Consumer satisfaction analysis method and device based on graph convolutional network

Abstract

The invention relates to the technical field of big data analysis, and discloses a consumer satisfaction analysis method and device based on a graph convolutional network, and the method comprises the steps: capturing initial consumer comment data of a target product through employing a distributed network crawler technology; preprocessing the initial consumer comment data to obtain target consumer comment data; performing theme recognition on the target consumer comment data by adopting a pre-trained large language model to obtain a theme; performing sentiment analysis based on the target consumer comment data and the theme to obtain a sentiment tendency; and performing satisfaction analysis based on the target consumer comment data, the theme and the emotional tendency by adopting a graph convolutional network to obtain a satisfaction score of the target consumer comment data to the target product. According to the method, themes in the comments and emotional attitudes of consumers to target products are deeply mined, the data processing efficiency is improved, accurate prediction of consumer satisfaction is realized, and a scientific basis is provided for decision making of enterprises.

Description

Technical Field

[0001] The present invention relates to the technical field of big data analysis, and in particular to a consumer satisfaction analysis method and device based on a graph convolutional network. Background Art

[0002] As market competition intensifies, consumer satisfaction with products directly affects the market performance and long-term development of products. With the rapid development of the Internet and social media platforms, consumers are increasingly expressing their opinions and feedback on products through online platforms, such as posting comments on online platforms to share their usage experience, product evaluation and brand preferences.

[0003] However, consumer review data is characterized by large data volume, wide data sources and strong real-time nature, making it difficult to directly analyze through traditional methods such as questionnaires and face-to-face interviews. Therefore, how to conduct satisfaction analysis based on consumer review data is a problem that needs to be solved. Summary of the invention

[0004] In view of this, the present invention provides a consumer satisfaction analysis method and device based on graph convolutional network to solve the problem of performing satisfaction analysis based on consumer review data.

[0005] In a first aspect, the present invention provides a method for analyzing consumer satisfaction based on a graph convolutional network, the method comprising:

[0006] Use distributed web crawler technology to capture multiple initial consumer review data for target products;

[0007] Preprocessing multiple pieces of initial consumer review data to obtain multiple pieces of target consumer review data;

[0008] Use a pre-trained large language model to identify topics of multiple target consumer review data and obtain the topics of multiple target consumer review data;

[0009] Using a pre-trained large language model, sentiment analysis is performed based on multiple target consumer review data and the topics of multiple target consumer review data to obtain the sentiment tendency of multiple target consumer review data;

[0010] A graph convolutional network is used to perform satisfaction analysis based on multiple target consumer review data, the topics and sentiment tendencies of multiple target consumer review data, and obtain the satisfaction scores of multiple target consumer review data for the target product.

[0011] The embodiment of the present invention provides a consumer satisfaction analysis method based on a graph convolutional network. The method captures initial consumer review data in real time through distributed web crawler technology, can quickly understand market dynamics and the latest feedback from consumers, and then uses a pre-trained large language model to perform topic recognition and sentiment analysis on the pre-processed target consumer review data. It can deeply mine the implicit topics in the reviews and the real intentions of consumers, and can also carefully distinguish the emotional attitudes of consumers towards target products. Finally, the characteristics of the graph convolutional network are used to perform in-depth analysis and prediction of consumer satisfaction, which can accurately identify the correlation and similarity between reviews, and obtain satisfaction scores for target products, thereby improving data processing efficiency, achieving accurate prediction of consumer satisfaction, and providing a scientific basis for corporate decision-making.

[0012] In an optional implementation, multiple pieces of initial consumer review data are preprocessed to obtain multiple pieces of target consumer review data, including:

[0013] For any initial consumer review data, a large language model is used to generate a semantic embedding vector of the initial consumer review data;

[0014] Perform noise detection on the semantic embedding vector of the initial consumer review data to obtain the noise score of the initial consumer review data;

[0015] Initial consumer review data with noise scores greater than a preset noise threshold and initial consumer review data with null values ​​are removed from the plurality of initial consumer review data to obtain a plurality of target consumer review data.

[0016] An embodiment of the present invention provides a consumer satisfaction analysis method based on a graph convolutional network. By performing noise detection on each initial consumer review data, the method can remove noise from the initial consumer review data and remove null values ​​at the same time, thereby obtaining target consumer review data suitable for data analysis, thereby providing data support for analyzing consumer satisfaction.

[0017] In an optional implementation, before using a pre-trained large language model to perform topic identification on multiple target consumer review data and obtaining the topics of the multiple target consumer review data, the method further includes:

[0018] Acquire a plurality of preprocessed sample text data, wherein any sample text data includes a topic label and a sentiment tendency label;

[0019] For any sample text data, a dimension reduction matrix and a dimension increase matrix are introduced into the large language model to adjust the weight matrix of the large language model;

[0020] Based on the large language model after adjusting the weight matrix, the sample text data is subjected to topic identification and sentiment analysis to obtain the sample topic and sample sentiment tendency of the sample text data;

[0021] Based on the difference between the topic label of the sample text data and the sample topic, and the difference between the sentiment label and the sample sentiment, update the dimension reduction matrix and the dimension increase matrix;

[0022] Repeat the above weight matrix adjustment process and the dimension reduction matrix and dimension increase matrix update process until the preset end condition is reached, and use the large language model after the last weight matrix adjustment as the pre-trained large language model.

[0023] An embodiment of the present invention provides a consumer satisfaction analysis method based on a graph convolutional network. A large language model is pre-trained based on sample text data and its labels. The continuous iterative weight adjustment and matrix optimization process ensures that the model can gradually reduce the gap between the predicted results and the actual labels, so that the pre-trained large language model has higher accuracy when performing topic identification and sentiment analysis. At the same time, compared with training from scratch, fine-tuning using the existing large language model saves time and computing resources and improves data processing efficiency.

[0024] In an optional implementation, a pre-trained large language model is used to perform topic identification on multiple target consumer review data to obtain the topics of the multiple target consumer review data, including:

[0025] For any target consumer review data, a pre-trained large language model is used to extract features of the target consumer review data to obtain the topic feature vector of the target consumer review data;

[0026] Perform topic clustering on the topic feature vectors of the target consumer review data to obtain multiple candidate topics;

[0027] Determine the probability distribution of multiple candidate topics, and select the candidate topic with the highest probability as the topic of the target consumer review data.

