Intelligent decision generation system based on user prior information guidance

Abstract

The invention relates to the field of intelligent decision-making, and particularly discloses an intelligent decision-making generation system based on user prior information guidance, which is characterized in that historical decision-making case data and to-be-audited decision-making contents of a target object are acquired, each decision-making case data in the historical decision-making case data comprises a case content and a decision-making label, and the case content and the decision-making label are acquired; semantic coding and global semantic aggregation are carried out on the historical decision case data by adopting a data processing and analysis technology based on artificial intelligence, and semantic understanding is carried out on the to-be-audited decision content; and intelligently obtaining an auxiliary decision-making result of the decision-making type label according to the semantic association matching feature between the historical decision-making case data and the to-be-audited decision-making content. Through the mode, the system can better utilize past decision-making experience to guide the current decision-making process, so that more accurate decision-making suggestions are provided, and the decision-making quality and efficiency are improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent decision-making, and more specifically, to an intelligent decision-making generation system guided by user prior information. Background Art

[0002] As problems become more complex, there is an increasing need for advanced decision support tools to handle multi-variable, multi-objective, and multi-constrained decision problems. Intelligent decision-making systems are artificial intelligence systems that can simulate the human decision-making process. These systems analyze large amounts of data, identify patterns and relationships, and then use this information to generate decisions or predictions. They play an important role in improving decision quality, accelerating decision-making, and processing large amounts of complex data.

[0003] However, the existing decision-making systems lack the support of practical experience, such as the lack of intelligent decision-making based on historical data (users' historical information), which can easily lead to inefficient decision-making processes and cause decisions to deviate from actual conditions, thereby affecting the reliability and accuracy of the decision-making process.

[0004] Therefore, an intelligent decision-making system guided by user prior information is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an intelligent decision generation system guided by user prior information, which obtains the historical decision case data and the decision content to be reviewed of the target object, wherein each decision case data in the historical decision case data contains case content and decision labels, and uses artificial intelligence-based data processing and analysis technology to semantically encode and globally semantically aggregate the historical decision case data, and semantically understand the decision content to be reviewed, so as to intelligently obtain the auxiliary decision results of the decision type label according to the semantic association matching features between the historical decision case data and the decision content to be reviewed. In this way, the system can better utilize past decision-making experience to guide the current decision-making process, thereby providing more accurate decision-making suggestions and improving the quality and efficiency of decision-making.

[0006] According to one aspect of the present application, there is provided an intelligent decision making system guided by user prior information, which includes:

[0007] A historical decision case data acquisition module is used to acquire historical decision case data of a target object, wherein each decision case data in the historical decision case data includes case content and decision labels;

[0008] A case data semantic coding module, used to semantically code each decision case data in the historical decision case data to obtain a set of decision case data semantic coding vectors;

[0009] A case-by-case semantic feature extraction module is used to extract the associated semantic features between decision cases after arranging the set of semantic encoding vectors of the decision case data into a global semantic matrix of historical decision case data to obtain a local semantic dimension associated representation vector between decision cases;

[0010] A semantic aggregation module, used for inputting the set of semantic encoding vectors of the decision case data into a node message interaction aggregation network integrating node importance to obtain a semantic aggregation representation vector of sample dimensions between decision cases;

[0011] A multi-scale semantic fusion module, used to fuse the local semantic dimension association representation vector between the decision cases and the sample dimension semantic aggregation representation vector between the decision cases to obtain a multi-dimensional and multi-scale semantic aggregation representation vector between the decision cases;

[0012] A module for obtaining decision content to be reviewed, used to obtain decision content to be reviewed;

[0013] A semantic coding module for the decision content to be reviewed, used for semantically coding the decision content to be reviewed to obtain a semantic coding vector for the decision content to be reviewed;

[0014] A semantic matching association module, used to use the semantic encoding vector of the decision content to be reviewed as a query vector, calculate the query result relative to the multi-dimensional and multi-scale semantic aggregation representation vector between the decision cases to obtain a semantic matching association matrix as a semantic matching association feature;

[0015] The result generation module is used to obtain auxiliary decision results based on the semantic matching association features.

[0016] In the above-mentioned intelligent decision-making system guided by user prior information, the case-to-case semantic feature extraction module is used to: arrange the set of semantic encoding vectors of the decision case data into a global semantic matrix of historical decision case data and then pass it through the decision case-to-decision case correlation semantic feature extractor based on the text convolutional neural network model to obtain the local semantic dimension correlation representation vector between the decision cases.

[0017] In the above-mentioned intelligent decision-making system guided by user prior information, the semantic aggregation module includes: a significant factor calculation unit, which is used to use a scoring function to measure the significant factors of the first k-1 decision case data semantic coding vectors in the set of the decision case data semantic coding vectors to obtain k-1 decision case data semantic significant factors; a feature weighted aggregation unit, which is used to use the k-1 decision case data semantic significant factors as weights to calculate the position-weighted sum of the first k-1 decision case data semantic coding vectors in the set of the decision case data semantic coding vectors to aggregate the first k-1 decision case data semantic coding vectors in the set of the decision case data semantic coding vectors to obtain a rear node set. A semantic aggregate representation vector of decision case data; a feature modulation unit, used to use preset hyperparameters to linearly modulate the kth decision case data semantic coding vector in the set of decision case data semantic coding vectors to obtain the kth decision case data semantic coding vector after modulation; a full-node aggregation unit, used to calculate the positional sum of the kth decision case data semantic coding vector after modulation and the semantic aggregate representation vector of the decision case data of the rear node set to obtain the semantic full-node aggregation representation vector of the decision case data; and a multi-layer perceptron processing unit, used to input the semantic full-node aggregation representation vector of the decision case data semantics into a multi-layer perceptron model to obtain the semantic aggregation representation vector of the sample dimension between the decision cases.

