Information processing method, device and system, electronic equipment and readable medium

Abstract

The embodiment of the invention discloses an information processing method, device and system, electronic equipment and a readable medium. The method comprises the following steps: acquiring to-be-processed audio information; converting the audio information into text information based on a first large language model; obtaining a user consultation association degree and a product improvement association degree of the dialogue text based on a second large language model, and further extracting a user consultation dialogue text set and a product improvement dialogue text set; processing the user consultation dialogue text set based on a third language model to obtain a first intelligent customer service training material, and processing the product improvement dialogue text set to obtain a first product function improvement list; respectively processing the first intelligent customer service training material and the first product function improvement list by using a regular expression to obtain a second intelligent customer service training material and a second product function improvement list; outputting the second intelligent customer service training material to the intelligent customer service system; and outputting the second product function improvement list to a product iteration system.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of natural language processing, and particularly relates to an information processing method, device, system, electronic equipment and readable medium. BACKGROUND

[0002] In the modern customer service industry, customer service recording as an important data source, its analysis and utilization are crucial to the operation and product iteration of enterprises. In the history of customer service recording analysis, true semantic understanding has always been a challenge. Existing technical analysis methods often only focus on superficial text information, rather than deep meaning and emotion. In terms of analysis methods, traditional customer service recording analysis mainly relies on manual work, and customer service representatives need to spend a lot of time playing back and summarizing the recording, which greatly reduces their efficiency in handling user inquiries, and is prone to omissions and biases. At the same time, due to the limitations of manual filling, existing customer service operations can only record the feedback of part of the users, and the valuable information hidden in most of the recording data is ignored. This leads to the fact that the user feedback provided by the customer service department to the product design department is based on partial or even one-sided information, which affects the efficiency and timeliness of product iteration. SUMMARY

[0003] The embodiments of the present disclosure provide an information processing method, device, system, electronic equipment and readable medium, which effectively identify user emotions by using large language models, improve the accuracy of two types of topic classification of user inquiries and product improvement, and based on topic classification, extract multi-dimensional information contained in customer service recordings, and use regular expressions to improve the accuracy of information extraction, which can provide more comprehensive and in-depth user feedback for the product design department, and also provide a large amount of business data for the training of intelligent customer service.

[0004] To achieve the above technical purposes, the embodiments of the present disclosure adopt the following technical solutions:

[0005] In a first aspect, the embodiments of the present disclosure provide an information processing method, which comprises:

[0006] obtaining audio information to be processed, wherein the audio information comprises one or more dialogue audios;

[0007] converting the audio information into text information based on a first large language model, wherein the text information comprises one or more dialogue texts corresponding to the dialogue audios one by one;

[0008] obtaining user inquiry relevance and product improvement relevance of the dialogue texts based on a second large language model, and further extracting a user inquiry dialogue text set and a product improvement dialogue text set;

[0009] Based on the third language model, user consultation dialogue text sets are processed to obtain the first intelligent customer service training material, and product improvement dialogue text sets are processed to obtain the first product feature improvement list.

[0010] Using regular expressions, process the first intelligent customer service training materials and the first product function improvement list respectively to obtain the second intelligent customer service training materials and the second product function improvement list.

[0011] Output the second intelligent customer service training materials to the intelligent customer service system; output the second product function improvement list to the product iteration system.

[0012] In some possible implementations, the first, second, and third language models are obtained by fine-tuning a general large model, and the three can be the same or different.

[0013] In some possible implementations, the user inquiry relevance and product improvement relevance of the dialogue text are obtained based on the second largest language model, including:

[0014] Input the first set of prompt words into the second language model and process the dialogue text. Calculate the weighted sum of the correlation coefficients of the user consultation correlation factors in the model output to obtain the user consultation correlation degree. Calculate the weighted sum of the correlation coefficients of the product improvement correlation factors in the model output to obtain the product improvement correlation degree.

[0015] In some possible implementations, the user inquiry relevance and product improvement relevance of the dialogue text are obtained based on the second largest language model, including:

[0016] Multiple sets of prompt words are input into the second language model and the dialogue text is processed. The weighted sum of the correlation coefficients of the user consultation correlation factors in the output results of each set of prompt words is calculated, and the user consultation correlation degree is obtained by further weighting. The weighted sum of the correlation coefficients of the product improvement correlation factors in the output results of each set of prompt words is calculated, and the product improvement correlation degree is obtained by further weighting.

[0017] In some possible implementations, the text sets of user inquiry dialogues and product improvement dialogues are extracted, including:

[0018] Filter dialogue texts with a user consultation relevance greater than the consultation relevance threshold, and generate a set of user consultation dialogue texts;

[0019] Filter dialogue texts with a product improvement relevance greater than the improvement relevance threshold, and generate a product improvement dialogue text set.

[0020] In some possible implementations, the user inquiry dialogue text set is processed based on the third largest language model to obtain the first intelligent customer service training material, and the product improvement dialogue text set is processed to obtain the first product feature improvement list, including:

[0021] The customer service training prompts are input into the third language model to tag the user consultation dialogue text set and extract customer service guidance data to obtain the first intelligent customer service training material, which includes a list of product question tags.

[0022] Input the product improvement prompts into the third language model to tag the product improvement dialogue text set and obtain the first product feature improvement list, in which the product improvement prompts include the product feature tag list.

[0023] In some possible implementations, the method further includes:

[0024] Based on the corrections made to the second and first intelligent customer service training materials, and / or the corrections made to the second and first product function improvement lists,

[0025] Adjust the customer service training prompts and / or product improvement prompts.

[0026] In some possible implementations, the second intelligent customer service training material is output to the intelligent customer service system; the second product feature improvement list is output to the product iteration system, including:

[0027] The second intelligent customer service training materials are filtered and then output to the intelligent customer service system; the second product function improvement list is sorted and then output to the product iteration system.

