CN118153684A - Knowledge-driven large model emotion tracing and propagation path analysis method - Google Patents

Abstract

The invention discloses a knowledge-driven large model emotion tracing and propagation path analysis method, which relates to the technical field of artificial intelligence and is used for conveying events and user emotions into a tracing path model so as to output causal relations between the events and the user emotions and shortest paths for user emotion propagation; the training process of the tracing path model is as follows: constructing a training set, wherein the training set comprises events and user moods; transmitting the event and the user emotion to a causality model, and generating whether causality exists between the event and the user emotion and the strength of the causality; transmitting the causal relationship and the strength of the causal relationship to a propagation path model to predict the shortest path of user emotion propagation, and adjusting model parameters in a tracing path model based on the generated shortest path of user emotion propagation; the emotion tracing and propagation path analysis method improves the accuracy of causal prediction of the event and the user emotion and effectively realizes tracing of the user emotion.

Description

A knowledge-driven large-scale model-based emotion tracing and propagation path analysis method

Technical Field

The present invention relates to the field of electronic mathematical data processing technology, and in particular to a knowledge-driven large-model emotion tracing and propagation path analysis method.

Background technique

In terms of predicting the causal relationship between individual user emotions and events, due to the uniqueness of events, existing methods find it difficult to effectively characterize and represent the characteristics of new events, making it difficult to determine the causal relationship between user-specific emotions and new events.

In terms of emotion propagation paths, existing methods generally construct a graph structure between emotions and events to realize emotion propagation path analysis. In this process, the strength of the relationship between emotions and events in the graph structure is ignored, resulting in inaccurate propagation path analysis.

Summary of the invention

Based on the technical problems existing in the background technology, the present invention proposes a knowledge-driven large-model emotion tracing and propagation path analysis method to effectively realize the prediction of causal relationships and the tracing of emotions.

The knowledge-driven large-model emotion tracing and propagation path analysis method proposed in the present invention transmits events and user emotions to the tracing path model to output the causal relationship between events and user emotions and the shortest path for user emotion propagation;

The traceability path model includes a causal relationship model and a propagation path model. The training process of the traceability path model is as follows:

S1: Construct a training set, which includes events and user sentiment/> ;

S2: Event and user sentiment/> The data is sent to the causal relationship model to determine whether there is a causal relationship between the generated event and the user's emotion, and the strength of the causal relationship.

S21: Events Perform feature encoding to obtain event representation/> , build event knowledge base based on historical event representation/> , use the key mapping and value mapping in the semantic enhancement model to transform the event knowledge base/> In the /> A representation of a historical event/> Mapped to key vector /> Sum value vector/> ;

S22: Computational event representation and key vector/> The weight between , and the weights/> Applied to value vector/> , weighted average to get similar features/> , similar features/> and event representation/> Connect in series to get enhanced event representation/> ;

S23: Enhanced event representation and user emotions/> Input into the big model to determine whether there is a causal relationship between the generated event and the user's emotion/> and the strength of the causal relationship/> ;

S3: transmitting the causal relationship and the strength of the causal relationship to the propagation path model to predict the shortest path of user emotion propagation, and adjusting the model parameters in the traceability path model based on the generated shortest path of user emotion propagation;

S31: Constructing a directed graph , directed graph/> The nodes include events and user emotions. Events at adjacent moments are connected by event directed edges, and user emotions at adjacent moments are connected by emotion directed edges. If there is a causal relationship between events and user emotions at the same moment, , then the event at that moment is connected to the user's emotion through an event-continuing directed edge;

S32: Set weights on both event directed edges and sentiment directed edges As a static weight, set the strength of the causal relationship on the event-sequential directed edge/> As a dynamic weight, weight /> is a preset value greater than 0;

S33: Based on directed graph , and the emotion at time T/> , for each event/> , find the event-continuation paths connecting all events with user emotions, and calculate the weights of the event-continuation paths;

S34: Using the shortest path algorithm, the emotion is calculated by taking the opposite number of the weight of the event path in the calculation. Source event with the shortest propagation path/> , and get the shortest source event/> The corresponding shortest path Z.