[0018] In the above-mentioned intelligent decision-making system guided by user prior information, the significant factor calculation unit includes: a feature addition secondary sub-unit, which is used to respectively calculate the matrix product between each weight coefficient matrix and the first k-1 decision case data semantic coding vectors to obtain the first k-1 decision case data semantic coding weighted feature vectors; a feature bias calculation secondary sub-unit, which is used to add the corresponding bias vector to each decision case data semantic coding weighted feature vector in the first k-1 decision case data semantic coding weighted feature vectors to obtain the first k-1 decision case data semantic coding bias feature vectors; a feature activation secondary sub-unit, which is used to add the first k-1 decision case data semantic coding weighted feature vectors to the first k-1 decision case data semantic coding weighted feature vectors; The semantic coding bias feature vectors of the first k-1 decision case data are input into a sigmoid function for activation to obtain the semantic coding activation feature vectors of the first k-1 decision case data; a semantic relevance calculation secondary subunit is used to calculate the product between each decision case data semantic coding activation feature vector in the first k-1 decision case data semantic coding activation feature vector and a weight coefficient vector to obtain the semantic relevance of the first k-1 decision case data; and a semantic saliency factor calculation secondary subunit is used to perform nonlinear processing on the semantic relevance of the first k-1 decision case data to obtain the semantic saliency factor of the k-1 decision case data.

[0019] In the above-mentioned intelligent decision-making system guided by user prior information, the semantic saliency factor calculation secondary sub-unit is used to: input the semantic associations of the first k-1 decision case data into the softmax function to obtain the semantic saliency factors of the k-1 decision case data.

[0020] In the above-mentioned intelligent decision-making system guided by user prior information, the semantic matching association module is used to: use the semantic encoding vector of the decision content to be reviewed as the query vector, calculate the vector product between the query vector and the transposed vector of the multi-dimensional and multi-scale semantic aggregation representation vector between the decision cases to obtain the semantic matching association matrix.

[0021] In the above-mentioned intelligent decision generation system guided by user prior information, the result generation module is used to: pass the semantic matching association matrix through the classifier-based intelligent decision generation module to obtain an auxiliary decision result, and the auxiliary decision result is a decision type label.

[0022] In the above-mentioned intelligent decision-making system guided by user prior information, it also includes a training module for training the decision case correlation semantic feature extractor based on the text convolutional neural network model, the node message interaction aggregation network for fusion node importance, and the classifier-based intelligent decision-making module.

[0023] In the above-mentioned intelligent decision-making system guided by user prior information, the training module includes: a training data acquisition unit, which is used to acquire training data, and the training data includes historical training decision case data of the target object and training decision content to be reviewed, wherein each training decision case data in the historical training decision case data contains training case content and training decision labels, as well as the real label of the decision type; a historical training decision case semantic encoding unit, which is used to semantically encode each training decision case data in the historical training decision case data to obtain a set of training decision case data semantic encoding vectors; a local semantic extraction unit between training decision cases, which is used to arrange the set of training decision case data semantic encoding vectors into a global semantic matrix of training historical decision case data and then pass it through the decision case correlation semantic feature extractor based on the text convolutional neural network model to obtain a local semantic dimension correlation representation vector between training decision cases; a semantic aggregation unit between training decision cases, which is used to input the set of training decision case data semantic encoding vectors into the node message interaction aggregation network of the fusion node importance to obtain a training decision case local semantic dimension association representation vector. A sample dimension semantic aggregation representation vector; a training multi-scale semantic aggregation unit, used to fuse the local semantic dimension association representation vector between the training decision cases and the sample dimension semantic aggregation representation vector between the training decision cases to obtain a multi-dimensional multi-scale semantic aggregation representation vector between the training decision cases; a training decision content semantic encoding unit, used to semantically encode the training decision content to be reviewed to obtain a training decision content semantic encoding vector; a training semantic matching unit, used to use the training decision content semantic encoding vector as a query vector, calculate the query result relative to the multi-dimensional multi-scale semantic aggregation representation vector between the training decision cases to obtain a training semantic matching association matrix; a classification loss calculation unit, used to input the training semantic matching association matrix into the classifier-based intelligent decision generation module to obtain a classification loss function value; and a training unit, used to train the decision case association semantic feature extractor based on the text convolutional neural network model, the node message interaction aggregation network fused with node importance, and the classifier-based intelligent decision generation module based on the classification loss function value and through gradient descent back propagation.

[0024] Compared with the prior art, the present application provides an intelligent decision-making system guided by user prior information, which obtains the historical decision case data and the decision content to be reviewed of the target object, wherein each decision case data in the historical decision case data contains case content and decision labels, and uses artificial intelligence-based data processing and analysis technology to semantically encode and globally semantically aggregate the historical decision case data, and semantically understand the decision content to be reviewed, so as to intelligently obtain the auxiliary decision results of the decision type label based on the semantic association matching features between the historical decision case data and the decision content to be reviewed. In this way, the system can better utilize past decision-making experience to guide the current decision-making process, thereby providing more accurate decision-making suggestions and improving the quality and efficiency of decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0026] Figure 1 The block diagram is a system for making intelligent decisions based on user prior information according to an embodiment of the present application.

[0027] Figure 2 Schematic diagram of the architecture of an intelligent decision-making system guided by user prior information according to an embodiment of the present application.

[0028] Figure 3 It is a block diagram of a semantic aggregation module in an intelligent decision-making generation system guided by user prior information according to an embodiment of the present application.

[0029] Figure 4 The block diagram is a significant factor calculation unit in the intelligent decision making system guided by user prior information according to an embodiment of the present application.