[0028] Secondly, embodiments of this disclosure provide an information processing apparatus, the apparatus comprising:

[0029] An information acquisition unit is configured to acquire audio information to be processed, wherein the audio information includes one or more dialogue audios;

[0030] The model processing unit is configured to convert audio information into text information based on the first major language model, wherein the text information includes one or more dialogue texts that correspond one-to-one with the dialogue audio; obtain the user consultation relevance and product improvement relevance of the dialogue texts based on the second major language model, and further extract the user consultation dialogue text set and the product improvement dialogue text set; process the user consultation dialogue text set based on the third major language model to obtain the first intelligent customer service training material, and process the product improvement dialogue text set to obtain the first product function improvement list;

[0031] The information verification unit is configured to use regular expressions to process the first intelligent customer service training material and the first product function improvement list respectively, and obtain the second intelligent customer service training material and the second product function improvement list.

[0032] The result output unit is configured to output the second intelligent customer service training material to the intelligent customer service system and output the second product function improvement list to the product iteration system.

[0033] Thirdly, embodiments of this application provide an information processing system, which includes the information processing device, intelligent customer service system, and product iteration system as described in the second aspect.

[0034] Fourthly, embodiments of this application provide an electronic device, including: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in the first aspect.

[0035] Fifthly, embodiments of this application provide a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect.

[0036] The technical solution of the first aspect provided by the embodiments of this disclosure brings at least the following beneficial effects: acquiring audio information to be processed; converting audio information into text information based on a first large language model; acquiring user consultation relevance and product improvement relevance of the dialogue text based on a second large language model, and further extracting user consultation dialogue text sets and product improvement dialogue text sets; processing the user consultation dialogue text sets based on a third large language model to obtain first intelligent customer service training materials, and processing the product improvement dialogue text sets to obtain a first product function improvement list; processing the first intelligent customer service training materials and the first product function improvement list respectively using regular expressions to obtain second intelligent customer service training materials and a second product function improvement list; outputting the second intelligent customer service training materials to the intelligent customer service system; and outputting the second product function improvement list to the product iteration system. The embodiments of this disclosure effectively identify user emotions using a large language model, improve the accuracy of the two topic classifications of user consultation and product improvement, and extract multi-dimensional information contained in customer service recordings based on topic classification, and improve the accuracy of information extraction using regular expressions. This can provide product design departments with more comprehensive and in-depth user feedback, and can also provide a large amount of business data for intelligent customer service training, realizing accurate, efficient, and multi-dimensional utilization of customer service information.

[0037] It should be noted that the technical effects of any of the implementation methods in aspects two through five can be found in the technical effects of the corresponding implementation methods in aspect one, and will not be repeated here.

[0038] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Clearly, the drawings described below only relate to some embodiments of this disclosure and are not intended to limit the scope of this disclosure.

[0040] Figure 1 A schematic diagram of an information processing system configuration according to at least one embodiment of the present disclosure is shown;

[0041] Figure 2 A schematic diagram illustrating the main steps of an information processing method according to at least one embodiment of the present disclosure is shown.

[0042] Figure 3 A flowchart of an information processing method according to at least one embodiment of the present disclosure is shown;

[0043] Figure 4 A schematic diagram of the main modules of an information processing apparatus according to at least one embodiment of the present disclosure is shown;

[0044] Figure 5 A schematic diagram of an electronic device according to at least one embodiment of the present disclosure is shown;

[0045] Figure 6 A schematic diagram of a computer-readable medium according to at least one embodiment of the present disclosure is shown. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the described embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0047] In the following text, any methods, apparatus, examples, and contents that do not fully correspond to the scope of the claims are not derived from the present invention. Such methods, apparatus, examples, and contents, as well as all subsequent descriptions, are for illustrative purposes only, or to highlight specific aspects or features of the claims.

[0048] Note that the examples described below are merely specific examples and are not intended to limit the embodiments of this disclosure to the specific shapes, hardware, connections, operations, values, conditions, data, sequences, etc., shown and described. Those skilled in the art can utilize the concepts of this disclosure to construct further embodiments not mentioned herein by reading this specification.

[0049] The terminology used in this disclosure is that which is currently widely used in the art in consideration of the functionality of this disclosure; however, these terms may vary depending on the intent, precedent, or new technology of those skilled in the art. Furthermore, specific terms may be chosen by the applicant, and in such cases, their detailed meanings will be described in the detailed description of this disclosure. Therefore, the terminology used in this specification should not be construed as simple names, but rather based on the meaning of the terms and the overall description of this disclosure.

[0050] To better understand the embodiments of this disclosure, the relevant terms involved in this disclosure will first be defined and explained.

[0051] Large language models refer to machine learning models with a large number of parameters and complex computational structures. These models are usually built from deep neural networks, such as the Transformer architecture, and have billions or even hundreds of billions of parameters. This complex computational structure provides the foundation for the model to process large-scale data and complex tasks, enabling effective feature extraction, transformation, and generation of input data, thereby achieving various tasks such as natural language processing, computer vision, and speech recognition.

[0052] Regular expressions are a tool for describing string patterns. It is a formal language composed of ordinary characters (such as letters and numbers) and special characters (called metacharacters). These characters are combined according to specific syntax rules to match, find, and replace strings that conform to a certain pattern.

[0053] It should be noted that the technical solutions in this disclosure, including the collection, updating, analysis, processing, use, transmission, and storage of user personal information, all comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken to prevent unauthorized access to user personal information data and to safeguard user personal information security, network security, and national security.

[0054] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0055] In the modern customer service industry, customer service recordings serve as an important data source. Common shallow analysis methods include:

[0056] Keyword matching: This is the most basic and common method. It captures information from the recorded text based on a predefined list of keywords. For example, if a user mentions "dissatisfaction" during a call, the system will flag the call as negative feedback. However, this method cannot capture more complex situations, such as "I wasn't very happy with the previous version, but the current update is great."

[0057] Pattern matching: This method is slightly more complex, capturing information based on predefined sentence patterns. For example, a pattern like "I hope to have" can be defined to capture a user's needs. However, this method still cannot understand more complex semantic structures and emotions.

[0058] Simple statistical methods, such as word frequency statistics, focus only on the frequency of word occurrences in a text, rather than the relationships between them or their underlying meanings.

