature distance between the teaching and taught estimated panoramic semantic vector sequences and the target sequence, and adjust the parameters of the taught text mining network based on this distance. Through continuous optimization and adjustment, the performance of the taught text mining network will gradually improve, and the estimated panoramic semantic vector sequence it generates will become closer and closer to the target sequence.

[0108] When the performance of the taught text mining network reaches the preset standard, or the optimization index (such as the loss function value) in the training process converges to a stable level, the server will stop training and consider the taught text mining network to be trained. At this point, the network has similar text mining capabilities as the taught text mining network, and can accurately generate a panoramic semantic vector representation of the sample supply chain settlement feedback text based on the focused attention vector sequence.

[0109] In a possible implementation, the supply chain text mining network includes a teaching text mining network and a taught text mining network, and the teaching text mining network is generated by optimizing parameter configuration information in the taught text mining network.

[0110] Step S110 may include:

[0111] Step S111, focusing attention vector representation is performed on the example supply chain settlement feedback text through the teaching text mining network to generate X teaching focusing attention vector sequences, and focusing attention vector representation is performed on the example supply chain settlement feedback text through the taught text mining network to generate X taught focusing attention vector sequences.

[0112] Step S112, wandering the X taught focus attention vector sequences, and for the currently wandered target taught focus attention vector sequence, extracting a target taught focus attention vector sequence having the smallest feature distance with the target taught focus attention vector sequence from the X taught focus attention vector sequences.

[0113] Step S113, configuring a focus attention vector sequence combination including the target teaching focus attention vector sequence and the target taught focus attention vector sequence, until the walk is completed, to generate X focus attention vector sequence combinations.

[0114] Step S114: determining a text mining distinction parameter between the taught text mining network and the teaching text mining network based on the semantic relevance corresponding to each of the X focused attention vector sequence combinations. The text mining distinction parameter is negatively correlated with the semantic relevance corresponding to each of the focused attention vector sequence combinations.

[0115] Step S140 may include: training the taught text mining network based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence, and minimizing the text mining distinction parameter between the taught text mining network and the teaching text mining network, to generate the trained supply chain text mining network.

[0116] In this embodiment, first, the server uses the teaching text mining network to process the sample supply chain settlement feedback text to generate X teaching focus attention vector sequences, which capture key information of different aspects in the text, such as delivery time, product quality, customer feedback, etc. Similarly, the server also uses the taught text mining network to process the same text to generate X taught focus attention vector sequences.

[0117] Next, the server will start to walk through the X taught focus attention vector sequences. For the target taught focus attention vector sequence currently walked to, the server will find the target taught focus attention vector sequence with the smallest feature distance from the X taught focus attention vector sequences. This process is to find the most similar vector representation of the same text content under two different network representations.

[0118] After finding the target taught focused attention vector sequence that is most similar to the target taught focused attention vector sequence, the server will configure the two vector sequences into a focused attention vector sequence combination. This process will continue until all X taught focused attention vector sequences have been walked, and finally X focused attention vector sequence combinations are generated.

[0119] The server will determine the text mining difference parameter between the taught text mining network and the teaching text mining network based on the semantic relevance of each combination of the X focus attention vector sequence combinations. This parameter actually reflects the degree of difference between the two networks when processing the same text. The higher the semantic relevance, the smaller the difference between the two networks, and therefore the smaller the text mining difference parameter; conversely, the larger the parameter, the negatively correlated with the difference between the taught network and the teaching network.

[0120] After obtaining the text mining distinguishing parameters, the next goal of the server is to train the taught text mining network so that it can get closer to the performance of the teaching network when processing supply chain text. To this end, the server sets two training goals: one is to minimize the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence; the other is to minimize the text mining distinguishing parameters between the taught text mining network and the teaching text mining network. Through the joint optimization of these two goals, the server trains the taught text mining network until a trained supply chain text mining network is generated. When processing supply chain text, this network will be able to more accurately capture the key information in the text and generate a more accurate panoramic semantic vector representation.

[0121] In a possible implementation, the method further includes:

[0122] Step S150, based on the text mining instruction for the target supply chain settlement feedback text in the supply chain settlement feedback big data, the target supply chain settlement feedback text is obtained.

[0123] In this embodiment, the server first receives a text mining instruction for supply chain settlement feedback big data, which may come from a supply chain management system or a data analyst, and the purpose is to conduct in-depth analysis on a specific target supply chain settlement feedback text. According to this text mining instruction, the server retrieves and obtains the target supply chain settlement feedback text from the massive supply chain settlement feedback data.