Furthermore, similar features The calculation formula is as follows:

in, Representing event knowledge base/> The total number of representations of historical events in .

Further, in step S33: finding all the event-continuation paths connecting the events and the user's emotions, and calculating the weights of the event-continuation paths, for the The correspondence between individual event paths and event path weights is as follows:

in, Indicates an event/> The first/> A continuous path, /> It is a directed graph/> The nodes in Represents a directed graph/> The total number of nodes, /> Indicates the continuation path/> the weight of, Indicates an event/> The first/> The first/> Each event has a directed edge, /> , that is, from/> To/> There is a directed edge with a weight of /> , weight/> The value is the event /> and user sentiment/> The strength of the causal relationship calculated in step S2/> ,/> Indicates an event/> The first/> The corresponding event in the event path, /> Indicates an event/> The first/> The corresponding user emotions in each event path, /> It is a sequential path/> Length weight penalty weight, /> .

Furthermore, the strength of the causal relationship is a real number between 0 and 1.

The advantages of the knowledge-driven large-model emotion tracing and propagation path analysis method provided by the present invention are: the knowledge-driven large-model emotion tracing and propagation path analysis method provided in the structure of the present invention, by introducing historical event knowledge and using the historical event representation of historical events, enhances the representation of current events, enriches the features contained in current events, helps the causal relationship model to extract new event features, and thus improves the accuracy of causal prediction of events and user emotions. In addition, by constructing a graph structure and modeling the weights of the edges between graphs based on causal relationships, the propagation path model can evaluate the impact of different events on emotions differently, thereby more effectively tracing the user's emotions.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a schematic diagram of the structure of the present invention;

Figure 2 is a training flowchart of the traceability path model.

Detailed ways

Below, the technical solution of the present invention is described in detail through specific embodiments. Many specific details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific implementation disclosed below.

This embodiment improves the existing large model, which refers to a machine learning model with large-scale parameters and complex computing structure. These large models are usually built from deep neural networks and have billions or even hundreds of billions of parameters. The purpose of designing large models is to improve the expressiveness and predictive performance of the model and to be able to handle more complex tasks and data. Large models are widely used in various fields, including natural language processing, computer vision, speech recognition, and recommendation systems. Large models learn complex patterns and features by training massive data, have more powerful generalization capabilities, and can make accurate predictions on unseen data.

ChatGPT's explanation of the big model is more understandable and more similar to that of humans in induction and thinking: the big model is essentially a deep neural network model trained using massive amounts of data. Its huge data and parameter scale enable the emergence of intelligence and demonstrate human-like intelligence.

As shown in Figures 1 and 2, the knowledge-driven large-model emotion tracing and propagation path analysis method proposed in the present invention transmits events and user emotions to the tracing path model to output the causal relationship between events and user emotions and the shortest path for user emotion propagation. This embodiment is mainly used in social media scenarios to identify the causes of individual user emotions and analyze the emotion propagation path.

The traceability path model includes the causal relationship model and the propagation path model. The causal relationship model is mainly used for causal prediction of events and user emotions, and the propagation path model is mainly used to predict the propagation path of user emotion paths.

In the causal model, Moment, given event /> , user sentiment/> , and event knowledge base/> , predict events/> and user emotions/> Is there a causal relationship between , and the strength of the causal relationship/> .

In the propagation path model, given the user sentiment at time T , and the strength of the causal relationship from time 1 to time T/> , analyze user emotions/> The propagation path, get the shortest path/> and emotions/> Source event with the shortest propagation path/> , realizing emotion tracing and path analysis.