[0059] Sentiment analysis: Early sentiment analysis methods often relied on simple dictionaries or rules, such as lists of positive and negative words. This approach failed to accurately capture the subtle emotions within a text, especially in complex contexts.

[0060] Due to the limitations of the methods described above, true semantic understanding and deep sentiment analysis have remained largely unresolved issues. This has resulted in the inability to effectively classify and utilize the rich information in customer service recordings. For example, when a customer service conversation reveals a user's clearly negative sentiment regarding product features, the conversation is generally related to product improvement and can serve as input for product iteration systems; similarly, when a customer service conversation includes inquiries about product usage, it can serve as input for intelligent customer service systems. Therefore, because existing technologies cannot accurately classify conversation content, a large amount of valuable information is overlooked, while also increasing the workload and complexity of manual analysis. Specifically, the following problems exist:

[0061] Impact on agent efficiency: Due to the lack of automated recording and text analysis tools, traditional customer service recording analysis mainly relies on manual work. Customer service agents need to spend a lot of time playing back and summarizing the recordings, which greatly reduces their efficiency in handling user inquiries and is prone to omissions and biases. The limitations of manual analysis are even more obvious when faced with a large amount of recording data.

[0062] Limitations of Product Feedback: Due to the inefficiency of manual data entry, existing customer service operations often only record feedback from a portion of users, while valuable information hidden within most of the recorded data is overlooked. This results in customer service providing product design with user feedback based on partial or even incomplete information.

[0063] The training cost of intelligent customer service is high: While the application of intelligent customer service is becoming increasingly widespread, its training still requires a large amount of data and resources. Currently, the cost of acquiring this training data is relatively high, and the quality and coverage of the data are also problematic. Enterprises often need to invest significant additional resources to meet the training data requirements for intelligent customer service.

[0064] In summary, effectively analyzing and utilizing customer service recordings to extract valuable information while reducing related costs has become a pressing issue for the customer service industry. This disclosure utilizes a large language model to effectively identify user emotions, improving the accuracy of two main categories: user inquiries and product improvements. Based on these categories, it extracts multi-dimensional information from customer service recordings and uses regular expressions to further enhance the accuracy of information extraction. This provides product design departments with more comprehensive and in-depth user feedback and also offers a wealth of business data for training intelligent customer service systems, achieving accurate, efficient, and multi-dimensional utilization of customer service information.

[0065] Figure 1 A schematic diagram of an information processing system 100 according to at least one embodiment of the present disclosure is shown. The system 100 includes a user 110, a customer service representative 120, a large language model 130, an intelligent customer service system 140, a product iteration system 150, and a regular expression 160. The user 110 and the customer service representative 120 communicate and generate audio dialogue. The dialogue content can include various types of content such as complaints, business inquiries, and business processing. The customer service representative 120 can be a human customer service representative or an intelligent customer service representative; this disclosure does not limit this. One or more audio dialogues constitute audio information to be processed. The audio information is input into the large language model 130, where speech recognition, topic classification, and text processing are performed. The output of the large language model 130 is then subjected to quality control, verification, and key information extraction using the regular expression 160. The results are then output to the intelligent customer service system 140 and the product iteration system 150, respectively.

[0066] Figure 3A flowchart of an information processing method 300 according to at least one embodiment of the present disclosure is shown. The method is applied to... Figure 1 The information processing system 100 shown includes the following specific steps:

[0067] Step S301: Obtain audio information to be processed, wherein the audio information includes one or more dialogue audios. User 110 and customer service 120 communicate and generate dialogue audio; one or more dialogue audios constitute the audio information to be processed.

[0068] Step S302: Based on the first large language model, the audio information is converted into text information, which includes one or more dialogue texts corresponding one-to-one with the dialogue audio. During the transcription process, to improve transcription quality, a batch of recordings can be manually transcribed according to common customer service call types (complaints, business inquiries, business processing, etc.) and used as training data to fine-tune the general large model. Fine-tuning can significantly improve the probability of correctly transcribing common business vocabulary (product function names, product usage methods, precautions, etc.) and terms (customer service language, etc.) in customer service dialogues. Optionally, to improve the large model's ability to recognize key information from different dialects, dialect information of common business vocabulary can be used as training data to fine-tune the general large model.

[0069] It is understood that the audio information adopts a format supported by the first major language model, such as common audio formats like WAV and MP3, and the output text information can also adopt a format commonly used in this field, such as TXT, DOC, and JSON, without limitation in this disclosure. It is understood that the output text information format can be supported by the second major language model to continue subsequent steps; if the second major language model does not support the format of the text information, the text information can be formatted before being input into the second major language model.

[0070] Step S303: Obtain the user inquiry relevance and product improvement relevance of the dialogue text based on the second major language model, and further extract the user inquiry dialogue text set and the product improvement dialogue text set. Optionally, the second major language model can be fine-tuned using business-related data as training data based on the general pre-trained model. The fine-tuned second major language model is more suitable for the analysis task of customer service call recordings. The first and second major language models can be the same or different. Both user inquiry relevance and product improvement relevance are used to characterize the topic classification of a dialogue text and its corresponding audio. Specifically, user inquiry relevance characterizes whether the audio segment belongs to the topic of customer inquiry about product usage, or the degree of relevance between the dialogue and the topic of customer inquiry about product usage (hereinafter referred to as CIPU, Customer Inquiry about Product Usage); product improvement relevance characterizes whether the audio segment expresses negative emotions of the user regarding product functions, or the degree of relevance between the dialogue and the topic of product function improvement based on negative user emotions (hereinafter referred to as NEPF, Negative Emotion of Product Function).