[0030] Figure 5 The block diagram is a training module in the intelligent decision-making generation system guided by user prior information according to an embodiment of the present application. DETAILED DESCRIPTION

[0031] The embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, but rather these embodiments are provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0032] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0033] In the description of the embodiments of the present disclosure, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0034] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0035] As problems become more complex, there is an increasing need for advanced decision support tools to handle multi-variable, multi-objective, and multi-constrained decision problems. Intelligent decision-making systems are artificial intelligence systems that can simulate the human decision-making process. These systems analyze large amounts of data, identify patterns and relationships, and then use this information to generate decisions or predictions. They play an important role in improving decision quality, accelerating decision-making, and processing large amounts of complex data.

[0036] However, the existing decision-making systems lack the support of practical experience, such as the lack of intelligent decision-making based on historical data (users' historical information), which can easily lead to inefficient decision-making processes and cause decisions to deviate from actual conditions, thereby affecting the reliability and accuracy of the decision-making process.

[0037] It should be understandable that corporate management has to approve and decide on a large number of matters every day. Most of these matters are repetitive or highly similar, and the relevant personnel have made a large number of decision-making precedents. Therefore, intelligent decision generation can be based on historical data (that is, user prior information) to improve office efficiency and office intelligence.

[0038] Based on this, the present application proposes an intelligent decision-making system guided by user prior information, which obtains the historical decision case data and the decision content to be reviewed of the target object, wherein each decision case data in the historical decision case data contains case content and decision labels, and uses artificial intelligence-based data processing and analysis technology to semantically encode and globally semantically aggregate the historical decision case data, and semantically understand the decision content to be reviewed, so as to intelligently obtain auxiliary decision results of decision type labels based on the semantic association matching features between the historical decision case data and the decision content to be reviewed. In this way, the system can better utilize past decision-making experience to guide the current decision-making process, thereby providing more accurate decision-making suggestions and improving the quality and efficiency of decision-making.

[0039] Figure 1 The block diagram is a system for making intelligent decisions based on user prior information according to an embodiment of the present application. Figure 2 FIG. 1 is a schematic diagram of the architecture of an intelligent decision-making system based on user prior information guidance according to an embodiment of the present application. Figure 1 and Figure 2As shown, according to an embodiment of the present application, an intelligent decision-making system 100 guided by user prior information includes: a historical decision case data acquisition module 110, which is used to acquire historical decision case data of a target object, wherein each decision case data in the historical decision case data includes case content and a decision label; a case data semantic encoding module 120, which is used to semantically encode each decision case data in the historical decision case data to obtain a set of decision case data semantic encoding vectors; an inter-case semantic feature extraction module 130, which is used to extract the associated semantic features between decision cases after arranging the set of decision case data semantic encoding vectors into a global semantic matrix of historical decision case data to obtain a local semantic dimension associated representation vector between decision cases; a semantic aggregation module 140, which is used to input the set of decision case data semantic encoding vectors into a node message interaction aggregation module that fuses the importance of a node. A network is used to obtain a sample dimension semantic aggregation representation vector between decision cases; a multi-scale semantic fusion module 150 is used to fuse the local semantic dimension association representation vector between the decision cases and the sample dimension semantic aggregation representation vector between the decision cases to obtain a multi-dimensional and multi-scale semantic aggregation representation vector between the decision cases; a decision content acquisition module 160 is used to obtain the decision content to be reviewed; a decision content semantic encoding module 170 is used to semantically encode the decision content to be reviewed to obtain a semantic encoding vector of the decision content to be reviewed; a semantic matching association module 180 is used to use the semantic encoding vector of the decision content to be reviewed as a query vector, and calculate the query result relative to the multi-dimensional and multi-scale semantic aggregation representation vector between the decision cases to obtain a semantic matching association matrix as a semantic matching association feature; and a result generation module 190 is used to obtain an auxiliary decision result based on the semantic matching association feature.

[0040] In an embodiment of the present application, the historical decision case data acquisition module 110 is used to acquire the historical decision case data of the target object, wherein each decision case data in the historical decision case data includes case content and decision labels. It should be understood that considering that the historical decision case data refers to a collection of cases that have occurred and made decisions in the past, these data usually include case content and corresponding decision labels. Specifically, the case content describes the decision context, relevant background information, issues involved and possible options in detail; the decision label is a mark of the decision result made on the case content, which can be a classification label (such as "approval", "rejection") or a continuous value (such as score, grade). Based on this, in the technical solution of the present application, the historical decision case data of the target object is acquired, wherein each decision case data in the historical decision case data includes case content and decision labels, and semantic analysis and understanding are performed on it. The system can learn decision patterns and trends, help identify potential risks and decision consequences, and better evaluate the decision content to be reviewed.

[0041] In an embodiment of the present application, the case data semantic encoding module 120 is used to semantically encode each decision case data in the historical decision case data to obtain a set of decision case data semantic encoding vectors. Accordingly, it is considered that each decision case data in the historical decision case data contains the semantic association between the context information of the decision cases. Therefore, in order to capture and extract the semantic feature information in each decision case data, in the technical solution of the present application, each decision case data in the historical decision case data is semantically encoded to mine and capture the semantic dependencies between the decision case data in each decision case data, thereby obtaining a set of decision case data semantic encoding vectors containing rich contextual semantic information.

[0042] In an embodiment of the present application, the case-to-case semantic feature extraction module 130 is used to extract the associated semantic features between decision cases after arranging the set of semantic coding vectors of the decision case data into a global semantic matrix of historical decision case data to obtain the local semantic dimension associated representation vector between decision cases. Specifically, in an embodiment of the present application, the case-to-case semantic feature extraction module is used to: after arranging the set of semantic coding vectors of the decision case data into a global semantic matrix of historical decision case data, obtain the local semantic dimension associated representation vector between decision cases through an inter-decision case associated semantic feature extractor based on a text convolutional neural network model. It should be understood that, considering that there is an associated relationship between the set of semantic coding vectors of the decision case data based on the global semantic information of the entire historical decision case data, therefore, in order to be able to more comprehensively and comprehensively analyze the semantic information of the entire historical decision case, so as to more comprehensively capture the semantic association between the historical decision case data, in the technical solution of the present application, the set of semantic coding vectors of the decision case data is arranged into a global semantic matrix of historical decision case data. Next, considering that there is local implicit semantic feature association information in the global semantic matrix of the historical decision case data, and the text convolutional neural network has good feature extraction capabilities in text processing, it can effectively capture the local features and semantic information of text data. Furthermore, in the technical solution of the present application, the global semantic matrix of the historical decision case data is passed through a decision case association semantic feature extractor based on a text convolutional neural network model to capture the association between local semantics in the global semantic matrix of the historical decision case data, thereby obtaining a local semantic dimension association representation vector between decision cases.

