Method for obtaining keywords through operator perception and combination of large model semantic analysis

Abstract

The invention discloses a method for obtaining keywords by operator perception in combination with large model semantic analysis, which comprises the following steps of: carrying out semantic extraction on an operator service chain through a ChatGPT large model to obtain a first related keyword, embedding the ChatGPT large model by adopting an error algorithm, carrying out secondary semantic extraction by utilizing the ChatGPT large model embedded with the error algorithm, and obtaining a second related keyword. Second-time related keywords are obtained, the keywords of the two times are compared, if the keywords are the same, the keywords are accurate, if the keywords are output, if the keywords are different, the keywords are inaccurate, and the keywords of the first time are corrected; the first keyword is substituted into a two-step spelling correction Detector-Corrector model for correction, the corrected first keyword and the corrected second keyword are compared again, if the first keyword and the second keyword are the same, the keyword is output, if the first keyword and the second keyword are different, output is abandoned, and semantic extraction is carried out again to obtain a new keyword. According to the invention, the keyword related to the demand can be accurately obtained from the service chain of the operator for analysis.

Description

Technical Field

[0001] The present invention relates to a keyword acquisition method, and in particular to a method for acquiring keywords by operator perception combined with large model semantic analysis. Background Art

[0002] Currently, network operators are generally faced with the problem of how to avoid being channelized and how to maintain their advantages in the competition with smart terminal manufacturers. The operator business chain is a series of business activities and links involved in the service provision process of telecom operators, including but not limited to the construction and maintenance of network infrastructure, customer service, marketing, billing and account management, user data management, business development and innovation, etc. Telecom operators use this business chain to ensure the smooth operation of their services, meet user needs, and maintain market competitiveness and profitability.

[0003] In this case, it is necessary to consider how to mine, perceive and analyze the massive data of the operator's business chain, including user behavior, device status, network traffic and other related data, to obtain keywords related to market dynamics and changes in user needs, so as to analyze and confirm changes in user needs and discover potential patterns and rules to improve the competitiveness of operators.

[0004] Therefore, there is an urgent need for a method that perceives operators and obtains keywords in combination with large model semantic analysis to solve the problem of how to obtain demand-related keywords from the operator's business chain. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a method for acquiring keywords by operator perception and combining large model semantic analysis, so as to solve the problem of how to acquire demand-related keywords from the operator's service chain.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A method for acquiring keywords by operator perception and combining with large model semantic analysis, characterized by comprising the following steps: The operator's business chain is semantically extracted through the ChatGPT large model to obtain the first relevant keywords. The error algorithm is used to embed the ChatGPT large model. The ChatGPT large model with the embedded error algorithm is used for secondary semantic extraction to obtain the second relevant keywords. The two keywords are compared. If the keywords are the same, the keywords are accurate and the keywords are output. If the keywords are different, the keywords are inaccurate and the first keywords are corrected. The first keywords are substituted into the two-step spelling correction Detector-Corrector model for correction. The corrected first keywords are compared with the second keywords again. If they are the same, the keywords are output. If they are different, the output is abandoned and semantic extraction is performed again to obtain new keywords.

[0007] To optimize the above technical solutions, the specific measures taken also include: Furthermore, the semantic extraction of the operator's service chain by using the ChatGPT large model to obtain the first relevant keywords includes the following steps: Prepare training data with business chain-related keyword labels, and use this data to train the ChatGPT big model to fine-tune the ChatGPT big model. Use the fine-tuned ChatGPT big model to perform semantic extraction on text data in the operator's business chain, which includes identifying key entities, relationships, events, or emotions in the text, and extracting the first keywords related to the operator's business chain needs.

[0008] Furthermore, the semantic extraction of text data in the operator service chain includes identifying key entities, relationships, events or emotions in the text, including the following steps: When performing semantic extraction, the NER function of ChatGPT is applied to extract entities with specific meanings from the text, including company names, product names, equipment names, and service types; the relationship extraction capability of ChatGPT is used to extract the relationship and status between entities from the text, and extract relevant keywords; ChatGPT is used to identify event information described in the text, including the launch of new services, marketing activities, and user complaints, and keywords are extracted from them; ChatGPT is used for sentiment analysis, and by analyzing the sentiment tendencies in user behavior, including the sentiment tendencies in user comments and feedback, the needs of users are indirectly identified, thereby extracting relevant keywords.