[0071] Optionally, the first set of prompts can be input into the second language model to process the dialogue text. For example, the first set of prompts might include "Did the customer service representative instruct the user on how to operate the system?", and the output of the second language model might be, for example, "The customer service representative instructed the user on how to resolve the issue of not receiving the SMS verification code, such as restarting the phone or moving to a location with better signal." Based on this output, it can be determined that the customer service audio message is related to the user's inquiry about how to use the product (CIPU). In this case, for example, the user inquiry relevance can be set to 1; otherwise, it can be set to 0. The dialogue text with a user inquiry relevance of 1 is extracted from the text information to obtain the user inquiry dialogue text set. Similarly, the first set of prompts can also include, for example, "Does the user have any dissatisfaction with the system's functions and offer suggestions?", and the output of the second language model might be, for example, "The user is dissatisfied with the system's functions; he expressed dissatisfaction that the system only provides SMS verification codes as a confirmation method; he suggests that the system add other confirmation methods." Based on the output, it can be determined that the customer service audio message is related to the NEPF (Neutral for Product Improvement) theme, which describes the user's negative emotions regarding product features. In this case, for example, the relevance score for product improvement is set to 1; otherwise, it is set to 0. The dialogue text with a relevance score of 1 for product improvement is extracted from the text information to obtain the product improvement dialogue text set.

[0072] Understandably, large language models may experience illusions due to their limited understanding of the accessed data. Using different question formats for the same content can help identify inaccurate or inconsistent analysis results when automating the processing of large amounts of recorded text, thus mitigating the impact of large model illusions. For example, the first set of prompts can include multiple questions for different topics. For instance, for the topic CIPU (User Inquiry about Product Usage), the relevance of user inquiries is represented by UCF (User Inquiry Relevance); for the topic NEPF (User Negative Emotions about Product Features), the relevance of product improvement is represented by PIF (Product Improvement Relevance). For the two topics CIPU and NEPF, the first set of prompts can include M and N prompts respectively to determine whether a dialogue text is relevant to CIPU or NEPF. Understandably, the M and N prompts can be the same or different, and each prompt can have its own weighting factor (e.g., W) for different topics (such as CIPU or NEPF). CIPU (i) or W NEPF (j)) The weighting factor is used to characterize the association between different prompt words and the topic. After inputting the first set of prompt words into the second language model and processing the dialogue text, the association coefficients of the prompt words for different topics can be obtained. By calculating the weighted sum of the association coefficients, the degree of association can be further obtained. That is, the degree of association of user consultation is obtained by calculating the weighted sum of the association coefficients of the user consultation association factors in the output of the second language model, and the degree of association of product improvement is obtained by calculating the weighted sum of the association coefficients of the product improvement association factors in the output of the second language model.

[0073] Taking the CIPU topic as an example, M prompts are set, where the weight factor corresponding to the i-th prompt is W. CIPU (i) Input the M prompt words into the second large language model to obtain the output results of each prompt word, that is, the correlation coefficient between the dialogue text and the CIPU topic based on the i-th prompt word, using CC CIPU (i) indicates that the user consultation relevance (UCF) can be obtained by weighting the relevance coefficients, i.e.:

[0074]

[0075] Similarly, for NEPF topics, N prompt words are set, where the weight factor corresponding to the j-th prompt word is W. NEPF (j). Input N prompt words into the second large language model to obtain the output results of each prompt word, that is, the correlation coefficient between the dialogue text and the NEPF topic based on the j-th prompt word, using CC. NEPF (j) represents the product improvement correlation factor (PIF). Therefore, the product improvement correlation factor (PIF) can be obtained by weighting the correlation coefficients, i.e.:

[0076]

[0077] Because large language models are more prone to illusions when processing long texts, the relevance obtained from a single set of prompt words may still be erroneous. Therefore, it is advisable to process the same dialogue text multiple times using the large language model, each time with a different combination of prompt words. This helps to further reduce illusions during the processing. Specifically, multiple sets of prompt words are input into a second large language model and processed to analyze the dialogue text. The weighted sum of the correlation coefficients of the user consultation relevance factors in the model outputs of each set of prompt words is calculated, and this weighted sum is used to obtain the user consultation relevance. Similarly, the weighted sum of the correlation coefficients of the product improvement relevance factors in the model outputs of each set of prompt words is calculated, and this weighted sum is used to obtain the product improvement relevance.

[0078] Taking the CIPU topic as an example, K cue word groups are set, and M cue words are set in each group, where the weight factor corresponding to the i-th cue word is W. CIPU (i,k). Input the M prompt words into the second largest language model to obtain the output results for each prompt word, that is, the correlation coefficient between the dialogue text and the CIPU topic based on the i-th prompt word, using CC... CIPU (i,k) represents the combination of different prompt words; different combinations of prompt words can also be assigned the same or different weights as needed, using WT. CIPU (k) indicates that, therefore, the User Consultation Relevance (UCF) can be calculated using the following formula:

[0079]

[0080] Similarly, for the NEPF topic, K suggestion word groups are set, and N suggestions words are set in each group, where the weight factor corresponding to the j-th suggestion word is W. NEPF (j,k). Input N prompt words into the second largest language model to obtain the output result for each prompt word, which is the correlation coefficient between the dialogue text and the NEPF topic based on the j-th prompt word, expressed as CC. NEPF (j,k) represents the combination of different prompt words; different combinations of prompt words can also be assigned the same or different weights as needed, using WT. NEPF (k) indicates that, therefore, the Product Improvement Relevance (PIF) can be calculated using the following formula:

[0081]

[0082] Optionally, for the different methods of calculating User Consultation Relevance (UCF) and Product Improvement Relevance (PIF) mentioned above, thresholds can be further set to determine the criteria for filtering dialogue texts. That is, dialogue texts with a UCF greater than the consultation relevance threshold are filtered to generate a set of user consultation dialogue texts; dialogue texts with a PIF greater than the improvement relevance threshold are filtered to generate a set of product improvement dialogue texts.

[0083] Understandably, if a feature receives a large number of inquiries about its usage despite no negative user feedback, it should be adjusted and optimized. Therefore, alternatively, a user inquiry threshold could be set. Within the CIPU topic, if the number of dialogue texts inquiring about a particular product feature exceeds this threshold within a certain timeframe, related inquiry texts would be added to the product improvement dialogue text set. It's important to note that only comparison texts not already included in the original product improvement dialogue text set should be added; for example, deduplication of dialogue texts within the product improvement dialogue text set is necessary.