The training process of the traceability path model is as follows:

S1: Construct a training set, which includes events and user sentiment/> ;

S2: Event and user sentiment/> The data is transmitted to the causal relationship model to determine whether there is a causal relationship between the generated event and the user's emotion and the strength of the causal relationship, specifically steps S21 to S23;

S21: Events Perform feature encoding to obtain event representation/> , build event knowledge base based on historical event representation/> , use the key mapping and value mapping in the semantic enhancement model to transform the event knowledge base/> In the /> A representation of a historical event/> Mapped to key vector /> Sum value vector/> ;

Use an event encoder (such as BERT) to encode events Perform feature encoding to obtain event representation/> ; Event Knowledge Base/> It is a package containing /> A collection of historical event representations, namely/> ,/> It is the first/> The representation of a historical event; the strength of the causal relationship/> is a real number between 0 and 1.

Key Vector Sum value vector/> The calculation is as follows:

in, Indicates a key mapping, /> Represents a value map.

S22: Computational event representation and key vector/> The weight between , and the weights/> Applied to value vector/> , weighted average to get similar features/> , similar features/> and event representation/> Connect in series to get enhanced event representation/> ;

Weights , similar features/> and enhanced event representation/> The calculation of is as follows:

in, Representing event knowledge base/> A collection of historical event representations in ,/> Indicates series connection, /> Representing event knowledge base/> The serial number of the historical event in.

S23: Enhanced event representation and user emotions/> Input into the big model, use the standard big model text generation method to generate whether there is a causal relationship between the event and the user's emotion/> and the strength of the causal relationship/> ;

Through steps S21 to S23, historical similar events are used as event knowledge base , using event knowledge base/> This knowledge enhances the representation of current events and enables the prediction of the causal relationship between events and user emotions.

S3: transmitting the causal relationship and the strength of the causal relationship to the propagation path model to predict the shortest path of user emotion propagation, and adjusting the model parameters in the tracing path model based on the generated shortest path of user emotion propagation, including steps S31 to S34;

S31: Constructing a directed graph , directed graph/> The nodes include events and user emotions. Events at adjacent moments are connected by event directed edges, and user emotions at adjacent moments are connected by emotion directed edges. If there is a causal relationship between events and user emotions at the same moment, , then the event at that moment is connected to the user's emotion through a directed edge;

For example, for adjacent Moment, /> Events and time Events at a certain moment are connected by directed event edges; /> User emotions at the moment and/> The user emotions at different moments are connected by directed emotion edges.

Through step S2, Is there a causal relationship between the event at that moment and the user's emotions/> , if/> There is a causal relationship between the event at that moment and the user's emotion, then in/> The event at the moment is connected to the user's emotion through a sequential directed edge. If/> There is no causal relationship between the event at that moment and the user's emotion, then in/> The event at a moment is not connected to the user's emotion through a sequential directed edge.

S32: Set weights on both event directed edges and sentiment directed edges As a static weight, set the strength of the causal relationship on the event-sequential directed edge/> As a dynamic weight, weight /> is a preset value greater than 0;

S33: Based on directed graph , and the emotion at time T/> , for each event/> , find the event-continuation paths connecting all events with user emotions, and calculate the weights of the event-continuation paths;

For The correspondence between individual event paths and event path weights is as follows:

in, Indicates an event/> The first/> A continuous path, /> It is a directed graph/> The nodes in Represents a directed graph/> The total number of nodes, /> Indicates the continuation path/> the weight of, Indicates an event/> The first/> The first/> Each event has a directed edge, /> , that is, from/> To/> There is a directed edge with a weight of /> , weight/> The value is the event /> and user emotions/> The strength of the causal relationship calculated in step S2/> ,/> Indicates an event/> The first/> The corresponding event in the event path, /> Indicates an event/> The first/> The corresponding user emotions in each event path, /> It is a sequential path/> Length weight penalty weight, /> .

S34: Using the shortest path algorithm, by taking the opposite number of the weight of the successive path in the calculation, the problem is converted into a problem of solving the path with the minimum weight, and the emotion is calculated. Source event with the shortest propagation path/> , and get the shortest source event/> The corresponding shortest path Z.

in, represents the minimum function, /> Indicates the continuation path/> The weight of Indicates an event/> The first/> A continuation path.