[0009] Furthermore, the error algorithm is embedded in the ChatGPT large model, comprising the following steps: Prepare training data with business chain-related keyword labels, and use this data to train the ChatGPT large model. Then, manually verify and mark the keywords output by the ChatGPT large model as correct or wrong for the ChatGPT large model training and learning, and collect and analyze the error information that exists when marking errors occur. Based on the collected error information, use the error algorithm to adjust and optimize the parameters or structure of the ChatGPT large model to obtain the ChatGPT large model embedded with the error algorithm. After each collection of new error information, adjust the parameters or structure of the ChatGPT large model again according to the new error information to iteratively update the ChatGPT large model.

[0010] Furthermore, based on the collected error information, the error algorithm is used to adjust and optimize the parameters or structure of the ChatGPT large model, including the following steps: Based on the collected error information, the error information is classified by type, including grammatical errors, semantic understanding errors, and knowledge gaps. The error frequency of each type is quantified, and the root cause of each error type is analyzed. According to the error analysis results, specific optimization goals are set. According to the impact and frequency of the errors, priorities are set for different optimization goals. Through grid search, random search, or Bayesian optimization methods, the hyperparameters of the ChatGPT large model are adjusted. The complexity and generalization ability of the model are balanced by adjusting the regularization term or weight attenuation.

[0011] Furthermore, the step of substituting the first keyword into the two-step spelling correction Detector-Corrector model for correction includes the following steps: Input the first keyword and use the pre-trained Detector-Corrector model to scan the input keyword. The Detector module in the Detector-Corrector model uses a sequence labeling method to identify each character or word, determine whether it has a spelling error, and label it correct or incorrectly, and output the labeling result of each character or word in the keyword; input the data output by the Detector module into the Corrector module in the Detector-Corrector model, and the Corrector module analyzes the characters or words marked as errors. The Corrector model uses a language model to predict its possible correct form. The language model predicts the correct vocabulary of the masked part based on context information. According to the prediction result, the candidate word with the highest confidence is selected as the correction result, and the corrected keyword is output.

[0012] Furthermore, before the semantic extraction of the operator service chain by the ChatGPT big model, the following step is also included: optimizing the construction of the operator service chain by optimizing the physical topology and application service topology of the SIB network.

[0013] Furthermore, the optimization of the construction of the operator service chain by optimizing the physical topology and application service topology of the SIB network includes the following steps: Remove unnecessary network devices and links, simplify the network structure, adopt hierarchical design, realize functional separation and simplify management, implement redundant design on key nodes and links, select the best transmission path through routing optimization algorithm, reduce the number of transmission hops and delay of data packets, and use SDN technology to realize dynamic scheduling and routing optimization of network traffic to optimize the physical topology of SIB network; Complex application services are split into multiple independent modules, each of which is responsible for specific functions. Communication and collaboration between modules are achieved through service interfaces. A load balancer is deployed at the application service layer to reasonably distribute user requests to multiple servers and evenly distribute traffic to optimize the application service topology of the SIB network.

[0014] Furthermore, a computer-readable storage medium stores a computer program, characterized in that: the computer program enables a computer to execute the above-mentioned method of acquiring keywords by operator perception and combining large model semantic analysis.

[0015] Furthermore, an electronic device is characterized in that it includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a method for acquiring keywords by operator perception and combining large model semantic analysis as described above is implemented.

[0016] The beneficial effects of the present invention are: The present invention uses the ChatGPT big model to perform semantic extraction on the operator's business chain, which can accurately capture and parse the semantic information of each link of the business chain, avoiding the limitations of traditional methods in parsing complex contexts. The ChatGPT big model can quickly extract key information related to the operator's business chain from massive data and improve the efficiency of information processing. The ChatGPT big model can not only extract information on the surface of the text, but also understand its deep semantic meaning, which enables it to perform deeper intelligent analysis on the data in the operator's business chain and dig out hidden business insights and potential risks; At the same time, the ChatGPT large model with embedded error algorithm is used for secondary semantic extraction. Through the adjustment of embedded error algorithm, the ChatGPT large model can more finely capture the fine-grained information in the text, such as subtle differences in context, implicit semantic associations, etc., to obtain more refined semantic understanding and improve the accuracy of information extraction. The inaccurate keywords are then input into the Detector-Corrector model for correction, reducing the risk of missing keywords in the final output due to incorrect information in the first keyword semantic extraction by the ChatGPT large model, or incorrect information in the keywords extracted twice, thereby ensuring the accuracy of this method in obtaining demand-related keywords from the operator's business chain. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of the overall process of a method proposed by the present invention for acquiring keywords by operator perception and combining large model semantic analysis; Figure 2 The figure is a schematic diagram of the correction processing flow of a method proposed by the present invention for acquiring keywords by operator perception combined with large model semantic analysis. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0019] As attached Figure 1 and attached Figure 2 As shown, a method for acquiring keywords by operator perception and combining large model semantic analysis in an embodiment of the present invention includes the following steps: The operator's business chain is semantically extracted through the ChatGPT large model to obtain the first relevant keywords, and the error algorithm is used to embed the ChatGPT large model. The ChatGPT large model with the embedded error algorithm is used for secondary semantic extraction to obtain the second relevant keywords. The two keywords are compared. If the keywords are the same, the keywords are accurate and the keywords are output. If the keywords are different, the keywords are inaccurate and the first keywords are corrected. The first keywords are substituted into the two-step spelling correction Detector-Corrector model for correction, and the corrected first keywords are compared with the second keywords. If they are the same, the keywords are output. If they are different, the output is abandoned and semantic extraction is performed again to obtain new keywords.