[0028] The embodiment of the present invention provides a consumer satisfaction analysis method based on a graph convolutional network. By extracting and clustering topic features of target consumer review data, it can not only identify obvious topics, but also discover potential, fine-grained topic classifications, thereby improving the accuracy of topic identification and providing support for analyzing consumer satisfaction.

[0029] In an optional implementation, a pre-trained large language model is used to perform sentiment analysis based on multiple target consumer review data and the topics of the multiple target consumer review data to obtain the sentiment tendencies of the multiple target consumer review data, including:

[0030] For any target consumer review data, a pre-trained large language model is used to extract features based on the target consumer review data and the subject of the target consumer review data to obtain the sentiment feature vector of the target consumer review data;

[0031] Based on the sentiment feature vector of the target consumer's comment data, the probability distribution of the sentiment categories of the target consumer's comment data is determined, and the sentiment category with the largest probability is taken as the sentiment tendency of the target consumer's comment data.

[0032] The embodiment of the present invention provides a consumer satisfaction analysis method based on a graph convolutional network. By first performing topic identification and then sentiment analysis, and extracting sentiment feature vectors based on topic information, it can capture more subtle sentiment changes, and can make sentiment judgments based on understanding the review topic, thereby improving the accuracy of sentiment analysis and providing support for analyzing consumer satisfaction.

[0033] In an optional implementation, a graph convolutional network is used to perform satisfaction analysis based on multiple target consumer review data, themes and sentiment tendencies of the multiple target consumer review data, and obtain satisfaction scores of the multiple target consumer review data for the target product, including:

[0034] Based on multiple target consumer review data, the topics and sentiment tendencies of multiple target consumer review data, a consumer review graph structure is constructed;

[0035] By using a graph convolutional network and based on the consumer review graph structure, we can obtain the satisfaction scores of multiple target consumer review data on the target product.

[0036] An embodiment of the present invention provides a consumer satisfaction analysis method based on a graph convolutional network. By constructing a consumer review graph structure, it is possible to capture the complex relationships between reviews. The graph convolutional network can comprehensively consider the influence of all reviews at the level of the entire review network, provide a more global and comprehensive target product satisfaction score, and improve the accuracy of consumer satisfaction analysis.

[0037] In an optional implementation, based on multiple target consumer review data, themes and sentiment tendencies of the multiple target consumer review data, a consumer review graph structure is constructed, including:

[0038] Using graph convolutional networks, multiple nodes in the consumer review graph structure are generated based on multiple target consumer review data;

[0039] Based on multiple target consumer review data, the topics and sentiment tendencies of the multiple target consumer review data, multiple nodes are assigned node features;

[0040] Based on the similarity between multiple target consumer review data, edges are constructed between multiple nodes.

[0041] A consumer satisfaction analysis method based on a graph convolutional network provided by an embodiment of the present invention reveals the direct and indirect connections between comments through a graph structure, helps understand which comments are discussing similar topics or sharing the same emotional attitudes, and provides a more intuitive and accurate way to analyze consumer satisfaction.

[0042] In an optional implementation, the graph convolutional network includes a multi-layer convolutional network and an output layer. The graph convolutional network is used to obtain satisfaction scores of multiple target consumer review data on a target product based on a consumer review graph structure, including:

[0043] Using graph convolutional networks, in each layer of convolutional networks, the node features of each node in the consumer review graph structure are updated;

[0044] For any node, the global feature vector of the node is obtained based on the node features updated by each layer of convolutional network;

[0045] The output layer of the graph convolutional network is used to make predictions based on the global feature vector of the node to obtain the satisfaction score of the target consumer review data corresponding to the node.

[0046] An embodiment of the present invention provides a consumer satisfaction analysis method based on a graph convolutional network. Through the node feature update mechanism in each layer of the convolutional network, it can deeply explore the complex patterns in the review content and structure, learn higher-level abstract features, and aggregate the node features of each node after the update of each layer of the convolutional network to form a feature vector from a global perspective. The satisfaction score is then determined based on the global feature vector, making the consumer satisfaction prediction more comprehensive and accurate.

[0047] In an optional embodiment, the method further includes:

[0048] For any target consumer review data, determine the expert satisfaction score of the target consumer review data;

[0049] Based on the difference between the expert satisfaction scores and the satisfaction scores of multiple target consumer review data, the network parameters of the graph convolutional network are updated.

[0050] An embodiment of the present invention provides a consumer satisfaction analysis method based on a graph convolutional network. The network parameters of the graph convolutional network are updated by using the difference between the satisfaction scores predicted by the graph convolutional network for consumer reviews and the satisfaction scores obtained based on expert experience, thereby improving the prediction performance of the graph convolutional network.

[0051] In a second aspect, the present invention provides a consumer satisfaction analysis device based on a graph convolutional network, the device comprising:

[0052] A crawling module is used to crawl multiple initial consumer review data of target products using distributed web crawler technology;

[0053] A preprocessing module, used to preprocess a plurality of initial consumer review data to obtain a plurality of target consumer review data;

[0054] A recognition module is used to use a pre-trained large language model to perform topic recognition on multiple target consumer review data to obtain the topics of the multiple target consumer review data;

[0055] An analysis module, for using a pre-trained large language model to perform sentiment analysis based on multiple target consumer review data and the topics of the multiple target consumer review data, to obtain the sentiment tendency of the multiple target consumer review data;

[0056] The scoring module is used to use a graph convolutional network to perform satisfaction analysis based on multiple target consumer review data, themes and sentiment tendencies of multiple target consumer review data, and obtain satisfaction scores of multiple target consumer review data for target products.

[0057] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the consumer satisfaction analysis method based on graph convolutional network of the above-mentioned first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0058] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the consumer satisfaction analysis method based on a graph convolutional network according to the first aspect or any corresponding embodiment thereof.