[0084] Step S304 involves processing the user inquiry dialogue text set based on the third major language model to obtain the first intelligent customer service training material, and processing the product improvement dialogue text set to obtain the first product feature improvement list. Specifically, the third major language model is obtained by fine-tuning the general pre-trained model using business-related data as training data. The fine-tuned third major language model is more suitable for the analysis task of customer service call recordings. The first, second, and third major language models can all be obtained by fine-tuning the general model, and the three can be the same or different.

[0085] Building upon the topic classification using the second language model, a third language model can be further utilized to extract key information from user consultation dialogue text sets and product improvement dialogue text sets for more effective input into subsequent systems. The output format of the third language model can optionally be JSON. To ensure more effective retrieval, sorting, and application of relevant information in subsequent systems, product issue tag lists and product function tag lists can be pre-set as input prompts for the third language model, achieving structured large-scale model output. Specifically, customer service training prompts are input into the third language model to tag the user consultation dialogue text set and extract customer service guidance data to obtain the first intelligent customer service training material, where the customer service training prompts include the product issue tag list. Product improvement prompts are input into the third language model to tag the product improvement dialogue text set to obtain the first product function improvement list, where the product improvement prompts include the product function tag list.

[0086] Step S305: Regular expressions are used to process the first intelligent customer service training materials and the first product function improvement list respectively to obtain the second intelligent customer service training materials and the second product function improvement list. This embodiment of the disclosure uses regular expressions to process the output of the third large language model, its functions including quality control, verification, and extraction of key information. Specifically, the large language model has certain illusions. For example, the input prompt words include a list of product problem tags, but the output results sometimes do not fall within the scope of the product problem tag list, and sometimes do not use a standard output format, such as JSON format. Therefore, it is necessary to further process the first intelligent customer service training materials and the first product function improvement list using regular expressions so that the obtained second intelligent customer service training materials and second product function improvement lists meet the input requirements of the intelligent customer service system 140 and the product iteration system 150 in terms of both content and format. Therefore, regular expressions can achieve the purpose of quality control.

[0087] Optionally, by comparing the correction content of the second intelligent customer service training material with that of the first intelligent customer service training material, and / or the correction content of the second product function improvement list with that of the first product function improvement list, the customer service training prompts and / or product improvement prompts can be adjusted to ensure that the output results of the third language model are more accurate.

[0088] Optionally, based on the output of the large model, regular expressions can be used to further parse the text and extract key information elements. In particular, for highly structured information, using regular expressions for matching and extraction can further improve accuracy. This is suitable for extracting information with fixed formats such as phone numbers, order numbers, and dates.

[0089] Step S306: Output the second intelligent customer service training material to the intelligent customer service system; output the second product function improvement list to the product iteration system. Specifically, the audio information from the communication between user 110 and customer service 120 contains multi-dimensional information. After topic classification, it can be used in multiple dimensions through large language models and regular expressions. First, in the training of the intelligent customer service system 140, a rich training corpus can be formed by organizing the user questions and customer service personnel's answers. This data can be used to train and optimize the intelligent customer service system, enabling it to more accurately understand user questions and provide effective solutions, thereby improving the service quality of automated customer service. At the same time, by organizing common questions and answers, the enterprise's knowledge base can be enriched, providing reference for customer service personnel and the intelligent customer service system, and improving the overall service level. Second, in terms of product iteration and user feedback analysis, the collected user suggestions and complaints regarding product functions will be statistically analyzed to identify common problems and needs, providing direct user feedback to the product development and design teams. These real user voices can serve as a reference for product improvement and function optimization, ensuring that product iteration meets user needs. Through in-depth user research, businesses can gain a better understanding of user expectations and behavioral habits, providing support for marketing strategies and user experience design.

[0090] Optionally, since the original audio information may contain sensitive information (such as personal identification information, inappropriate words and phrases, etc.), and during the processing of the large language model, some inappropriate words and phrases may be added due to illusion (such as inappropriate replies or promises to users). Therefore, the training material for the second intelligent customer service can be further filtered before being output to the intelligent customer service system.

[0091] Optionally, the output of the second product feature improvement list can be sorted and output to the product iteration system, which can provide product development and design teams with more direct and clear product improvement suggestions.

[0092] Figure 2 A schematic diagram illustrating the main steps of an information processing method 200 according to at least one embodiment of the present disclosure is shown. The specific steps are as follows:

[0093] Step S201 involves fine-tuning the large language model 130. A batch of recordings can be selected according to common types of customer service calls (complaints, business inquiries, business processing, etc.) and transcribed manually as training data. Through fine-tuning, the probability of correctly transcribing common business terms (product function names, product usage methods, precautions, etc.) and phrases (customer service language, etc.) from customer service calls can be greatly improved.

[0094] Step S202: The customer service recording 210 is transcribed using the fine-tuned large language model 130 to generate a customer service file 220. An example of the generated customer service text is as follows:

[0095] Customer Service Representative: Hello, thank you for calling xx Payment Customer Service Center. I am a customer service representative, employee number XXXXX. How can I help you?

[0096] User: Hello! I just tried to make a large transfer in your system, but I haven't received the SMS verification code! What's going on?

[0097] Customer Service: I'm so sorry to hear you're having trouble. To better assist you, could you please provide your username and registered mobile phone number? I'll help you check.

[0098] User: (Provides username and mobile phone number) I think your system should provide other confirmation methods. SMS verification codes are too unreliable!

[0099] Customer service representative: I understand your concerns and thank you for your valuable suggestion. We will record your feedback and forward it to the relevant department. Also, could you try restarting your phone or moving to a location with better signal and trying again? This usually resolves the issue of not receiving the verification code.

[0100] User: I've tried everything, but I still haven't received it! Seriously!

[0101] Customer Service Representative: I completely understand how you feel right now, and I'm sorry for the inconvenience. Please check if the text message was mistakenly identified as spam, or if your phone has SMS blocking enabled. You can also try using the SIM card in another phone. We will record your situation and forward your suggestions.

[0102] User: Okay, I'll give it a try. You should also consider adding other confirmation methods!