[0043] In an embodiment of the present application, the semantic aggregation module 140 is used to input the set of decision case data semantic coding vectors into a node message interaction aggregation network that integrates node importance to obtain a semantic aggregation representation vector of sample dimensions between decision cases. Accordingly, it is considered that the decision case data semantic coding vectors have semantic influence and dependency on the entire historical decision case data, and the different decision case data semantic coding features have different influence and importance in the entire historical decision data. Based on this, in order to consider the importance of each decision case data semantic coding vector node, the network pays more attention to the important decision case data semantic coding node feature information, so as to help the network better capture the important key semantic information in the set of decision case data semantic coding vectors. In the technical solution of the present application, the set of decision case data semantic coding vectors is input into a node message interaction aggregation network that integrates node importance to obtain a semantic aggregation representation vector of sample dimensions between decision cases. Specifically, since the different decision case data semantic coding vectors have different importance in the entire decision case data, the node message interaction aggregation network that fused the node importance can adaptively adjust according to the contribution degree of different decision case data semantic coding nodes in the set of decision case data semantic coding vectors, propagate the feature information of different nodes in the network, and realize information transmission and interaction between different decision case data semantic coding nodes, and effectively integrate and fuse the semantic information of different features, so as to generate a more comprehensive and rich semantic aggregation representation vector of sample dimensions between decision cases, thereby improving the model's understanding and representation capabilities of decision case data, and further providing more accurate and comprehensive information support for the decision generation system.

[0044] Figure 3 is a block diagram of a semantic aggregation module in an intelligent decision-making system based on user prior information guidance according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 3As shown, the semantic aggregation module 140 includes: a significant factor calculation unit 141, which is used to use a scoring function to measure the significant factors of the first k-1 decision case data semantic coding vectors in the set of the decision case data semantic coding vectors to obtain k-1 decision case data semantic significant factors; a feature weighted aggregation unit 142, which is used to use the k-1 decision case data semantic significant factors as weights to calculate the position-weighted sum of the first k-1 decision case data semantic coding vectors in the set of the decision case data semantic coding vectors to aggregate the first k-1 decision case data semantic coding vectors in the set of the decision case data semantic coding vectors to obtain the decision case data semantic aggregation of the rear node set. A representation vector; a feature modulation unit 143, used to use a preset hyperparameter to linearly modulate the kth decision case data semantic coding vector in the set of decision case data semantic coding vectors to obtain the kth decision case data semantic coding vector after modulation; a full-node aggregation unit 144, used to calculate the positional sum of the kth decision case data semantic coding vector after modulation and the decision case data semantic aggregation representation vector of the rear node set to obtain the decision case data semantic full-node aggregation representation vector; and a multi-layer perceptron processing unit 145, used to input the decision case data semantic full-node aggregation representation vector into a multi-layer perceptron model to obtain the sample dimension semantic aggregation representation vector between the decision cases

[0045] Figure 4 is a block diagram of a significant factor calculation unit in an intelligent decision-making system based on user prior information guidance according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 4As shown, the significant factor calculation unit 141 includes: a feature addition secondary subunit 1411, which is used to respectively calculate the matrix product between each weight coefficient matrix and the first k-1 decision case data semantic coding vectors to obtain the first k-1 decision case data semantic coding weighted feature vectors; a feature bias calculation secondary subunit 1412, which is used to add the corresponding bias vector to each decision case data semantic coding weighted feature vector in the first k-1 decision case data semantic coding weighted feature vectors to obtain the first k-1 decision case data semantic coding bias feature vectors; a feature activation secondary subunit 1413, which is used to add the first k-1 decision case data semantic coding weighted feature vectors to the first k-1 decision case data semantic coding weighted feature vectors. The case data semantic coding bias feature vector is input into a sigmoid function for activation to obtain the semantic coding activation feature vectors of the first k-1 decision case data; a semantic relevance calculation secondary subunit 1414 is used to calculate the product between each decision case data semantic coding activation feature vector in the first k-1 decision case data semantic coding activation feature vector and a weight coefficient vector to obtain the semantic relevance of the first k-1 decision case data; and a semantic significance factor calculation secondary subunit 1415 is used to perform nonlinear processing on the semantic relevance of the first k-1 decision case data to obtain the semantic significance factor of the k-1 decision case data.

[0046] More specifically, in an embodiment of the present application, the semantic saliency factor calculation secondary subunit is used to: input the semantic associations of the first k-1 decision case data into a softmax function to obtain the semantic saliency factors of the k-1 decision case data.

[0047] In the embodiment of the present application, specifically, the semantic aggregation module is used to: use the node message interaction aggregation network of the fusion node importance to process the set of semantic encoding vectors of the decision case data with the following aggregation formula to obtain the semantic aggregation representation vector of the sample dimension between the decision cases; wherein the aggregation formula is:

[0048]

[0049] Among them, h i represents the i-th decision case data semantic encoding vector in the set of decision case data semantic encoding vectors, W i represents the weight coefficient matrix, B i is the bias vector, represents the weight coefficient vector, sigmoid(·) represents the sigmoid function, softmax(·) represents the softmax function, e(h i) represents the semantic saliency factor of the decision case data at each position in the sequence of the semantic saliency factor of the decision case data, k is the number of feature vectors in the set of the semantic encoding vector of the decision case data, h k represents the kth decision case data semantic encoding vector in the set of decision case data semantic encoding vectors, ∈ is the preset hyperparameter, MLP(·) represents a multi-layer perceptron, h c is a semantic aggregation representation vector of sample dimensions between the decision cases.