[0124] Step S160, performing panoramic semantic vector representation on the target supply chain settlement feedback text through the trained supply chain text mining network to generate a target panoramic semantic vector sequence of the target supply chain settlement feedback text.

[0125] Once the target supply chain settlement feedback text is obtained, the server will use the previously trained supply chain text mining network to perform a panoramic semantic vector representation on the target supply chain settlement feedback text. This process involves inputting the target supply chain settlement feedback text into the supply chain text mining network. Through the processing of the supply chain text mining network, the semantic information in the target supply chain settlement feedback text is converted into a target panoramic semantic vector sequence. The target panoramic semantic vector sequence can comprehensively and accurately capture the deep semantic information in the text, providing a basis for subsequent supply chain feedback intention mining.

[0126] Step S170, performing supply chain feedback intention mining on the target supply chain settlement feedback text according to the target panoramic semantic vector sequence of the target supply chain settlement feedback text, and generating a supply chain feedback intention mining result.

[0127] After obtaining the target panoramic semantic vector sequence, the server will mine the supply chain feedback intention of the target supply chain settlement feedback text based on this target panoramic semantic vector sequence. This process mainly analyzes the patterns and associations in the target panoramic semantic vector sequence to identify and interpret the implicit supply chain feedback intention in the target supply chain settlement feedback text. For example, the target supply chain settlement feedback text may mention delivery delays, product quality issues, or payment disputes, which are all key feedback points in the supply chain. The server will use advanced algorithms and models to accurately identify and interpret these feedback intentions.

[0128] Step S180: executing corresponding information push according to the supply chain feedback intention mining result.

[0129] After the above mining process, the server will generate a supply chain feedback intention mining result, which can include a series of specific descriptions of supply chain problems, cause analysis, and possible solution suggestions. Subsequently, the server will execute corresponding information push based on this supply chain feedback intention mining result. For example, if the supply chain feedback intention mining result shows that there is a problem of delayed delivery, the server can automatically send early warning information to relevant supply chain managers to remind them to take timely measures to solve the problem. Such information push can help enterprises quickly respond to problems in the supply chain and improve the efficiency and accuracy of supply chain management.

[0130] In a possible implementation, step S170 includes:

[0131] Step S171, pre-build an intention classification model based on deep learning, and the intention classification model is used to identify supply chain feedback intentions based on the target panoramic semantic vector sequence of the target supply chain settlement feedback text.

[0132] First, the server needs to build a deep learning-based intent classification model, which is designed to receive a target panoramic semantic vector sequence as input and identify the implicit supply chain feedback intent. The process of building the intent classification model can include selecting an appropriate network structure (such as a convolutional neural network, a recurrent neural network, etc.) and determining the parameters and training strategy of the intent classification model.

[0133] Step S172, using the existing annotated data set to train the intent classification model to obtain a trained intent classification model.

[0134] To train this intent classification model, the server uses existing annotated datasets that contain a large number of supply chain settlement feedback texts with known feedback intents and their corresponding target panoramic semantic vector sequences. By inputting these annotated data into the intent classification model and adjusting the parameters of the intent classification model to minimize the difference between the predicted intent and the actual intent, a model that can accurately identify the supply chain feedback intent is trained.

[0135] Step S173, input the target panoramic semantic vector sequence into the trained intent classification model, extract the key semantic features in the target panoramic semantic vector sequence through the intent classification model, and map them to predefined intent categories to output one or more intent labels and the confidence corresponding to each intent label, wherein the intent label represents the supply chain feedback intention expressed in the target supply chain settlement feedback text.

[0136] When the server receives a target panoramic semantic vector sequence of a target supply chain settlement feedback text, it inputs the target panoramic semantic vector sequence into the trained intent classification model. The intent classification model first extracts the key semantic features in the target panoramic semantic vector sequence, which can represent the main information and intent in the text.

[0137] Next, the extracted key semantic features are mapped to predefined intent categories, which may include common supply chain feedback intents such as delivery delays, product quality issues, and payment disputes. The intent classification model outputs an intent label for each identified intent and gives the confidence level corresponding to the label, which indicates the degree of certainty of the model's judgment on the intent.

[0138] Step S174, remove the intention tags with confidence less than the preset confidence, merge the similar intention tags in the remaining intention tags, and output the processed intention tags as the supply chain feedback intention mining results.