[0020] The present invention uses the ChatGPT big model to perform semantic extraction on the operator's business chain, which can accurately capture and parse the semantic information of each link of the business chain, avoiding the limitations of traditional methods in parsing complex contexts. The ChatGPT big model can quickly extract key information related to the operator's business chain from massive data, and improve the efficiency of information processing. The ChatGPT big model can not only extract information on the surface of the text, but also understand its deep semantic meaning, which enables it to perform a deeper intelligent analysis of the data in the operator's business chain and dig out hidden business insights and potential risks; at the same time, the ChatGPT big model with embedded error algorithm is used for secondary semantic extraction. By adjusting the embedded error algorithm, the ChatGPT big model The model can more finely capture fine-grained information in the text, such as subtle differences in context, implicit semantic associations, etc., to obtain more refined semantic understanding and improve the accuracy of information extraction; then the inaccurate keywords are input into the Detector-Corrector model for correction, reducing the risk of missing the final output keyword accuracy due to the presence of erroneous information in the first keyword extracted by the ChatGPT large model semantics, or erroneous information in both extracted keywords, thereby ensuring the accuracy of this method in obtaining demand-related keywords from the operator's business chain; finally, when comparing the first and second keywords after correction, if they are still different, the output is abandoned, which can effectively avoid obtaining wrong keywords and interfering with subsequent analysis. The keywords obtained by this solution can be used to analyze and confirm changes in user needs, discover potential operating models and rules to improve the competitiveness of operators.

[0021] In this embodiment, the semantic extraction of the operator service chain through the ChatGPT large model to obtain the first related keywords includes the following steps: Select a suitable ChatGPT big model version, such as GPT-3.5 or higher, which has strong natural language processing and semantic understanding capabilities; prepare training data with business chain-related keyword labels, and use these data to train the ChatGPT big model to fine-tune the ChatGPT big model, and use the fine-tuned ChatGPT big model to perform semantic extraction on text data in the operator's business chain, which includes identifying key entities, relationships, events, or emotions in the text, and extracting the first keywords related to the operator's business chain needs.

[0022] The above-mentioned semantic extraction of text data in the operator's service chain includes identifying key entities, relationships, events or emotions in the text, including the following steps: When performing semantic extraction, the NER function of ChatGPT is applied to extract entities with specific meanings from the text. Entities include company names, product names, equipment names, service types, etc. These entities are often demand-related keywords; the relationship extraction capability of ChatGPT is used to extract the relationships and states between entities from the text. These relationships can help us understand business needs more deeply and extract relevant keywords; ChatGPT is used to identify event information described in the text. Event information includes new business launches, marketing activities, user complaints, etc. These events are often closely related to business needs, and keywords are extracted from them; ChatGPT is used for sentiment analysis. By analyzing the sentiment tendencies in user behavior, including the sentiment tendencies in user comments and feedback, the demand points that users are concerned about are indirectly identified, thereby extracting relevant keywords.

[0023] In this embodiment, the above error algorithm is embedded in the ChatGPT large model, including the following steps: Prepare training data with business chain-related keyword labels, and use this data to train the ChatGPT large model. Then, manually verify and mark the keywords output by the ChatGPT large model as correct or wrong for the ChatGPT large model training and learning, and collect and analyze the error information that exists when marking errors occur. Based on the collected error information, use the error algorithm to adjust and optimize the parameters or structure of the ChatGPT large model to obtain the ChatGPT large model embedded with the error algorithm. After each collection of new error information, adjust the parameters or structure of the ChatGPT large model again according to the new error information to iteratively update the ChatGPT large model.