[0050] In an embodiment of the present application, the multi-scale semantic fusion module 150 is used to fuse the local semantic dimension association representation vector between the decision cases and the sample dimension semantic aggregation representation vector between the decision cases to obtain a multi-dimensional multi-scale semantic aggregation representation vector between the decision cases. It should be understood that, considering that the local semantic dimension association representation vector between the decision cases and the sample dimension semantic aggregation representation vector between the decision cases respectively capture the local semantic association information and global sample dimension semantic aggregation information of the decision case data. Therefore, in the technical solution of the present application, the local semantic dimension association representation vector between the decision cases and the sample dimension semantic aggregation representation vector between the decision cases are fused to comprehensively utilize the local detail semantic correlation and overall semantic information of the decision cases to obtain a multi-dimensional multi-scale semantic aggregation representation vector between decision cases with more comprehensive and rich semantic information.

[0051] In the embodiment of the present application, the decision content acquisition module 160 is used to obtain the decision content to be reviewed. Accordingly, the decision content to be reviewed refers to the cases that currently need to make decisions. These cases have not yet had decision results and need to be evaluated and decided. Therefore, in the technical solution of the present application, obtaining the decision content to be reviewed and combining it with historical decision case data can more accurately analyze the decision content with review, which helps to improve the speed, accuracy and reliability of decision-making.

[0052] In an embodiment of the present application, the decision content to be reviewed semantic coding module 170 is used to semantically encode the decision content to be reviewed to obtain a semantic coding vector of the decision content to be reviewed. It should be understood that, considering that the decision content to be reviewed also contains a large amount of decision content semantic information, there are mutual correlations and dependencies between these semantic information. Based on this, in the technical solution of the present application, the decision content to be reviewed is semantically encoded to capture and extract the semantic correlation between each semantic in the decision content to be reviewed, thereby obtaining a semantic coding vector of the decision content to be reviewed.

[0053] In an embodiment of the present application, the semantic matching association module 180 is used to use the semantic encoding vector of the decision content to be reviewed as a query vector, and calculate the query result relative to the multi-dimensional multi-scale semantic aggregation representation vector between the decision cases to obtain a semantic matching association matrix as a semantic matching association feature. Specifically, in an embodiment of the present application, the semantic matching association module is used to: use the semantic encoding vector of the decision content to be reviewed as a query vector, and calculate the vector product between the query vector and the transposed vector of the multi-dimensional multi-scale semantic aggregation representation vector between the decision cases to obtain the semantic matching association matrix. Accordingly, in order to more accurately determine the decision type of the decision content to be reviewed, in the technical solution of the present application, the semantic encoding vector of the decision content to be reviewed is used as a query vector, and the query result relative to the multi-dimensional multi-scale semantic aggregation representation vector between the decision cases is calculated to obtain a semantic matching association matrix. In particular, in a specific embodiment of the present application, the semantic encoding vector of the decision content to be reviewed is used as a query vector, and the vector product between the query vector and the transposed vector of the multi-dimensional multi-scale semantic aggregation representation vector between the decision cases is calculated to obtain the semantic matching association matrix. It should be understandable that by calculating the semantic matching association matrix between the two, the semantic similarity or matching degree between the decision content to be reviewed and each historical decision case can be quantified, so as to find the decision case most relevant to the content to be reviewed, and provide recommendations or suggestions to decision makers, helping them to find previous decision cases related to the current review content more quickly for reference or reference, thereby optimizing the decision-making process and improving decision-making efficiency.

[0054] In an embodiment of the present application, the result generation module 190 is used to obtain auxiliary decision results based on the semantic matching association features. Specifically, in an embodiment of the present application, the result generation module is used to: pass the semantic matching association matrix through a classifier-based intelligent decision generation module to obtain auxiliary decision results, and the auxiliary decision results are decision type labels. That is, the semantic matching association features calculated by using the semantic matching between the semantic coding vector of the decision content to be reviewed and the multi-dimensional and multi-scale semantic aggregation representation vector between the decision cases are classified and processed to intelligently obtain auxiliary decision results of decision type labels. In this way, the system can better utilize past decision-making experience to guide the current decision-making process, thereby providing more accurate decision suggestions and improving the quality and efficiency of decision-making.

[0055] In particular, considering that the multi-dimensional and multi-scale semantic aggregation representation vector between the decision cases performs a multi-dimensional fusion representation of the semantic coding association features and the semantic interaction aggregation features of each decision case data, there will be a semantic distribution misalignment with the single-dimensional semantic coding features of the decision content to be reviewed, so that the semantic matching association matrix as a position-by-position association representation of the semantic coding vector of the decision content to be reviewed and the multi-dimensional and multi-scale semantic aggregation representation vector between the decision cases will have local distribution differences caused by vector dimension misalignment, and the semantic matching association matrix will have a diversified set expression distribution of cross-modal semantic aggregation. Therefore, it is expected to improve the balance between the regression mapping accuracy and completeness when the semantic matching association matrix is ​​input into the classifier-based intelligent decision generation module for classification regression, thereby improving the accuracy of the auxiliary decision results obtained.