[0059] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions, which are used to enable a computer to execute the consumer satisfaction analysis method based on a graph convolutional network according to the above-mentioned first aspect or any corresponding embodiment. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0061] Figure 1 is a flowchart of a method for analyzing consumer satisfaction based on a graph convolutional network according to an embodiment of the present invention;

[0062] Figure 2 is a flowchart of another method for analyzing consumer satisfaction based on a graph convolutional network according to an embodiment of the present invention;

[0063] Figure 3 is a flowchart of another method for analyzing consumer satisfaction based on a graph convolutional network according to an embodiment of the present invention;

[0064] Figure 4 is a structural block diagram of a consumer satisfaction analysis device based on a graph convolutional network according to an embodiment of the present invention;

[0065] Figure 5 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0067] Consumers often post comments on online platforms to share their usage experience, product evaluations, and brand preferences. However, consumer comment data has the characteristics of large data volume, wide data sources, and strong real-time performance, making it difficult to directly analyze using traditional methods. Therefore, how to conduct satisfaction analysis based on consumer comment data is a problem that needs to be solved. A consumer satisfaction analysis method based on a graph convolutional network provided in an embodiment of the present invention uses distributed web crawler technology to capture consumer comment data in real time, and then uses a pre-trained large language model to perform topic recognition and sentiment analysis on the pre-processed consumer comment data. Finally, using the characteristics of the graph convolutional network, consumer satisfaction is deeply analyzed and predicted to obtain a satisfaction score for the target product, thereby improving data processing efficiency, achieving accurate prediction of consumer satisfaction, and providing a scientific basis for corporate decision-making.

[0068] According to an embodiment of the present invention, an embodiment of a consumer satisfaction analysis method based on a graph convolutional network is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0069] In this embodiment, a consumer satisfaction analysis method based on a graph convolutional network is provided, which can be used in electronic devices. Figure 1 is a flowchart of a method for analyzing consumer satisfaction based on a graph convolutional network according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0070] Step S101, using distributed web crawler technology to crawl multiple initial consumer comment data of the target product. Specifically, the user satisfaction analysis method based on graph convolutional network provided by the embodiment of the present invention is described by taking the cigarette industry as an example, and the target product is cigarettes. Any consumer comment data includes consumer nickname, consumer IP (Internet Protocol Address), comment time and comment content, etc. Distributed web crawler technology is implemented based on a crawler program, and the crawler program includes a master node and multiple slave nodes. The master node is responsible for the allocation and scheduling of crawling tasks, and the slave node is responsible for executing crawling tasks. The master node generates crawling tasks based on the keywords entered by the user, such as the brand name of cigarettes and the target social platform, and assigns them to the slave nodes. Each slave node submits the keywords to the target social platform for retrieval by simulating user operations or calling the platform API (Application Programming Interface) according to the assigned tasks, automatically finds videos and graphic links related to the keywords entered by the user, and then obtains page data through HTTP requests, parses the page content, and extracts the comment content. Multiple slave nodes can realize parallel crawling, and dynamically increase or decrease the number of threads through the thread pool management unit to adapt to the request restrictions and load conditions of different social platforms, and ensure the efficiency and stability of data crawling. For example, when the response speed of the target social platform is slow, the number of threads is automatically reduced. The task scheduling unit of the crawler optimizes the crawling strategy according to the load and network conditions of the target social platform. For example, when the network condition is poor, the crawling tasks of each slave node are automatically reduced to avoid data collection failure. At the same time, it supports task priority management to ensure that the comment data of important social platforms or specific keywords are captured first. When facing the consumer comment crawling tasks of multiple target products at the same time, slave nodes can be assigned for priority crawling according to user needs. For example, the comment data involving "cigarette brand A" is captured first. When all slave nodes have completed the crawling task, the captured initial consumer comment data will be passed to the master node, which will be processed uniformly by the master node. Optionally, Table 1 below is an example of the captured initial consumer comment data.

[0071] Table 1 Initial consumer review data

[0072] Comments Comment time Consumer nickname Consumer IP This cigarette tastes very good 2024-07-31 Xiao Zhang City A Brand A cigarettes are a bit expensive 2024-07-30 Newbie Xiao Wang City B B brand cigarettes are cost-effective and recommended 2024-07-27 Xiao Li who sells cigarettes City A Nice packaging 2024-07-25 Don't buy cigarettes City C Bought it as a gift, my friend said it was good 2024-07-24 Cigarette Master City D

[0073] Step S102, preprocessing the multiple initial consumer review data to obtain multiple target consumer review data. Specifically, the initial consumer review data usually contains noise, such as emoticons, special symbols, etc., or null values, etc. The initial consumer review data is converted into a structured form through the preprocessing process to facilitate subsequent analysis.

[0074] Step S103, using a pre-trained large language model, the topic identification is performed on the multiple target consumer comment data to obtain the topics of the multiple target consumer comment data. Specifically, a large language model, such as BERT, Qwen, etc., is used, and after pre-training, the main discussion topics of each target consumer comment data are analyzed, such as price, taste, packaging, etc., which is helpful for subsequent sentiment analysis and satisfaction analysis.

[0075] Step S104, using the pre-trained large language model, based on the multiple target consumer comment data and the themes of the multiple target consumer comment data, sentiment analysis is performed to obtain the sentiment tendency of the multiple target consumer comment data. Specifically, using the pre-trained large language model in the above step S103, on the basis of understanding the theme of each target consumer comment data, focusing on the emotional attitude of the comment, judging the consumer's emotional tendency, such as positive, negative or neutral, is helpful for subsequent sentiment analysis and satisfaction analysis.

[0076] Step S105, using a graph convolutional network, based on multiple target consumer comment data, themes and sentiment tendencies of multiple target consumer comment data, a satisfaction analysis is performed to obtain satisfaction scores of multiple target consumer comment data for the target product. Specifically, through the graph convolutional network, a comprehensive analysis is performed based on the theme and sentiment tendencies of each target consumer comment data to obtain a satisfaction score, which reflects the comprehensive satisfaction of consumers with the target product.

[0077] The embodiment of the present invention provides a consumer satisfaction analysis method based on a graph convolutional network. The method captures initial consumer review data in real time through distributed web crawler technology, can quickly understand market dynamics and the latest feedback from consumers, and then uses a pre-trained large language model to perform topic recognition and sentiment analysis on the pre-processed target consumer review data. It can deeply mine the implicit topics in the reviews and the real intentions of consumers, and can also carefully distinguish the emotional attitudes of consumers towards target products. Finally, the characteristics of the graph convolutional network are used to perform in-depth analysis and prediction of consumer satisfaction, which can accurately identify the correlation and similarity between reviews, and obtain satisfaction scores for target products, thereby improving data processing efficiency, achieving accurate prediction of consumer satisfaction, and providing a scientific basis for corporate decision-making.