[0024] Among them, based on the collected error information, an error algorithm is used to adjust and optimize the parameters or structure of the ChatGPT large model, including the following steps: Based on the collected error information, the error information is classified by type, including grammatical errors, semantic understanding errors, knowledge gaps, etc. The error frequency of each type is quantified, and the root cause of each error type is analyzed, such as in which types of data or contexts the model performs poorly. According to the error analysis results, specific optimization goals are set, such as reducing the error rate of a specific type and improving overall performance indicators such as BLEU, ROUGE, Perplexity, etc. According to the impact and frequency of the errors, priorities are set for different optimization goals. Through grid search, random search or Bayesian optimization methods, the hyperparameters of the ChatGPT large model, such as learning rate, batch size, number of iterations, etc., are adjusted to observe the impact on the error rate. The complexity and generalization ability of the model are balanced by adjusting the regularization term or weight decay to reduce overfitting.

[0025] In this embodiment, the above-mentioned step of substituting the first keyword into the two-step spelling correction Detector-Corrector model for correction includes the following steps: Input the first keyword and use the pre-trained Detector-Corrector model to scan the input keyword. The Detector-Corrector model is usually based on deep learning technology and is trained with a large amount of text data. It can identify spelling errors in the text. The Detector module in the Detector-Corrector model uses a sequence labeling method to identify each character or word, determine whether it has a spelling error, and mark the correct or incorrect result. The labeling result can be binary, such as 0 for correct and 1 for error, or a more complex label system to distinguish different types of errors. The labeling result of each character or word in the keyword is output to mark possible spelling errors; the data output by the Detector module is input into the Corrector module in the Detector-Corrector model, and the Corrector module analyzes the characters or words marked as errors. The Corrector model uses a language model, such as Masked Language Model, MLM, to predict its possible correct form. The language model predicts the correct vocabulary of the masked part based on context information. According to the prediction result, the candidate word with the highest confidence is selected as the correction result, and the corrected keyword is output.

[0026] In this embodiment, before the semantic extraction of the operator service chain through the ChatGPT big model, the following steps are also included: by optimizing the physical topology and application service topology of the SIB network, the construction of the operator service chain is optimized to improve the semantic extraction effect.

[0027] The above-mentioned optimization of the construction of the operator service chain by optimizing the physical topology and application service topology of the SIB network includes the following steps: Remove unnecessary network devices and links, simplify network structure, reduce network complexity and potential failure points, adopt hierarchical design, divide the network into core layer, aggregation layer and access layer, realize functional separation and simplify management, implement redundancy design on key nodes and links, such as dual-machine hot standby and link aggregation, to improve network reliability and availability, select the best transmission path through routing optimization algorithms, such as shortest path first SPF, OSPF, etc., reduce the number of transmission hops and delay of data packets, use SDN software-defined network technology to realize dynamic scheduling and routing optimization of network traffic, improve network flexibility and performance, and optimize the physical topology of SIB network; Complex application services are split into multiple independent modules, each of which is responsible for specific functions. Communication and collaboration between modules are achieved through service interfaces to improve the maintainability and scalability of services. Load balancers are deployed at the application service layer to reasonably distribute user requests to multiple servers for balanced traffic distribution. The distribution strategy is dynamically adjusted according to the server load and performance to ensure service stability and efficiency, thereby optimizing the application service topology of the SIB network.

[0028] In another embodiment, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the above-mentioned method of acquiring keywords by operator perception and combining large model semantic analysis.

[0029] In another embodiment, the present invention proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a method for acquiring keywords by operator perception and combining large model semantic analysis as described above is implemented.

[0030] In the embodiments disclosed in the present application, the computer storage medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or apparatus. The computer storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. More specific examples of computer storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0031] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0032] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A method for obtaining keywords by operator perception and combining large model semantic analysis, characterized in that: The steps include: The operator's business chain is semantically extracted through the ChatGPT large model to obtain the first relevant keywords. The error algorithm is used to embed the ChatGPT large model. The ChatGPT large model with the embedded error algorithm is used for secondary semantic extraction to obtain the second relevant keywords. The two keywords are compared. If the keywords are the same, the keywords are accurate and the keywords are output. If the keywords are different, the keywords are inaccurate and the first keywords are corrected. The first keywords are substituted into the two-step spelling correction Detector-Corrector model for correction. The corrected first keywords are compared with the second keywords again. If they are the same, the keywords are output. If they are different, the output is abandoned and semantic extraction is performed again to obtain new keywords.

2. The method for acquiring keywords by operator perception and combining large model semantic analysis according to claim 1, characterized in that: The method of performing semantic extraction on the operator service chain through the ChatGPT large model to obtain the first relevant keywords includes the following steps: Prepare training data with business chain-related keyword labels, and use this data to train the ChatGPT big model to fine-tune the ChatGPT big model. Use the fine-tuned ChatGPT big model to perform semantic extraction on text data in the operator's business chain, which includes identifying key entities, relationships, events, or emotions in the text, and extracting the first keywords related to the operator's business chain needs.