[0056] In a preferred example, the semantic matching association matrix is ​​passed through a classifier-based intelligent decision generation module to obtain an auxiliary decision result, which includes:

[0057] Determine the maximum eigenvalue and the minimum eigenvalue of the semantic matching association matrix, and calculate the semantic matching association mean and the semantic matching association standard deviation of the feature set of the semantic matching association matrix;

[0058] Calculating the quotient of the semantic matching association mean and the semantic matching association standard deviation to obtain a semantic matching association statistical normalization value;

[0059] Calculate the inverse of each eigenvalue of the semantic matching association matrix, multiply it by the difference between the maximum eigenvalue of the semantic matching association and the minimum eigenvalue of the semantic matching association, and then perform point subtraction from the semantic matching association statistical normalization value to obtain a semantic matching association distribution approximation matrix;

[0060] Calculate an exponential function with a natural constant as the base and each eigenvalue of the semantic matching association distribution approximation matrix as the exponent to obtain a semantic matching association distribution class approximation matrix;

[0061] Performing point addition on the semantic matching association distribution class approximation matrix and the semantic matching association statistical normalized value, and calculating the base-two logarithm of the absolute value of each eigenvalue of the point addition matrix to obtain an optimized semantic matching association matrix;

[0062] The optimized semantic matching association matrix is ​​passed through a classifier-based intelligent decision generation module to obtain an auxiliary decision result.

[0063] The semantic matching association matrix, denoted as M, is optimized as follows:

[0064] M'=log2|η⊕M e |

[0065]

[0066] δ=m max -m min

[0067] Among them, m max Indicates the maximum eigenvalue of semantic matching association, m min represents the minimum eigenvalue of semantic matching association, μ represents the mean of semantic matching association, σ represents the standard deviation of semantic matching association, η is the statistical normalization value of semantic matching association, and δ is the difference between the maximum eigenvalue of semantic matching association and the minimum eigenvalue of semantic matching association, M represents the semantic matching association matrix, (·) ⊙-1 Indicates the calculation of the inverse of each eigenvalue of the matrix, ⊙ indicates point multiplication by position, ⊕ indicates addition by position, Subtract by position, M e represents the semantic matching association distribution class approximation matrix, and M' represents the optimized semantic matching association matrix.

[0068] Therefore, by performing random statistical normalization on the diverse feature set of the semantic matching association matrix, the confidence space constructed based on the overall eigenvalues ​​of the semantic matching association matrix is ​​approximated by a standardized continuous probability density distribution relative to the response hypothesis test of each eigenvalue of the semantic matching association matrix, thereby establishing the target reachability of the diverse feature distribution of the semantic matching association matrix to a unified regression target, so as to achieve a balanced executable performance between mapping accuracy and mapping completeness in the classification regression process based on the diverse feature distribution of the semantic matching association matrix, and improve the accuracy of the auxiliary decision results obtained by the intelligent decision generation module based on the classifier when the semantic matching association matrix is ​​input.

[0069] It is worth mentioning that those skilled in the art should be aware that before applying a deep neural network model for inference, the deep neural network model must be trained so that the deep neural network can implement specific functional capabilities.

[0070] Specifically, in the technical solution of the present application, the intelligent decision generation system guided by user prior information also includes a training module for training the semantic feature extractor for association between decision cases based on the text convolutional neural network model, the node message interaction aggregation network for fusion node importance, and the classifier-based intelligent decision generation module.

[0071] Figure 5 is a block diagram of a training module in an intelligent decision-making system guided by user prior information according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 5As shown, the training module 200 includes: a training data acquisition unit 210, which is used to acquire training data, wherein the training data includes historical training decision case data and training decision content to be reviewed of the target object, wherein each training decision case data in the historical training decision case data includes training case content and training decision labels, as well as the true label of the decision type; a historical training decision case semantic encoding unit 220, which is used to semantically encode each training decision case data in the historical training decision case data to obtain a set of training decision case data semantic encoding vectors; a local semantic extraction unit 230 between training decision cases, which is used to arrange the set of training decision case data semantic encoding vectors into a global semantic matrix of training historical decision case data and then pass it through the decision case correlation semantic feature extractor based on the text convolutional neural network model to obtain a local semantic dimension correlation representation vector between training decision cases; a semantic aggregation unit 240 between training decision cases, which is used to input the set of training decision case data semantic encoding vectors into the node message interaction aggregation network of the fusion node importance to obtain a sample dimension semantic aggregation representation between training decision cases. vector; a training multi-scale semantic aggregation unit 250, which is used to fuse the local semantic dimension association representation vector between the training decision cases and the sample dimension semantic aggregation representation vector between the training decision cases to obtain the multi-dimensional multi-scale semantic aggregation representation vector between the training decision cases; a training decision content semantic encoding unit 260, which is used to semantically encode the training decision content to be reviewed to obtain the training decision content semantic encoding vector; a training semantic matching unit 270, which is used to use the training decision content semantic encoding vector to be reviewed as a query vector, and calculate the query result relative to the multi-dimensional multi-scale semantic aggregation representation vector between the training decision cases to obtain a training semantic matching association matrix; a classification loss calculation unit 280, which is used to input the training semantic matching association matrix into the classifier-based intelligent decision generation module to obtain a classification loss function value; and a training unit 290, which is used to train the decision case association semantic feature extractor based on the text convolutional neural network model, the node message interaction aggregation network of the fusion node importance, and the classifier-based intelligent decision generation module based on the classification loss function value and through gradient descent back propagation.

[0072] It should be understandable that, considering that the multi-dimensional and multi-scale semantic aggregation representation vector between the training decision cases performs a multi-dimensional fusion representation of the semantic coding association features and the semantic interaction aggregation features of each training decision case data, there will be a semantic distribution misalignment with the single-dimensional semantic coding features of the training decision content to be reviewed, so that the training semantic matching association matrix, as a position-by-position association representation between the semantic coding vector of the training decision content to be reviewed and the multi-dimensional and multi-scale semantic aggregation representation vector between the training decision cases, will have local distribution differences caused by the misalignment of the vector dimensions, thereby affecting the classification iteration effect.