[0078] In this embodiment, a consumer satisfaction analysis method based on a graph convolutional network is provided, which can be used in the above-mentioned electronic device. Figure 2 is a flowchart of another method for analyzing consumer satisfaction based on a graph convolutional network according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:

[0079] Step S201: Use distributed web crawler technology to crawl multiple initial consumer review data of the target product. Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0080] Step S202: pre-process the multiple pieces of initial consumer review data to obtain multiple pieces of target consumer review data.

[0081] Specifically, the above step S202 includes:

[0082] Step S2021: For any initial consumer review data, a large language model is used to generate a semantic embedding vector of the initial consumer review data. Specifically, the large language model is used to extract features of the initial consumer review data using the following formula (1) to obtain a semantic embedding vector representing the semantic information contained in the initial consumer review data.

[0083] E(x i )=LM encode (x i ) (1)

[0084] Among them, x i represents the i-th initial consumer review data; E(x i ) represents the semantic embedding vector of the i-th initial consumer review data; LM encode Represents a large language model.

[0085] Step S2022: noise detection is performed on the semantic embedding vector of the initial consumer review data to obtain a noise score of the initial consumer review data. Specifically, a large language model is used to perform context understanding and semantic analysis on the semantic embedding vector through the following formula (2) to identify noise in the review that is irrelevant to the target product.

[0086] NoiseScore(x i ) = f LM (E(x i )) (2)

[0087] Among them, x i represents the i-th initial consumer review data; NoiseScore(x i ) represents the noise score of the i-th initial consumer review data; f LM Represents the characteristic function of the large language model; E(x i ) represents the semantic embedding vector of the i-th initial consumer review data.

[0088] Step S2023, clear the initial consumer review data with noise scores greater than a preset noise threshold and the initial consumer review data with null values ​​from the multiple initial consumer review data to obtain multiple target consumer review data. Specifically, the preset noise threshold is determined by the following formula (3). When the noise score of the initial consumer review data is greater than the preset noise threshold, it means that there is a lot of noise data in the review, which affects the accuracy of the subsequent consumer satisfaction analysis and needs to be cleared. At the same time, there may be null values ​​in the initial consumer review data, which also need to be cleared to save computing resources. By clearing noise and null values ​​from the initial consumer review data, target consumer review data suitable for data analysis is obtained, providing data support for analyzing consumer satisfaction.

[0089] τ=μ+σk(3)

[0090] Among them, τ represents the preset noise threshold; μ represents the average value of the noise scores of multiple initial consumer review data; σ represents the standard deviation of the noise scores of multiple initial consumer review data; k represents the adjustment factor, which takes the value of 1 when the standard deviation is less than the average value, and takes the value of 2 when the standard deviation is not less than the average value.

[0091] Step S203, obtain a plurality of preprocessed sample text data, any sample text data includes a topic label and a sentiment tendency label. Specifically, in order to improve the accuracy of consumer satisfaction analysis, the large language model used in the above step S202 can be pre-trained through steps S203 to S207. First, obtain a plurality of preprocessed sample text data. These sample text data have a wide range of sources, a large number and carry known labels to ensure the pre-training effect. The sample text data is pre-processed before pre-training, such as de-duplication, normalization, missing value processing, etc., which is not limited in the embodiment of the present invention.

[0092] Step S204, for any sample text data, introduce a dimensionality reduction matrix and a dimensionality increase matrix into the large language model, and adjust the weight matrix of the large language model. Specifically, the LoRA (Low-Rank Adaptation) technology is used to train the large language model to reduce the number of parameter updates and achieve efficient model training. When the large language model processes any sample text data, LoRA is used to introduce a low-rank matrix, that is, a dimensionality reduction matrix and a dimensionality increase matrix, and the original weight matrix of the large language model is adjusted by the following formula (4), reducing the demand for computing resources and speeding up the training speed, while increasing the expressive power of the model and capturing potential complex relationships.

[0093] W'=W+A*B (4)

[0094] Among them, W' represents the adjusted weight matrix; W represents the original weight matrix; A represents the reduced dimension matrix; B represents the increased dimension matrix.

[0095] Step S205, based on the large language model after adjusting the weight matrix, subject identification and sentiment analysis are performed on the sample text data to obtain sample subjects and sample sentiment tendencies of the sample text data. Specifically, after the weight matrix of the large language model is adjusted, subject identification is first performed on the sample text data to obtain sample subjects, and then sentiment analysis is performed based on the sample text data and the sample subjects to obtain sample sentiment tendencies.

[0096] Step S206, based on the difference between the topic label of the sample text data and the sample topic, and the difference between the sentiment label and the sample sentiment label, update the dimension reduction matrix and the dimension increase matrix. Specifically, the sample topic is compared with the topic label, and the sample sentiment label is compared with the sentiment label, and the loss is calculated by the loss function. Based on the loss between the predicted result and the actual label, the dimension reduction matrix and the dimension increase matrix are updated so that after the weight matrix of the large language model is adjusted by the updated matrix, the prediction error of the large language model is reduced, and the model output is closer to the actual label.

[0097] Step S207, repeat the adjustment process of the weight matrix and the update process of the dimension reduction matrix and the dimension increase matrix until the preset end condition is reached, and use the large language model after the last weight matrix adjustment as the pre-trained large language model. Specifically, for the sample text data currently input into the large language model, based on the currently updated dimension reduction matrix and dimension increase matrix, the weight matrix of the large language model is adjusted again when the next sample text data is input, and then the dimension reduction matrix and dimension increase matrix are continued to be updated by the difference between the prediction result and the actual label of the next input sample text data. For each sample text data input into the large language model, the above steps S204 to S206 are executed until the preset end condition is reached, and the pre-training of the large language model is completed. Among them, the preset end condition can be training a preset number of sample text data or the model performance of the large language model meets the requirements, etc., and the embodiment of the present invention does not limit this. By pre-training the large language model, the pre-trained large language model has higher accuracy when performing topic identification and sentiment analysis. At the same time, compared with training from scratch, fine-tuning the existing large language model saves time and computing resources and improves data processing efficiency.