According to steps S1 to S3, by introducing historical event knowledge and using historical event representations of historical events, the representation of the current event is enhanced, making the features contained in the current event richer, helping the causal relationship model to extract new event features, thereby improving the accuracy of causal prediction of events and user emotions. In addition, by constructing a graph structure and modeling the weights of the edges between graphs based on causal relationships, the propagation path model can evaluate the impact of different events on emotions differently, thereby more effectively tracing the source of emotions.

The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims (4)

1. A large-model emotion tracing and propagation path analysis method based on knowledge-driven, characterized in that events and user emotions are fed into the tracing path model to output the causal relationship between events and user emotions and the shortest path for user emotion propagation; The traceability path model includes a causal relationship model and a propagation path model. The training process of the traceability path model is as follows: S1: Construct a training set, which includes events and user sentiment/> ; S2: Event and user sentiment/> Send it to the causal relationship model to determine whether there is a causal relationship between the generated event and the user's emotion/> and the strength of the causal relationship/> ; S21: Events Perform feature encoding to obtain event representation/> , build event knowledge base based on historical event representation/> , use the key mapping and value mapping in the semantic enhancement model to transform the event knowledge base/> In the /> A representation of a historical event/> Mapped to key vector /> Sum value vector/> ; S22: Computational event representation and key vector/> The weight between , and the weights/> Applied to value vector/> , weighted average to get similar features/> , similar features/> and event representation/> Connect in series to get enhanced event representation/> ; S23: Enhanced event representation and user emotions/> Input into the big model to determine whether there is a causal relationship between the generated event and the user's emotion/> and the strength of the causal relationship/> ; S3: transmitting the causal relationship and the strength of the causal relationship to the propagation path model to predict the shortest path of user emotion propagation, and adjusting the model parameters in the traceability path model based on the generated shortest path of user emotion propagation; S31: Constructing a directed graph , directed graph/> The nodes include events and user emotions. Events at adjacent moments are connected by event directed edges, and user emotions at adjacent moments are connected by emotion directed edges. If there is a causal relationship between events and user emotions at the same moment, , then the event at that moment is connected to the user's emotion through a directed edge; S32: Set weights on both event directed edges and sentiment directed edges As a static weight, set the strength of the causal relationship on the event-sequential directed edge/> As a dynamic weight, weight /> is a preset value greater than 0; S33: Based on directed graph , and the emotion at time T/> , for each event/> , find the event-continuation paths connecting all events with user emotions, and calculate the weights of the event-continuation paths; S34: Using the shortest path algorithm, the emotion is calculated by taking the opposite number of the weight of the event path in the calculation. Source event with the shortest propagation path/> , and get the shortest source event/> The corresponding shortest path Z. 2. The knowledge-driven large-model emotion tracing and propagation path analysis method according to claim 1 is characterized in that similar features The calculation formula is as follows: in, Representing event knowledge base/> The total number of representations of historical events in . 3. The method for tracing the source of emotion and propagation path of a large model based on knowledge-driven analysis according to claim 1 is characterized in that, in step S33: finding the event-continuation paths connecting all events with user emotions and calculating the weights of the event-continuation paths, for the first The correspondence between individual event paths and event path weights is as follows: in, Indicates an event/> The first/> A continuous path, /> It is a directed graph/> Nodes in,/> Represents a directed graph/> The total number of nodes, /> Indicates the continuation path/> the weight of, Indicates an event/> The first/> The first/> Each event has a directed edge, /> , that is, from/> To/> There is a directed edge with a weight of /> , weight/> The value is the event /> and user emotions/> The strength of the causal relationship calculated in step S2/> ,/> Indicates an event/> The first/> The corresponding event in the event path, /> Indicates an event/> The first/> The corresponding user emotions in each event path, /> It is a sequential path/> Length weight penalty weight, /> . 4. The knowledge-driven large-model emotion tracing and propagation path analysis method according to claim 3 is characterized in that the strength of the causal relationship is a real number between 0 and 1.