3. The method for acquiring keywords by operator perception and combining large model semantic analysis according to claim 2, characterized in that: The semantic extraction of text data in the operator service chain includes identifying key entities, relationships, events or emotions in the text, including the following steps: When performing semantic extraction, the NER function of ChatGPT is applied to extract entities with specific meanings from the text, including company names, product names, equipment names, and service types; the relationship extraction capability of ChatGPT is used to extract the relationship and status between entities from the text, and extract relevant keywords; ChatGPT is used to identify event information described in the text, including the launch of new services, marketing activities, and user complaints, and keywords are extracted from them; ChatGPT is used for sentiment analysis, and by analyzing the sentiment tendencies in user behavior, including the sentiment tendencies in user comments and feedback, the needs of users are indirectly identified, thereby extracting relevant keywords.

4. The method for acquiring keywords by operator perception and combining large model semantic analysis according to claim 1, characterized in that: The error algorithm is embedded in the ChatGPT large model, including the following steps: Prepare training data with business chain-related keyword labels, and use these data to train the ChatGPT large model. Then manually check and mark the keywords output by the ChatGPT large model as correct or wrong for ChatGPT large model training and learning, and collect and analyze the error information when marking errors occur. Based on the collected error information, use the error algorithm to adjust and optimize the parameters or structure of the ChatGPT large model to obtain the ChatGPT large model embedded with the error algorithm. After each new error information is collected, the parameters or structure of the ChatGPT large model are adjusted again according to the new error information to perform iterative updates of the ChatGPT large model.

5. The method for acquiring keywords by operator perception and combining large model semantic analysis according to claim 4, characterized in that: Based on the collected error information, the error algorithm is used to adjust and optimize the parameters or structure of the ChatGPT large model, including the following steps: Based on the collected error information, the error information is classified by type, including grammatical errors, semantic understanding errors, and knowledge gaps. The error frequency of each type is quantified, and the root cause of each error type is analyzed. According to the error analysis results, specific optimization goals are set. According to the impact and frequency of the errors, priorities are set for different optimization goals. Through grid search, random search, or Bayesian optimization methods, the hyperparameters of the ChatGPT large model are adjusted. The complexity and generalization ability of the model are balanced by adjusting the regularization term or weight attenuation.

6. The method for acquiring keywords by operator perception and combining large model semantic analysis according to claim 1, characterized in that: Substituting the first keyword into the two-step spelling correction Detector-Corrector model for correction includes the following steps: Input the first keyword and use the pre-trained Detector-Corrector model to scan the input keyword. The Detector module in the Detector-Corrector model uses a sequence labeling method to identify each character or word, determine whether it has a spelling error, and label it correct or incorrectly, and output the labeling result of each character or word in the keyword; input the data output by the Detector module into the Corrector module in the Detector-Corrector model, and the Corrector module analyzes the characters or words marked as errors. The Corrector model uses a language model to predict its possible correct form. The language model predicts the correct vocabulary of the masked part based on context information. According to the prediction result, the candidate word with the highest confidence is selected as the correction result, and the corrected keyword is output.

7. The method for acquiring keywords by operator perception and combining large model semantic analysis according to claim 1, characterized in that: Before the semantic extraction of the operator service chain is performed through the ChatGPT large model, the following steps are also included: optimizing the construction of the operator service chain by optimizing the physical topology and application service topology of the SIB network.

8. The method for acquiring keywords by operator perception and combining large model semantic analysis according to claim 7, characterized in that: The optimization of the construction of the operator service chain by optimizing the physical topology and application service topology of the SIB network includes the following steps: Remove unnecessary network devices and links, simplify the network structure, adopt hierarchical design, realize functional separation and simplify management, implement redundant design on key nodes and links, select the best transmission path through routing optimization algorithm, reduce the number of transmission hops and delay of data packets, and use SDN technology to realize dynamic scheduling and routing optimization of network traffic to optimize the physical topology of SIB network; Complex application services are split into multiple independent modules, each of which is responsible for specific functions. Communication and collaboration between modules are achieved through service interfaces. A load balancer is deployed at the application service layer to reasonably distribute user requests to multiple servers and evenly distribute traffic to optimize the application service topology of the SIB network.

9. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables the computer to execute the method for acquiring keywords by operator perception and combining large model semantic analysis as described in any one of claims 1 to 8.

10. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a method for acquiring keywords by operator perception and combining large model semantic analysis as described in any one of claims 1 to 8 is implemented.

Images