[0098] Step S208: Use the pre-trained large language model to perform topic identification on the multiple target consumer review data to obtain the topics of the multiple target consumer review data. Figure 1 Step S103 of the illustrated embodiment will not be described in detail here.

[0099] Step S209: Use the pre-trained large language model to perform sentiment analysis based on the multiple target consumer review data and the themes of the multiple target consumer review data to obtain the sentiment tendency of the multiple target consumer review data. Figure 1 Step S104 of the illustrated embodiment will not be described in detail here.

[0100] Step S210, using a graph convolutional network, based on multiple target consumer review data, themes and sentiment tendencies of multiple target consumer review data, to perform satisfaction analysis, and obtain satisfaction scores of multiple target consumer review data for the target product. For details, please refer to Figure 1 Step S105 of the illustrated embodiment will not be described in detail here.

[0101] The embodiment of the present invention provides a consumer satisfaction analysis method based on a graph convolutional network. The method captures initial consumer review data in real time through distributed web crawler technology, can quickly understand market dynamics and the latest feedback from consumers, and then uses a pre-trained large language model to perform topic recognition and sentiment analysis on the pre-processed target consumer review data. It can deeply mine the implicit topics in the reviews and the real intentions of consumers, and can also carefully distinguish the emotional attitudes of consumers towards target products. Finally, the characteristics of the graph convolutional network are used to perform in-depth analysis and prediction of consumer satisfaction, which can accurately identify the correlation and similarity between reviews, and obtain satisfaction scores for target products, thereby improving data processing efficiency, achieving accurate prediction of consumer satisfaction, and providing a scientific basis for corporate decision-making.

[0102] In this embodiment, a consumer satisfaction analysis method based on a graph convolutional network is provided, which can be used in the above-mentioned electronic device. Figure 3 is a flowchart of another method for analyzing consumer satisfaction based on a graph convolutional network according to an embodiment of the present invention. Figure 3 As shown, the process includes the following steps:

[0103] Step S301: Use distributed web crawler technology to crawl multiple initial consumer review data of the target product. Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0104] Step S302: pre-process the multiple initial consumer review data to obtain multiple target consumer review data. Figure 2 Step S202 of the illustrated embodiment will not be described in detail here.

[0105] Step S303: Use the pre-trained large language model to perform topic identification on the multiple target consumer comment data to obtain the topics of the multiple target consumer comment data.

[0106] Specifically, the above step S303 includes:

[0107] Step S3031: For any target consumer comment data, use the pre-trained large language model to extract features from the target consumer comment data to obtain the topic feature vector of the target consumer comment data. Specifically, the large language model used in step S202 is fine-tuned through the above steps S203 to 207 to obtain a pre-trained large language model that can more accurately perform topic identification and sentiment analysis. Use the pre-trained large language model to extract features from any target consumer comment data through the following formula (5) to obtain a topic feature vector.

[0108] H t =Model LoRA (y t ) (5)

[0109] Where yt represents the tth target consumer comment data; H t Represents the topic feature vector of the tth target consumer review data; Model LoRA Represents a pre-trained large language model.

[0110] Step S3032, clustering the topic feature vectors of the target consumer review data to obtain multiple topics to be selected. Specifically, clustering methods such as K-means or hierarchical clustering are used to cluster the topic feature vectors through the following formula (6) to obtain multiple topics to be selected.

[0111] Cluster({H t})→{C1,C2,…,C k} (6)

[0112] Among them, H t Represents the topic feature vector; Cluster() represents the clustering function; C k represents the kth candidate topic.

[0113] Step S3033, determine the probability distribution of multiple topics to be selected, and use the topic with the highest probability as the topic of the target consumer comment data. Specifically, after subject identification of a target consumer comment data, three topics to be selected may be obtained: "taste", "price" and "packaging". At this time, the probability distribution of the three topics to be selected can be obtained by the following formula (7). Assuming that the probabilities of the three topics to be selected are {taste (0.8), price (0.1), packaging (0.1)}, "taste" corresponding to the highest probability of 0.8 is used as the topic of the target consumer comment data. By extracting and clustering thematic features of the target consumer comment data, not only can obvious topics be identified, but also potential and fine-grained topic classifications can be discovered, which improves the accuracy of topic identification and provides support for analyzing consumer satisfaction.

[0114] θ t,k =softmax(W model *H t +b) (7)

[0115] Among them, θ t,k represents the probability of the kth selected topic of the tth target consumer review data; H t represents the topic feature vector; W model represents the weight matrix of the pre-trained large language model; b represents the bias term of the pre-trained large language model.

[0116] Step S304, using a pre-trained large language model, sentiment analysis is performed based on the multiple target consumer comment data and the topics of the multiple target consumer comment data to obtain the sentiment tendency of the multiple target consumer comment data.

[0117] Specifically, the above step S304 includes:

[0118] Step S3041: For any target consumer comment data, a pre-trained large language model is used to extract features based on the target consumer comment data and the subject of the target consumer comment data to obtain a sentiment feature vector of the target consumer comment data. Specifically, the pre-trained large language model is used to obtain a sentiment feature vector by the following formula (8) when the subject of the target consumer comment data is known.

[0119] S t =Model LoRA (y t ,Topic t )(8)

[0120] Among them, y t represents the tth target consumer review data; S tRepresents the sentiment feature vector of the tth target consumer comment data; Topic t Represents the subject of the tth target consumer review data; Model LoRA Represents a pre-trained large language model.

[0121] Step S3042, based on the emotional feature vector of the target consumer comment data, determine the emotional category probability distribution of the target consumer comment data, and use the emotional category with the largest probability as the emotional tendency of the target consumer comment data. Specifically, the output layer of the pre-trained large language model uses the softmax function to generate the emotional category probability distribution through the following formula (9). Among them, the emotional categories include positive, negative and neutral. Assuming that the emotional category probability distribution is {positive (0.9), neutral (0.01), negative (0.01)}, the "positive" corresponding to the maximum probability of 0.9 is used as the emotional tendency of the target consumer comment data. By first performing topic identification and then performing sentiment analysis, and extracting the emotional feature vector in combination with the topic information, it is possible to capture more subtle emotional changes, and to make emotional judgments based on the understanding of the comment topic, thereby improving the accuracy of sentiment analysis and providing support for analyzing consumer satisfaction.