[0103] Customer Service Representative: Thank you so much for your suggestion. We will definitely record your feedback and experience and provide feedback. We are constantly striving to improve our service. Is there anything else I can help you with?

[0104] User: That's all.

[0105] Customer Service: Okay, please feel free to contact us if you have any questions or concerns in the future. We apologize again for any inconvenience caused, and thank you for your understanding and patience. Have a wonderful day!

[0106] User: Okay, thank you.

[0107] Customer service representative: Goodbye, and best of luck.

[0108] Step S203 involves fine-tuning the large language model 130 to better perform the topic classification of customer service texts. Fine-tuning includes common business terms (product function names, product usage methods, precautions, etc.), language (customer service terms, etc.), and may also include indexing results from manual topic classification of different customer service texts.

[0109] Step S204: Classify the customer service text 220 into topics using the fine-tuned large language model. Input prompt words to ask whether the customer service text 220 relates to a specific topic. For example, a single question can be used to directly determine relevance to a topic; alternatively, a questionnaire-style question list can be designed to reduce the potential impact of the large language model's illusions on the accuracy and consistency of the evaluation results. An example of the first set of prompt words is as follows:

[0110] 1. Did the user mention any specific functions of the system?

[0111] 2. Does the user have negative emotions?

[0112] 3. Did the user exhibit intense emotions?

[0113] 4. Are users dissatisfied with the system's functionality?

[0114] 5. Did the user mention the system's X1 function?

[0115] 6. Did the user mention the system's X2 feature?

[0116] 7. Did the user mention the system's X3 feature?

[0117] 8. Did the user mention the system's X4 function?

[0118] 9. Did customer service personnel instruct users on how to operate the system?

[0119] 10. Did the user discuss the operation of the system with customer service personnel?

[0120] 11. Did customer service personnel respond to the user's questions?

[0121] 12. Did the customer service personnel resolve the user's problem?

[0122] 13. Have users provided any suggestions regarding system functionality?

[0123] 14. What system improvement suggestions did users make?

[0124] 15. How do customer service personnel guide users through the operation?

[0125] 16. What were the main contents of this communication?

[0126] The first set of prompts above relates to CIPU and / or NEPF, with each question having a different weighting factor associated with each topic. In the example above, for the CIPU topic, the weighting factor is as follows:

[0127] [0.1,0,0,0,0.05,0.05,0.05,0.05,0.2,0.2,0.05,0.05,0,0,0.2,0];

[0128] For NEPF topics, the weighting factors are as follows:

[0129] [0.1,0.1,0.1,0.1,0.05,0.05,0.05,0.05,0,0,0.05,0.05,0.1,0.1,0,0.1].

[0130] It is understandable that the above prompts can be set separately for different topics, or, as shown in this example, the prompts for different topics can be set together.

[0131] The output of the large language model can determine the correlation coefficient. Then, by using the correlation coefficient and weighting factors, the correlation degree between customer service text 220 and each topic can be calculated, and customer service texts with correlation degrees exceeding the correlation threshold can be filtered out.

[0132] The output for the above problem is, for example:

[0133] 1. The user did not mention the specific functions of the system.

[0134] 2. Users do indeed have negative emotions.

[0135] 3. The user displayed strong emotions because they said, "I'm very angry right now!"

[0136] 4. The user is dissatisfied with the system's functionality, expressing dissatisfaction that the system only provides SMS verification codes as a confirmation method.

[0137] 5. The user did not mention the system's X1 function.

[0138] 6. The user did not mention the system's X2 function.

[0139] 7. The user did not mention the system's X3 function.

[0140] 8. The user did not mention the system's X4 function.

[0141] 9. Customer service personnel guided users on how to resolve the issue of not receiving SMS verification codes, such as restarting the phone or moving to a location with better signal.

[0142] 10. The user did discuss the operation of the system with customer service personnel.

[0143] 11. Customer service personnel responded to the user's question and offered possible solutions and suggestions.

[0144] 12. The customer service staff did their best to resolve the user's problem and proposed several possible solutions, but we have no way of knowing whether the user ultimately resolved the problem.

[0145] 13. Users have made suggestions regarding system functions, such as adding other confirmation methods.

[0146] 14. The user suggested that the system add other confirmation methods because he felt that the existing SMS verification code was unreliable.

[0147] 15. Customer service personnel will guide users to restart their phones or move to a location with better signal to check if the text messages were mistakenly identified as spam or if the phone has text message blocking enabled.

[0148] 16. The main content of this communication was the user's issue of not receiving the SMS verification code. The user expressed dissatisfaction and suggested adding alternative confirmation methods to the system. Meanwhile, customer service personnel answered the user's questions and apologized for the inconvenience caused.

[0149] In the example above, for the CIPU topic, the output correlation coefficient is as follows:

[0150] [0,0,0,0,0,0,0,0,1,1,1,1,0,0,1,0];

[0151] For NEPF topics, the output correlation coefficients are as follows:

[0152] [0,1,1,1,0,0,0,0,0,0,1,1,1,1,0,1].

[0153] Therefore, UCF = 0.7 and PIF = 0.7 can be calculated. Assuming that both the consultation relevance threshold and the improvement relevance threshold are set to 0.6, this customer service text 220 involves both the topics CIPU and NEPF. Therefore, this customer service text 220 can be extracted into the user consultation dialogue text set and the product improvement dialogue text set, respectively.

[0154] To reduce the illusion of output from large language models in cases such as long texts, multiple sets of cue words from different perspectives can be set for calculation to improve the accuracy of topic classification. For example, in terms of representing emotions, negative emotions can include various angles such as anger, disappointment, complaint, anxiety, sadness, frustration, impatience, doubt, and threat; conversely, positive emotion cue angles can also be set, such as satisfaction, gratitude, thanks, trust, peace of mind, expectation, support, and surprise.

[0155] In step S205, the large language model is fine-tuned based on the information extraction requirements of the intelligent customer service system and the product iteration system.