[0139] After obtaining a series of intent tags and corresponding confidence levels, the server will perform further processing. First, those intent tags with confidence levels less than the preset confidence levels are removed because these intent tags have low reliability. Then, the server will merge similar intent tags from the remaining tags to avoid duplicate or redundant intent outputs. Finally, the processed intent tags will be output as supply chain feedback intent mining results, which can clearly reflect the supply chain feedback intent expressed in the target supply chain settlement feedback text, thereby helping companies quickly understand and respond to problems in the supply chain.

[0140] In a possible implementation, step S180 includes:

[0141] Step S181, establishing a mapping table, through which the supply chain feedback intention is associated with the corresponding information push content.

[0142] In this embodiment, the server will first establish a mapping table, which is used to associate specific supply chain feedback intents with corresponding information push content. For example, if the feedback intent is "delivery delay", the mapping table can correspond this intent to a series of suggestions for solving delivery delay problems, relevant supply chain optimization articles or case studies, etc. Similarly, if the feedback intent is "product quality issues", it will be associated with push content related to quality management, return procedures, etc.

[0143] Step S182, analyzing the historical behavior pattern data of the target user corresponding to the supply chain feedback intention mining result, and generating a personalized push strategy for the target user.

[0144] Next, the server will deeply analyze the historical behavior pattern data of the target users corresponding to the supply chain feedback intention mining results. This data may include the user's browsing history, click behavior, search history, etc., so as to understand the user's preferences, needs and habits. Based on this data, the server generates a personalized push strategy for the target user. For example, if the server finds that a user often pays attention to content related to delivery time, then when encountering feedback intentions about delivery delays, the server can prioritize pushing strategies or suggestions on how to optimize logistics distribution and reduce delivery time.

[0145] Step S183, based on the supply chain feedback intention mining results, select candidate push content information that matches the current supply chain feedback intention of the target user from the mapping table, and extract target push content information from the candidate push content information based on the personalized push strategy for push.

[0146] When the server receives the supply chain feedback intention mining results, it will filter out candidate push content information that matches the target user's current supply chain feedback intention from the mapping table based on this result. Then, based on the personalized push strategy generated previously, the server will extract the target push content information that best suits the target user from these candidate push content information for push. For example, if the user's personalized push strategy indicates that they prefer to receive practical operation guides, then when encountering the feedback intention of "product quality problem", the server can push a detailed operation guide on how to deal with product quality problems instead of a theoretical analysis article.

[0147] Therefore, this process ensures the pertinence and effectiveness of information push and improves the efficiency of users receiving and processing supply chain feedback.

[0148] Figure 2 The hardware structure of the supply chain settlement feedback big data analysis system 100 based on artificial intelligence for implementing the supply chain settlement feedback big data analysis method based on artificial intelligence provided in the embodiment of the present application is shown as follows: Figure 2 As shown, the artificial intelligence-based supply chain settlement feedback big data analysis system 100 may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0149] In one possible design, the supply chain settlement feedback big data analysis system 100 based on artificial intelligence can be a single server or a server group. The server group can be centralized or distributed (for example, the supply chain settlement feedback big data analysis system 100 based on artificial intelligence can be a distributed system). In some embodiments, the supply chain settlement feedback big data analysis system 100 based on artificial intelligence can be local or remote. For example, the supply chain settlement feedback big data analysis system 100 based on artificial intelligence can access information and / or data stored in the machine-readable storage medium 120 via a network. For another example, the supply chain settlement feedback big data analysis system 100 based on artificial intelligence can be directly connected to the machine-readable storage medium 120 to access the stored information and / or data. In some embodiments, the supply chain settlement feedback big data analysis system 100 based on artificial intelligence can be implemented on the supply chain settlement feedback big data analysis system based on artificial intelligence. As an example only, the supply chain settlement feedback big data analysis system based on artificial intelligence can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any combination thereof.

[0150] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the artificial intelligence-based supply chain settlement feedback big data analysis system 100 uses to execute or use to complete the exemplary method described in this application.

[0151] During the specific implementation process, at least one processor 110 executes computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the supply chain settlement feedback big data analysis method based on artificial intelligence as described in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0152] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned artificial intelligence-based supply chain settlement feedback big data analysis system 100. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0153] In addition, an embodiment of the present application also provides a readable storage medium, in which computer executable instructions are preset. When the processor executes the computer executable instructions, the above-mentioned supply chain settlement feedback big data analysis method based on artificial intelligence is implemented.