[0122] P t =softmax(W o *S t +b o )(9)

[0123] Among them, P t W represents the probability distribution of sentiment category of the tth target consumer comment data; o represents the weight matrix of the output layer; S t b represents the sentiment feature vector of the tth target consumer review data; o Represents the bias term of the output layer.

[0124] Step S305, using a graph convolutional network, performing satisfaction analysis based on multiple target consumer review data, themes and sentiment tendencies of the multiple target consumer review data, and obtaining satisfaction scores of the multiple target consumer review data for the target product.

[0125] Specifically, the above step S305 includes:

[0126] Step S3051, constructing a consumer review graph structure based on multiple target consumer review data, themes and sentiment tendencies of the multiple target consumer review data.

[0127] In some optional implementations, the above step S3051 includes:

[0128] Step a1, using a graph convolutional network, generates multiple nodes in a consumer review graph structure based on multiple target consumer review data. Specifically, each target consumer review data is used as a node in the consumer review graph structure.

[0129] Step a2, assign node features to multiple nodes based on multiple target consumer comment data, themes and sentiment tendencies of multiple target consumer comment data. Specifically, the node features of each node include the comment time and consumer IP in the target consumer comment data and themes and sentiment tendencies of the target consumer comment data, which can be expressed by the following formula (10).

[0130] R t =Concat(Sentiment t ,Topic t ,Time t ,Location t )(10)

[0131] Among them, R t Represents the node feature of the tth node; Sentiment t Indicates the sentiment tendency of the tth node; Topic t Indicates the topic of the tth node; Time t Indicates the comment time of the tth node; Location t Represents the consumer IP of the tth node.

[0132] Step a3, based on the similarity between multiple target consumer review data, construct edges between multiple nodes. Specifically, for any two nodes in the consumer review graph structure, the cosine similarity between the two nodes is calculated as the similarity of the two reviews by the following formula (11). If the similarity is greater than the preset similarity threshold, an edge is constructed between the two nodes, and the weight of the edge is the value of the similarity. The graph structure reveals the direct and indirect connections between reviews, helps understand which reviews discuss similar topics or share the same emotional attitudes, and provides a more intuitive and accurate way to analyze consumer satisfaction.

[0133]

[0134] Among them, R m represents the mth node; R n Represents the nth node; cosine_similarity(R m ,R n ) represents the similarity between the mth node and the nth node.

[0135] Step S3052, using a graph convolutional network, based on the consumer review graph structure, obtains satisfaction scores of multiple target consumer review data for the target product, and the graph convolutional network includes a multi-layer convolutional network and an output layer.

[0136] In some optional implementations, the above step S3052 includes:

[0137] Step b1, using a graph convolutional network, in each layer of the convolutional network, update the node features of each node in the consumer review graph structure. Specifically, the graph convolutional network (GCN) includes multiple layers of convolutional networks. In each layer of the convolutional network, for each node, the node features of the node are updated based on the neighbor nodes of the node through the following formula (12). By combining the features of the node itself with the features of its neighboring nodes, the information in the local network structure can be effectively captured, which helps to understand the relative position of each node in the graph structure and enhance the ability to understand complex graph structure data and generalization performance.

[0138] Among them, the neighbor node is the node that is directly connected to the node through an edge in the consumer review graph structure.

[0139]

[0140] in, represents the updated node feature of the vth node in the lth layer of the convolutional network; φ represents the activation function; u represents the neighbor node of the vth node; U represents the set of neighbor nodes of the vth node; W l-1 Represents the weight matrix of the l-1th layer convolutional network; R u represents the node characteristics of neighbor nodes; b l-1 represents the bias term of the l-1th layer of the convolutional network; c uv Represents the normalization coefficient.

[0141] Step b2, for any node, based on the node features updated by each layer of the convolutional network, obtain the global feature vector of the node. Specifically, for any node, after each layer of the graph convolutional network updates the node features through the above step b1, the node features of each layer of the node are aggregated using the average pooling operation through the following formula (13) to achieve global prediction analysis of the node based on the entire consumer review graph structure.

[0142]

[0143] Among them, v represents the vth node; represents the global eigenvector of the vth node; Represents the node features of the vth node after being updated in the lth layer of the convolutional network.

[0144] Step b3, using the output layer of the graph convolutional network, predicts based on the global feature vector of the node to obtain the satisfaction score of the target consumer comment data corresponding to the node. Specifically, using the output layer of the graph convolutional network, the global feature vector of any node is mapped to the softmax function through the following formula (14) to obtain the satisfaction score of the comment corresponding to the node. Optionally, the following Table 2 is an example of the satisfaction score corresponding to the target consumer comment data. The node feature update mechanism in each layer of the convolutional network can deeply explore the complex patterns in the comment content and structure, learn higher-level abstract features, aggregate the node features of each node after each layer of the convolutional network is updated, and form a feature vector from a global perspective, thereby determining the satisfaction score based on the global feature vector, making the consumer satisfaction prediction more comprehensive and accurate.

[0145]

[0146] Among them, v represents the vth node; Score v represents the satisfaction score of the vth node; W o represents the weight matrix of the output layer; represents the global eigenvector of the vth node; b o Represents the bias term of the output layer.

[0147] Table 2 Satisfaction scores of target consumer review data

[0148]

[0149] Step S306: for any target consumer review data, determine the expert satisfaction score of the target consumer review data. Specifically, a business expert scores each target consumer review data based on experience to obtain an expert satisfaction score.

[0150] Step S307, based on the difference between the expert satisfaction score and the satisfaction score of the multiple target consumer review data, the network parameters of the graph convolutional network are updated. Specifically, the loss between the expert satisfaction score of the target consumer review data and the satisfaction score predicted in the above step S305 is calculated by using the mean square error loss function through the following formula (15), so as to adjust the parameters of the graph convolutional network through the gradient descent algorithm to minimize the loss, thereby reducing the prediction error and improving the accuracy of the satisfaction prediction.

[0151]

[0152] Where Ω represents the mean square error loss; t represents the tth target consumer review data; T represents the number of target consumer review data; Score pScore represents the expert satisfaction score of the tth target consumer review data; t Represents the satisfaction score of the tth target consumer review data.