[0156] Step S206: For the user inquiry dialogue text set, input customer service training prompts into the large language model, and output the user's product usage questions and customer service guidance, i.e., the first intelligent customer service training material. The large language model performs structured tagging on the customer service text, for example: determining a list of product problem tags, including: product quality issues, APP experience issues, logistics issues, etc., and tagging the text according to this list for easy retrieval and use.

[0157] Step S207 involves generating keyword A using regular expressions and performing quality control and verification on the first intelligent customer service training material. For example, the large model might not output according to the input label categories, or the output format might not meet requirements, such as not being in JSON format. Standardized second intelligent customer service training material can then be generated using regular expressions.

[0158] Step S208: Output the second intelligent customer service training material to the intelligent customer service system 140.

[0159] Step S209: For the product improvement dialogue text set, input product improvement prompts into the large language model and output the first product feature improvement list. The large language model then performs structured tagging of the customer service text according to the product feature tag list.

[0160] Step S210: Generate keyword B using regular expressions and perform quality control and verification on the first product feature improvement list. For example: the large model does not output according to the input label categories, or the output format does not meet the requirements, such as not being in JSON format. A standardized second product feature improvement list can be generated using regular expressions.

[0161] Step S211: Output the second product feature improvement list to the product iteration system 150.

[0162] Figure 4 A schematic diagram of the main modules of an information processing apparatus 400 according to at least one embodiment of the present disclosure is shown. The apparatus includes:

[0163] The information acquisition unit 410 is configured to acquire audio information to be processed, wherein the audio information includes one or more dialogue audios;

[0164] The model processing unit 420 is configured to convert audio information into text information based on the first major language model, wherein the text information includes one or more dialogue texts that correspond one-to-one with the dialogue audio; obtain the user consultation relevance and product improvement relevance of the dialogue texts based on the second major language model, and further extract the user consultation dialogue text set and the product improvement dialogue text set; process the user consultation dialogue text set based on the third major language model to obtain the first intelligent customer service training material, and process the product improvement dialogue text set to obtain the first product function improvement list;

[0165] Information verification unit 430 is configured to use regular expressions to process the first intelligent customer service training material and the first product function improvement list respectively to obtain the second intelligent customer service training material and the second product function improvement list.

[0166] The result output unit 440 is configured to output the second intelligent customer service training material to the intelligent customer service system and output the second product function improvement list to the product iteration system.

[0167] The embodiments of this disclosure utilize a large language model to effectively identify user emotions, improve the accuracy of two topic classifications: user inquiries and product improvements, and extract multi-dimensional information contained in customer service recordings based on topic classification. Regular expressions are used to improve the accuracy of information extraction, which can provide product design departments with more comprehensive and in-depth user feedback, and can also provide a large amount of business data for training intelligent customer service, realizing accurate, efficient and multi-dimensional utilization of customer service information.

[0168] It should be noted that the above application scenarios are merely exemplary, intended to describe one or more aspects of this disclosure in specific scenarios. However, these aspects are not essential, and various modifications can be made to the application scenario. It is readily understood that the specific application scenarios described in this disclosure are not limited.

[0169] At least some embodiments of this disclosure also provide an electronic device. Figure 5 A schematic diagram of an electronic device 500 according to at least one embodiment of the present disclosure is shown.

[0170] like Figure 5As shown, the electronic device 500 includes one or more processors 510 and a memory 520. The memory 520 includes one or more computer program modules 521. The one or more computer program modules 521 are stored in the memory 520 and are executed by the processor 510. These computer program modules 521 include instructions for executing an information processing method and its additional aspects according to at least one embodiment of the present disclosure. When executed by the processor 510, they can perform one or more steps of the information processing method and its additional aspects according to at least one embodiment of the present disclosure. The memory 520 and the processor 510 can be interconnected via a bus system and / or other forms of connection mechanisms (not shown). For example, the bus can be a Peripheral Component Interconnect Standard (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0171] For example, processor 510 may be a central processing unit (CPU), a digital signal processor (DSP), or other processing unit with data processing and / or program execution capabilities, such as a field-programmable gate array (FPGA). Processor 510 may be a general-purpose processor or a special-purpose processor, capable of controlling other components in electronic device 500 to perform desired functions.

[0172] Exemplarily, memory 520 may include any combination of one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, erasable programmable read-only memory (EPROM), portable compact disc read-only memory (CD-ROM), USB memory, flash memory, etc. One or more computer program modules 521 may be stored on the computer-readable storage medium, and processor 510 may run one or more computer program modules 521 to implement various functions of electronic device 500. The computer program modules include multiple computer-executable instructions. Various application programs and various data, as well as various data used and / or generated by the application programs, may also be stored in the computer-readable storage medium.

[0173] For example, electronic device 500 may also include input devices such as touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, and gyroscopes; output devices such as liquid crystal displays, speakers, and vibrators; storage devices such as magnetic tapes and hard disks (HDDs or SDDs); and communication devices such as network interface cards like LAN cards and modems. The communication devices allow electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data and perform communication processing via networks such as the Internet. A drive is connected to the I / O interface as needed. Removable storage media, such as disks, optical disks, magneto-optical disks, and semiconductor memories, are installed on the drive as needed so that computer programs read from them can be installed into the storage device as required.

[0174] For example, the electronic device 500 may further include a peripheral interface (not shown in the figure). This peripheral interface can be various types of interfaces, such as a USB interface, a Lightning interface, etc. The communication device can communicate wirelessly with networks and other devices, such as the Internet, intranets and / or wireless networks such as cellular telephone networks, wireless local area networks (LANs) and / or metropolitan area networks (MANs). Wireless communication can use any of a variety of communication standards, protocols, and technologies, including but not limited to Global System for Mobile Communications (GSM), Enhanced Data GSM Environment (EDGE), Wideband Code Division Multiple Access (W-CDMA), Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Bluetooth, Wi-Fi (e.g., based on IEEE 802.11a, IEEE 802.11b, IEEE 802.11g, and / or IEEE 802.11n standards), Voice over Internet Protocol (VoIP), Wi-MAX, protocols for email, instant messaging, and / or Short Message Service (SMS), or any other suitable communication protocol.