[0154] It should be noted that in order to simplify the description disclosed in this application and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof. Similarly, it should be noted that in order to simplify the description disclosed in this application and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof.

Claims

1. A supply chain settlement feedback big data analysis method based on artificial intelligence, characterized in that: The method comprises: Acquire a sample supply chain settlement feedback text, and acquire a plurality of focus attention vector sequences of the sample supply chain settlement feedback text, each focus attention vector sequence representing a conversation text understanding result expressed by each text attention unit in the sample supply chain settlement feedback text; Performing panoramic semantic vector representation on the example supply chain settlement feedback text through a supply chain text mining network to generate a target panoramic semantic vector sequence of the example supply chain settlement feedback text; Through the supply chain text mining network, a panoramic semantic vector is estimated for the example supply chain settlement feedback text according to the multiple focused attention vector sequences, and an estimated panoramic semantic vector sequence of the example supply chain settlement feedback text is generated, wherein the estimated panoramic semantic vector sequence at least represents a conversation text understanding result expressed by multiple text attention units in the example supply chain settlement feedback text; Based on a training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence, training the supply chain text mining network to generate a trained supply chain text mining network; The supply chain text mining network includes a teaching text mining network and a taught text mining network, and the teaching text mining network is generated by optimizing parameter configuration information in the taught text mining network; The step of performing panoramic semantic vector representation on the example supply chain settlement feedback text through a supply chain text mining network to generate a target panoramic semantic vector sequence of the example supply chain settlement feedback text includes: Through the teaching text mining network, the example supply chain settlement feedback text is represented by a panoramic semantic vector to generate a teaching panoramic semantic vector sequence of the example supply chain settlement feedback text; Performing a panoramic semantic vector representation on the example supply chain settlement feedback text through the taught text mining network to generate a taught panoramic semantic vector sequence of the example supply chain settlement feedback text; The step of training the supply chain text mining network based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence to generate a trained supply chain text mining network includes: Based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence, and the feature distance between the teaching panoramic semantic vector sequence and the taught panoramic semantic vector sequence, the taught text mining network is trained to generate the trained supply chain text mining network; The supply chain text mining network includes a teaching text mining network and a taught text mining network, and the teaching text mining network is generated by optimizing parameter configuration information in the taught text mining network; The step of obtaining a plurality of focus vector sequences of the example supply chain settlement feedback text includes: Performing a focused attention vector representation on the example supply chain settlement feedback text through the teaching text mining network to generate X teaching focused attention vector sequences, and performing a focused attention vector representation on the example supply chain settlement feedback text through the taught text mining network to generate X taught focused attention vector sequences; Walking the X taught focus attention vector sequences, and for the currently walked target taught focus attention vector sequence, extracting a target taught focus attention vector sequence having the smallest feature distance with the target taught focus attention vector sequence from the X taught focus attention vector sequences; configuring a focus attention vector sequence combination including the target teaching focus attention vector sequence and the target taught focus attention vector sequence until the walk is completed, generating X focus attention vector sequence combinations; Based on the semantic relevance corresponding to each of the X focused attention vector sequence combinations, determining a text mining distinction parameter between the taught text mining network and the teaching text mining network; the text mining distinction parameter is negatively correlated with the semantic relevance corresponding to each of the focused attention vector sequence combinations; The step of training the supply chain text mining network based on the training goal of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence to generate a trained supply chain text mining network includes: Based on the training objectives of minimizing the feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence, and minimizing the text mining distinction parameters between the taught text mining network and the teaching text mining network, the taught text mining network is trained to generate the trained supply chain text mining network.

2. The supply chain settlement feedback big data analysis method based on artificial intelligence according to claim 1 is characterized in that: The step of performing panoramic semantic vector representation on the example supply chain settlement feedback text through a supply chain text mining network to generate a target panoramic semantic vector sequence of the example supply chain settlement feedback text includes: Performing attention processing on each of the multiple focused attention vector sequences to generate an attention vector sequence corresponding to the corresponding focused attention vector sequence; The attention vector sequences corresponding to the multiple focused attention vector sequences are integrated to generate a target panoramic semantic vector sequence of the example supply chain settlement feedback text.

3. The supply chain settlement feedback big data analysis method based on artificial intelligence according to claim 1 is characterized in that: The step of estimating the panoramic semantic vector of the example supply chain settlement feedback text according to the multiple focus attention vector sequences through the supply chain text mining network to generate an estimated panoramic semantic vector sequence of the example supply chain settlement feedback text includes: The multiple focused attention vector sequences are fused to generate an estimated panoramic semantic vector sequence of the example supply chain settlement feedback text.