[0153] The embodiment of the present invention provides a consumer satisfaction analysis method based on a graph convolutional network. The method captures initial consumer review data in real time through distributed web crawler technology, can quickly understand market dynamics and the latest feedback from consumers, and then uses a pre-trained large language model to perform topic recognition and sentiment analysis on the pre-processed target consumer review data. It can deeply mine the implicit topics in the reviews and the real intentions of consumers, and can also carefully distinguish the emotional attitudes of consumers towards target products. Finally, the characteristics of the graph convolutional network are used to perform in-depth analysis and prediction of consumer satisfaction, which can accurately identify the correlation and similarity between reviews, and obtain satisfaction scores for target products, thereby improving data processing efficiency, achieving accurate prediction of consumer satisfaction, and providing a scientific basis for corporate decision-making.

[0154] In this embodiment, a consumer satisfaction analysis device based on a graph convolutional network is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0155] This embodiment provides a consumer satisfaction analysis device based on a graph convolutional network. Figure 4 As shown, including:

[0156] The crawling module 401 is used to crawl multiple initial consumer review data of the target product using distributed web crawler technology.

[0157] The preprocessing module 402 is used to preprocess the multiple pieces of initial consumer review data to obtain multiple pieces of target consumer review data.

[0158] The identification module 403 is used to use a pre-trained large language model to perform topic identification on multiple target consumer review data to obtain the topics of the multiple target consumer review data.

[0159] The analysis module 404 is used to use a pre-trained large language model to perform sentiment analysis based on multiple target consumer comment data and the topics of the multiple target consumer comment data to obtain the sentiment tendency of the multiple target consumer comment data.

[0160] The scoring module 405 is used to use a graph convolutional network to perform satisfaction analysis based on multiple target consumer review data, themes and sentiment tendencies of the multiple target consumer review data, and obtain satisfaction scores of the multiple target consumer review data for the target product.

[0161] In some optional implementations, the preprocessing module 402 includes:

[0162] The generating unit is used to generate a semantic embedding vector of any initial consumer review data by using a large language model.

[0163] The detection unit is used to perform noise detection on the semantic embedding vector of the initial consumer review data to obtain the noise score of the initial consumer review data.

[0164] The clearing unit is used to clear the initial consumer review data with noise scores greater than a preset noise threshold and the initial consumer review data with null values ​​from the multiple initial consumer review data to obtain multiple target consumer review data.

[0165] In some optional implementations, before the identification module 403, the device further includes:

[0166] The acquisition module is used to acquire a plurality of preprocessed sample text data, wherein any sample text data includes a topic label and a sentiment tendency label.

[0167] The adjustment module is used to introduce a dimensionality reduction matrix and a dimensionality increase matrix into the large language model for any sample text data, and adjust the weight matrix of the large language model.

[0168] The first determination module is used to perform topic identification and sentiment analysis on the sample text data based on the large language model after adjusting the weight matrix, so as to obtain the sample topic and sample sentiment tendency of the sample text data.

[0169] The first updating module is used to update the dimension reduction matrix and the dimension increase matrix based on the difference between the topic label of the sample text data and the sample topic and the difference between the sentiment tendency label and the sample sentiment tendency.

[0170] The second determination module is used to repeat the adjustment process of the weight matrix and the update process of the dimensionality reduction matrix and the dimensionality increase matrix until the preset end condition is reached, and the large language model after the last weight matrix adjustment is used as the pre-trained large language model.

[0171] In some optional implementations, the identification module 403 includes:

[0172] The first extraction unit is used to extract features of any target consumer comment data by using a pre-trained large language model to obtain a topic feature vector of the target consumer comment data.

[0173] The clustering unit is used to perform topic clustering on the topic feature vectors of the target consumer review data to obtain multiple topics to be selected.

[0174] The first determination unit is used to determine the probability distribution of multiple topics to be selected, and select the topic with the greatest probability as the topic of the target consumer review data.

[0175] In some optional implementations, the analysis module 404 includes:

[0176] The second extraction unit is used to use a pre-trained large language model to extract features based on the target consumer comment data and the subject of the target consumer comment data for any target consumer comment data, so as to obtain a sentiment feature vector of the target consumer comment data.

[0177] The second determination unit is used to determine the emotional category probability distribution of the target consumer comment data based on the emotional feature vector of the target consumer comment data, and take the emotional category with the largest probability as the emotional tendency of the target consumer comment data.

[0178] In some optional implementations, the scoring module 405 includes:

[0179] The construction unit is used to construct a consumer review graph structure based on multiple target consumer review data, topics and sentiment tendencies of the multiple target consumer review data.

[0180] The scoring unit is used to obtain satisfaction scores of multiple target consumer review data on the target product based on the consumer review graph structure by using a graph convolutional network.

[0181] In some optional embodiments, the building block includes:

[0182] The generating subunit is used to generate a plurality of nodes in a consumer review graph structure based on a plurality of target consumer review data by using a graph convolutional network.

[0183] The assigning subunit is used to assign node features to multiple nodes based on multiple target consumer review data, topics and sentiment tendencies of the multiple target consumer review data.

[0184] The construction subunit is used to construct edges between multiple nodes based on the similarities between multiple target consumer review data.

[0185] In some optional implementations, the scoring unit includes:

[0186] The updating subunit is used to adopt a graph convolutional network and update the node features of each node in the consumer review graph structure in each layer of the convolutional network.

[0187] A subunit is determined, which is used to obtain a global feature vector of any node based on the node features updated by each layer of the convolutional network.

[0188] The prediction subunit is used to use the output layer of the graph convolutional network to make predictions based on the global feature vector of the node to obtain the satisfaction score of the target consumer review data corresponding to the node.

[0189] In some optional embodiments, the device further comprises:

[0190] The third determination module is used to determine the expert satisfaction score of any target consumer review data.

[0191] The second updating module is used to update the network parameters of the graph convolutional network based on the difference between the expert satisfaction scores and the satisfaction scores of multiple target consumer review data.

[0192] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0193] The consumer satisfaction analysis device based on graph convolutional network in this embodiment is presented in the form of functional units, where the units refer to ASIC (Application Specific Integrated Circuit) circuits, processors and memories that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0194] The embodiment of the present invention also provides a computer device having the above Figure 4 The consumer satisfaction analysis device based on graph convolutional network is shown.

[0195] See also Figure 5 , Figure 5 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 5As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 5 A processor 10 is taken as an example.