[0073] Based on this, in a preferred embodiment, in the process of training the decision case correlation semantic feature extractor based on the text convolutional neural network model, the node message interaction aggregation network that fused the node importance, and the classifier-based intelligent decision generation module based on the classification loss function value and through gradient descent back propagation, each time the training semantic matching association matrix is ​​classified and iterated through the classifier-based intelligent decision generation module, the training semantic matching association matrix is ​​optimized, specifically including the following optimization steps: expanding the training semantic matching association matrix into a training semantic matching association vector; determining the first semantic matching association feature value and the second semantic matching association feature value of any two different positions of the training semantic matching association vector; determining the n label probability values ​​corresponding to the n decision type labels respectively obtained by the training semantic matching association vector through the classifier-based intelligent decision generation module; calculating the relationship between the n label probability values. The standard deviation is calculated to obtain the semantic matching association relative distribution value, and the nth root of the product between the n label probability values ​​is calculated to obtain the semantic matching association collaborative distribution value; the first semantic matching association feature value and the second semantic matching association feature value are respectively subtracted from the semantic matching association relative distribution value and then multiplied to obtain the semantic matching association relative joint representation value; the first semantic matching association feature value and the second semantic matching association feature value are added and then divided by the semantic matching association collaborative distribution value to obtain the semantic matching association system response representation value; the semantic matching association relative joint representation value is added to the semantic matching association system response representation value as the matrix value at the coordinates corresponding to different positions of the first semantic matching association feature value and the second semantic matching association feature value to obtain the semantic matching association correction matrix; and the inner product of each row vector of the semantic matching association correction matrix and the semantic matching association vector is calculated to obtain the optimized training semantic matching association vector.

[0074] The above preferred embodiment uses the eigenvalue of the training semantic matching association matrix as a unit, and for the relative class probability distribution form and the collaborative class probability distribution form of the class probabilities corresponding to different labels obtained by the training semantic matching association matrix as a feature set through the intelligent decision generation module based on the classifier, the eigenvalue pairs of the training semantic matching association matrix are respectively characterized based on relativity and based on synergy. The response characterization avoids the class induction bias caused by the corresponding local distribution differences between the eigenvalues ​​of the training semantic matching association matrix, so as to establish a robust understanding paradigm for class cognition of the training semantic matching association matrix as a feature set as a whole, thereby improving the iteration effect of the training semantic matching association matrix as a feature set in the classification process, that is, improving the speed of classification training and the accuracy of classification results when the training semantic matching association matrix is ​​classified by the intelligent decision generation module. In this way, the system can better utilize past decision-making experience to guide the current decision-making process, thereby providing more accurate decision-making suggestions and improving the quality and efficiency of decision-making.

[0075] In summary, the intelligent decision generation system 100 based on user prior information guidance according to the embodiment of the present application is explained, which obtains the historical decision case data and the decision content to be reviewed of the target object, wherein each decision case data in the historical decision case data contains case content and decision labels, and uses artificial intelligence-based data processing and analysis technology to semantically encode and globally semantically aggregate the historical decision case data, and semantically understand the decision content to be reviewed, so as to intelligently obtain the auxiliary decision results of the decision type label according to the semantic association matching features between the historical decision case data and the decision content to be reviewed. In this way, the system can better utilize past decision-making experience to guide the current decision-making process, so as to provide more accurate decision suggestions and improve the quality and efficiency of decision-making.

[0076] As described above, the intelligent decision-making system 100 guided by user prior information according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with an intelligent decision-making algorithm guided by user prior information. In a possible implementation, the intelligent decision-making system 100 guided by user prior information according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the intelligent decision-making system 100 guided by user prior information can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the intelligent decision-making system 100 guided by user prior information can also be one of the many hardware modules of the wireless terminal.

[0077] Alternatively, in another example, the intelligent decision-making system 100 guided by user prior information and the wireless terminal may also be separate devices, and the intelligent decision-making system 100 guided by user prior information may be connected to the wireless terminal via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0078] The above specific implementation methods further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific implementation methods of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent decision-making system based on user prior information guidance, characterized in that: include: A historical decision case data acquisition module is used to acquire historical decision case data of a target object, wherein each decision case data in the historical decision case data includes case content and decision labels; A case data semantic coding module, used to semantically code each decision case data in the historical decision case data to obtain a set of decision case data semantic coding vectors; A case-by-case semantic feature extraction module is used to extract the associated semantic features between decision cases after arranging the set of semantic encoding vectors of the decision case data into a global semantic matrix of historical decision case data to obtain a local semantic dimension associated representation vector between decision cases; A semantic aggregation module, used for inputting the set of semantic encoding vectors of the decision case data into a node message interaction aggregation network integrating node importance to obtain a semantic aggregation representation vector of sample dimensions between decision cases; A multi-scale semantic fusion module, used to fuse the local semantic dimension association representation vector between the decision cases and the sample dimension semantic aggregation representation vector between the decision cases to obtain a multi-dimensional and multi-scale semantic aggregation representation vector between the decision cases; A module for obtaining decision content to be reviewed, used to obtain decision content to be reviewed; A semantic coding module for the decision content to be reviewed, used for semantically coding the decision content to be reviewed to obtain a semantic coding vector for the decision content to be reviewed; A semantic matching association module, used to use the semantic encoding vector of the decision content to be reviewed as a query vector, calculate the query result relative to the multi-dimensional and multi-scale semantic aggregation representation vector between the decision cases to obtain a semantic matching association matrix as a semantic matching association feature; The result generation module is used to obtain auxiliary decision results based on the semantic matching association features.

2. The intelligent decision-making system based on user prior information guidance according to claim 1 is characterized in that: The case-to-case semantic feature extraction module is used to: arrange the set of semantic encoding vectors of the decision case data into a global semantic matrix of historical decision case data, and then pass it through a decision case-to-decision case association semantic feature extractor based on a text convolutional neural network model to obtain a local semantic dimension association representation vector between the decision cases.