4. The supply chain settlement feedback big data analysis method based on artificial intelligence according to claim 1 is characterized in that: The step of estimating the panoramic semantic vector of the example supply chain settlement feedback text according to the multiple focus attention vector sequences through the supply chain text mining network to generate an estimated panoramic semantic vector sequence of the example supply chain settlement feedback text includes: Through the teaching text mining network, a panoramic semantic vector is estimated for the example supply chain settlement feedback text according to the multiple focus attention vector sequences to generate a teaching estimated panoramic semantic vector sequence of the example supply chain settlement feedback text; By means of the taught text mining network, a panoramic semantic vector is estimated for the example supply chain settlement feedback text according to the multiple focused attention vector sequences, so as to generate a taught estimated panoramic semantic vector sequence of the example supply chain settlement feedback text; The step of training the taught text mining network based on a training objective of minimizing a feature distance between the estimated panoramic semantic vector sequence and the target panoramic semantic vector sequence comprises: Based on the training objectives of minimizing the feature distance between the teaching panoramic semantic vector sequence and the teaching estimated panoramic semantic vector sequence, and minimizing the feature distance between the taught panoramic semantic vector sequence and the taught estimated panoramic semantic vector sequence, the taught text mining network is trained to generate the trained supply chain text mining network.

5. The supply chain settlement feedback big data analysis method based on artificial intelligence according to claim 1 is characterized in that: The method further comprises: Based on a text mining instruction for a target supply chain settlement feedback text in the supply chain settlement feedback big data, obtaining the target supply chain settlement feedback text; Performing panoramic semantic vector representation on the target supply chain settlement feedback text through the trained supply chain text mining network to generate a target panoramic semantic vector sequence of the target supply chain settlement feedback text; According to the target panoramic semantic vector sequence of the target supply chain settlement feedback text, supply chain feedback intention mining is performed on the target supply chain settlement feedback text to generate a supply chain feedback intention mining result; Corresponding information push is performed according to the supply chain feedback intention mining results.

6. The supply chain settlement feedback big data analysis method based on artificial intelligence according to claim 5 is characterized in that: The step of performing supply chain feedback intention mining on the target supply chain settlement feedback text based on the target panoramic semantic vector sequence of the target supply chain settlement feedback text to generate a supply chain feedback intention mining result comprises: Pre-building an intention classification model based on deep learning, wherein the intention classification model is used to identify supply chain feedback intentions based on a target panoramic semantic vector sequence of the target supply chain settlement feedback text; Using the existing labeled data set to train the intent classification model to obtain a trained intent classification model; Inputting the target panoramic semantic vector sequence into the trained intent classification model, extracting key semantic features in the target panoramic semantic vector sequence through the intent classification model, and mapping them to predefined intent categories to output one or more intent labels and the confidence corresponding to each intent label, wherein the intent label represents the supply chain feedback intent expressed in the target supply chain settlement feedback text; The intention tags with confidence less than the preset confidence are eliminated, and the similar intention tags in the remaining intention tags are merged, and the processed intention tags are output as the supply chain feedback intention mining results.

7. The supply chain settlement feedback big data analysis method based on artificial intelligence according to claim 5 is characterized in that: The step of executing corresponding information push according to the supply chain feedback intention mining result includes: Establishing a mapping table, through which the supply chain feedback intention is associated with the corresponding information push content; Analyze the historical behavior pattern data of the target user corresponding to the supply chain feedback intention mining result, and generate a personalized push strategy for the target user; According to the supply chain feedback intention mining results, candidate push content information matching the current supply chain feedback intention of the target user is screened out from the mapping table, and target push content information is extracted from the candidate push content information based on the personalized push strategy for push.

8. A supply chain settlement feedback big data analysis system based on artificial intelligence, characterized in that: The artificial intelligence-based supply chain settlement feedback big data analysis system 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 artificial intelligence-based supply chain settlement feedback big data analysis method described in any one of claims 1 to 7.

Citation Information

Patent Citations

  • Processing method and system for power grid user service evaluation

    CN112541077A

  • Method and system for quickly constructing industry question and answer knowledge base

    CN117290489A

  • Natural language processing method and system based on machine learning

    CN117493491A

  • Digital enterprise service data mining method and system based on artificial intelligence

    CN117591662A

  • B2B shop intelligent searching and recommending method based on WeChat applet

    CN117763233A