[0196] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0197] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0198] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0199] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0200] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 5 The example of connecting through bus is taken in the following.

[0201] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0202] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0203] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0204] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A consumer satisfaction analysis method based on graph convolutional network, characterized in that: The method comprises: Use distributed web crawler technology to capture multiple initial consumer review data for target products; Preprocessing the plurality of initial consumer review data to obtain a plurality of target consumer review data; Using a pre-trained large language model, performing topic identification on the plurality of target consumer review data to obtain topics of the plurality of target consumer review data; Using a pre-trained large language model, sentiment analysis is performed based on the multiple target consumer review data and the topics of the multiple target consumer review data to obtain the sentiment tendency of the multiple target consumer review data; A graph convolutional network is used to perform satisfaction analysis based on the multiple target consumer review data, the topics and sentiment tendencies of the multiple target consumer review data, to obtain satisfaction scores of the multiple target consumer review data for the target product.

2. The method according to claim 1, characterized in that The preprocessing of the plurality of initial consumer review data to obtain a plurality of target consumer review data includes: For any piece of initial consumer review data, a large language model is used to generate a semantic embedding vector of the initial consumer review data; Performing noise detection on the semantic embedding vector of the initial consumer review data to obtain a noise score of the initial consumer review data; Initial consumer review data with noise scores greater than a preset noise threshold and initial consumer review data with null values ​​are removed from the plurality of initial consumer review data to obtain the plurality of target consumer review data.

3. The method according to claim 1, characterized in that Before using the pre-trained large language model to perform topic identification on the plurality of target consumer review data to obtain the topics of the plurality of target consumer review data, the method further includes: Acquire a plurality of preprocessed sample text data, wherein any sample text data includes a topic label and a sentiment tendency label; For any sample text data, a dimension reduction matrix and a dimension increase matrix are introduced into the large language model to adjust the weight matrix of the large language model; Based on the large language model after adjusting the weight matrix, performing topic identification and sentiment analysis on the sample text data to obtain a sample topic and a sample sentiment tendency of the sample text data; Based on the difference between the topic label of the sample text data and the sample topic and the difference between the sentiment label and the sample sentiment, the dimension reduction matrix and the dimension increase matrix are updated; Repeat the above-mentioned weight matrix adjustment process and the dimension reduction matrix and dimension increase matrix update process until the preset end condition is reached, and use the large language model after the last weight matrix adjustment as the pre-trained large language model.

4. The method according to claim 1, characterized in that The pre-trained large language model is used to perform topic identification on the plurality of target consumer review data to obtain the topics of the plurality of target consumer review data, including: For any target consumer review data, use a pre-trained large language model to extract features from the target consumer review data to obtain a topic feature vector of the target consumer review data; Performing topic clustering on the topic feature vectors of the target consumer review data to obtain a plurality of topics to be selected; The probability distribution of the multiple topics to be selected is determined, and the topic to be selected with the greatest probability is used as the topic of the target consumer review data.

5. The method according to claim 1, characterized in that The method of using a pre-trained large language model to perform sentiment analysis based on the multiple target consumer review data and the topics of the multiple target consumer review data to obtain the sentiment tendency of the multiple target consumer review data includes: For any target consumer comment data, a pre-trained large language model is used to extract features based on the target consumer comment data and the subject of the target consumer comment data to obtain a sentiment feature vector of the target consumer comment data; Based on the sentiment feature vector of the target consumer comment data, the sentiment category probability distribution of the target consumer comment data is determined, and the sentiment category with the greatest probability is taken as the sentiment tendency of the target consumer comment data.

6. The method according to claim 1, characterized in that The graph convolutional network is used to perform satisfaction analysis based on the multiple target consumer review data, the topics and sentiment tendencies of the multiple target consumer review data, and obtain satisfaction scores of the multiple target consumer review data for the target product, including: Constructing a consumer review graph structure based on the multiple target consumer review data, the topics and sentiment tendencies of the multiple target consumer review data; A graph convolutional network is used to obtain satisfaction scores of the target product from the multiple target consumer review data based on the consumer review graph structure.

7. The method according to claim 6, characterized in that The constructing a consumer review graph structure based on the multiple target consumer review data, the topics and sentiment tendencies of the multiple target consumer review data includes: Using a graph convolutional network, generating a plurality of nodes in the consumer review graph structure based on the plurality of target consumer review data; Assigning node features to the plurality of nodes based on the plurality of target consumer review data, the topics and sentiment tendencies of the plurality of target consumer review data; Based on the similarities between the multiple target consumer review data, edges are constructed between the multiple nodes.

8. The method according to claim 7, characterized in that The graph convolution network includes a multi-layer convolution network and an output layer. The graph convolution network is used to obtain satisfaction scores of the target product based on the consumer review graph structure, including: Using a graph convolutional network, in each layer of the convolutional network, updating the node features of each node in the consumer review graph structure; For any node, based on the node features updated by each layer of convolutional network, a global feature vector of the node is obtained; The output layer of the graph convolutional network is used to make predictions based on the global feature vector of the node to obtain the satisfaction score of the target consumer review data corresponding to the node.

9. The method according to claim 8, characterized in that The method further comprises: For any target consumer review data, determine the expert satisfaction score of the target consumer review data; Based on the difference between the expert satisfaction scores and the satisfaction scores of the multiple target consumer review data, the network parameters of the graph convolutional network are updated.

10. A consumer satisfaction analysis device based on graph convolutional network, characterized in that: The device comprises: A crawling module is used to crawl multiple initial consumer review data of target products using distributed web crawler technology; A preprocessing module, used for preprocessing the plurality of initial consumer review data to obtain a plurality of target consumer review data; an identification module, configured to use a pre-trained large language model to perform topic identification on the plurality of target consumer review data to obtain topics of the plurality of target consumer review data; An analysis module, configured to use a pre-trained large language model to perform sentiment analysis based on the plurality of target consumer review data and the topics of the plurality of target consumer review data, so as to obtain the sentiment tendency of the plurality of target consumer review data; The scoring module is used to use a graph convolutional network to perform satisfaction analysis based on the multiple target consumer review data, the topics and sentiment tendencies of the multiple target consumer review data, and obtain satisfaction scores of the multiple target consumer review data for the target product.

Images