3. The intelligent decision-making system based on user prior information guidance according to claim 2 is characterized in that: The semantic aggregation module includes: A saliency factor calculation unit, used to measure the saliency factors of the first k-1 decision case data semantic encoding vectors in the set of decision case data semantic encoding vectors using a scoring function to obtain k-1 decision case data semantic saliency factors; A feature weighted aggregation unit, used to calculate the weighted sum of the first k-1 decision case data semantic coding vectors in the set of the decision case data semantic coding vectors by taking the k-1 decision case data semantic significance factors as weights to aggregate the first k-1 decision case data semantic coding vectors in the set of the decision case data semantic coding vectors to obtain a decision case data semantic aggregation representation vector of the rear node set; A feature modulation unit, used for linearly modulating the kth decision case data semantic coding vector in the set of decision case data semantic coding vectors using a preset hyperparameter to obtain a modulated kth decision case data semantic coding vector; A full-node aggregation unit, used for calculating the positional sum of the k-th decision case data semantic coding vector after the modulation and the decision case data semantic aggregation representation vector of the rear node set to obtain the decision case data semantic full-node aggregation representation vector; The multi-layer perceptron processing unit is used to input the semantic full-node aggregation representation vector of the decision case data into a multi-layer perceptron model to obtain the semantic aggregation representation vector of the sample dimension between the decision cases.

4. The intelligent decision-making system based on user prior information guidance according to claim 3 is characterized in that: The significant factor calculation unit comprises: A feature addition secondary subunit is used to calculate the matrix product between each weight coefficient matrix and the first k-1 decision case data semantic coding vectors to obtain the first k-1 decision case data semantic coding weighted feature vectors; A feature bias calculation secondary subunit, used for adding a corresponding bias vector to each of the semantically coded weighted feature vectors of the first k-1 decision case data to obtain the semantically coded biased feature vectors of the first k-1 decision case data; A feature activation secondary subunit, used for inputting the semantic encoding bias feature vectors of the first k-1 decision case data into a sigmoid function for activation to obtain the semantic encoding activation feature vectors of the first k-1 decision case data; The semantic relevance calculation secondary subunit is used to calculate the product between each decision case data semantic coding activation feature vector in the first k-1 decision case data semantic coding activation feature vectors and the weight coefficient vector to obtain the semantic relevance of the first k-1 decision case data; The semantic saliency factor calculation secondary subunit is used to perform nonlinear processing on the semantic associations of the first k-1 decision case data to obtain the semantic saliency factors of the k-1 decision case data.

5. The intelligent decision-making system based on user prior information guidance according to claim 4 is characterized in that: The semantic saliency factor calculation secondary subunit is used to: input the semantic associations of the first k-1 decision case data into a softmax function to obtain the semantic saliency factors of the k-1 decision case data.

6. The intelligent decision-making system based on user prior information guidance according to claim 5 is characterized in that: The semantic matching association module is used to: use the semantic encoding vector of the decision content to be reviewed as a query vector, calculate the vector product between the query vector and the transposed vector of the multi-dimensional and multi-scale semantic aggregation representation vector between the decision cases to obtain the semantic matching association matrix.

7. The intelligent decision-making system based on user prior information guidance according to claim 6 is characterized in that: The result generation module is used to: pass the semantic matching association matrix through a classifier-based intelligent decision generation module to obtain an auxiliary decision result, and the auxiliary decision result is a decision type label.

8. The intelligent decision-making system based on user prior information guidance according to claim 7 is characterized in that: It also includes a training module for training the decision case correlation semantic feature extractor based on the text convolutional neural network model, the node message interaction aggregation network for fusion node importance, and the classifier-based intelligent decision generation module.

9. The intelligent decision-making system based on user prior information guidance according to claim 8, characterized in that: The training module comprises: A training data acquisition unit, used to acquire training data, wherein the training data includes historical training decision case data and training decision content to be reviewed of the target object, wherein each training decision case data in the historical training decision case data includes training case content and training decision labels, as well as the true label of the decision type; A historical training decision case semantic encoding unit, used to semantically encode each training decision case data in the historical training decision case data to obtain a set of training decision case data semantic encoding vectors; A local semantic extraction unit between training decision cases, used for arranging the set of semantic encoding vectors of the training decision case data into a global semantic matrix of training historical decision case data and then passing the associated semantic feature extractor between decision cases based on the text convolutional neural network model to obtain a local semantic dimension associated representation vector between training decision cases; A semantic aggregation unit between training decision cases, used for inputting the set of semantic encoding vectors of the training decision case data into the node message interaction aggregation network of the fusion node importance to obtain a semantic aggregation representation vector of sample dimensions between training decision cases; A training multi-scale semantic aggregation unit is used to fuse the local semantic dimension association representation vector between the training decision cases and the sample dimension semantic aggregation representation vector between the training decision cases to obtain a multi-dimensional multi-scale semantic aggregation representation vector between the training decision cases; A training decision content to be reviewed semantic coding unit, used for semantically coding the training decision content to be reviewed to obtain a training decision content to be reviewed semantic coding vector; A training semantic matching unit, used to use the semantic encoding vector of the training decision content to be reviewed as a query vector, and calculate the query result relative to the multi-dimensional and multi-scale semantic aggregation representation vector between the training decision cases to obtain a training semantic matching association matrix; A classification loss calculation unit, used for inputting the training semantic matching association matrix into the classifier-based intelligent decision generation module to obtain a classification loss function value; A training unit is used to train the decision case correlation semantic feature extractor based on the text convolutional neural network model, the node message interaction aggregation network of the fusion node importance, and the classifier-based intelligent decision generation module based on the classification loss function value and through back propagation of gradient descent.

Images