[0175] The electronic device 500 may be, for example, a system-on-a-chip (SOC) or a device including the SOC. For instance, it can be any device such as a mobile phone, tablet computer, laptop computer, e-reader, game console, television, digital photo frame, navigator, home appliance, communication base station, industrial controller, server, etc., or any combination of data processing devices and hardware. The embodiments of this disclosure do not limit this. The specific functions and technical effects of the electronic device 500 can be found in the foregoing description of the information processing method and its additional aspects according to at least one embodiment of this disclosure, and will not be repeated here.

[0176] Figure 6 A schematic diagram of a readable storage medium 600 according to at least one embodiment of the present disclosure is shown.

[0177] like Figure 6 As shown, a computer program 610 is stored on a readable storage medium 600, which is a computer-readable storage medium. When the computer program 610 is executed by a processor, it performs one or more steps of the information processing method and its additional aspects described above.

[0178] For example, when the program code is read by a computer, the computer can execute the program code stored in the computer storage medium to perform one or more steps of, for example, the information processing method and its additional aspects according to at least one embodiment of the present disclosure.

[0179] For example, the readable storage medium may include a memory card of a smartphone, a storage component of a tablet computer, a hard disk of a personal computer, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), portable compact disc read-only memory (CD-ROM), flash memory, and other readable storage media or any combination thereof. The readable storage medium 600 may be a non-transitory readable storage medium.

[0180] At least some of the embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other.

[0181] It should be noted that, in this disclosure, relational terms such as "first," "second," etc., are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0182] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved; that is, the preceding or following operations are not necessarily executed precisely in sequence. Instead, various steps may be processed in reverse order or simultaneously as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0183] The units described in the embodiments of this disclosure can be implemented in software or hardware. The described units can also be located in a processor. The names of these units do not, in some cases, constitute a limitation on the unit itself.

[0184] The following points should be noted regarding this disclosure:

[0185] (1) The accompanying drawings of the embodiments of this disclosure only involve the structures involved in the embodiments of this disclosure. Other structures can be referred to the general design.

[0186] (2) Where there is no conflict, the embodiments of this disclosure and the features in the embodiments can be combined with each other to obtain new embodiments.

[0187] The above are merely exemplary embodiments of this disclosure and are not intended to limit the scope of protection of this disclosure, which is determined by the appended claims.

Claims

1. An information processing method characterized by comprising: The method comprises: acquiring audio information to be processed, wherein the audio information comprises one or more dialogue audios; translating the audio information into text information based on a first large language model, wherein the text information comprises one or more dialogue texts corresponding to the dialogue audios; obtaining user consultation relevance and product improvement relevance of the dialogue texts based on a second large language model, and further extracting a user consultation dialogue text set and a product improvement dialogue text set; processing the user consultation dialogue text set based on a third large language model to obtain first intelligent customer service training materials, and processing the product improvement dialogue text set to obtain a first product function improvement list; processing the first intelligent customer service training materials and the first product function improvement list by using regular expressions respectively to obtain second intelligent customer service training materials and a second product function improvement list; outputting the second intelligent customer service training materials to an intelligent customer service system; and outputting the second product function improvement list to the product iteration system.

2. The method of claim 1, wherein, The first large language model, the second large language model, and the third large language model are obtained by fine-tuning a general large model, and the three can be the same or different.

3. The method of claim 1, wherein, The method for obtaining the user consultation relevance and the product improvement relevance of the dialogue texts based on the second large language model comprises: inputting a first set of prompt words into the second large language model and processing the dialogue texts to calculate a relevance sum weighted by relevance coefficients of user consultation association factors in model output results to obtain user consultation relevance, and to calculate a relevance sum weighted by relevance coefficients of product improvement association factors in model output results to obtain product improvement relevance.

4. The method of claim 1, wherein, The method for obtaining the user consultation relevance and the product improvement relevance of the dialogue texts based on the second large language model comprises: inputting multiple sets of prompt words into the second large language model and processing the dialogue texts to calculate a relevance sum weighted by relevance coefficients of user consultation association factors in model output results of each set of prompt words, and further obtaining user consultation relevance by weighted calculation; and calculating a relevance sum weighted by relevance coefficients of product improvement association factors in model output results of each set of prompt words, and further obtaining product improvement relevance by weighted calculation.

5. The method according to claim 3 or 4, characterized in that, The method for extracting the user consultation dialogue text set and the product improvement dialogue text set comprises: screening dialogue texts with user consultation relevance greater than a consultation relevance threshold to generate the user consultation dialogue text set; screening dialogue texts with product improvement relevance greater than an improvement relevance threshold to generate the product improvement dialogue text set.

6. The method of claim 1, wherein, The method for processing the user consultation dialogue text set based on the third large language model to obtain first intelligent customer service training materials, and processing the product improvement dialogue text set to obtain a first product function improvement list comprises: inputting customer service training prompt words into the third large language model to tag the user consultation dialogue text set, and extracting customer service guidance data to obtain the first intelligent customer service training materials, wherein the customer service training prompt words comprise a product question label list. The product improvement prompt word is input into the third large language model to mark the product improvement dialogue text set, and the first product function improvement list is obtained, wherein the product improvement prompt word includes a product function tag list.

7. The method of claim 6, wherein, The method further comprises: According to the correction content of the second intelligent customer service training material and the first intelligent customer service training material, and / or the correction content of the second product function improvement list and the first product function improvement list, The customer service training prompt word and / or the product improvement prompt word are adjusted.

8. The method of claim 1, wherein, The second intelligent customer service training material is output to the intelligent customer service system. The second product function improvement list is output to the product iteration system, comprising: The second intelligent customer service training material is filtered and then output to the intelligent customer service system. The second product function improvement list is sorted and then output to the product iteration system.

9. An information processing apparatus, characterized by comprising: The device comprises: An information acquisition unit configured to acquire audio information to be processed, wherein the audio information includes one or more dialogue audios; A model processing unit configured to convert the audio information into text information based on a first large language model, wherein the text information includes one or more dialogue texts corresponding to the dialogue audios; obtain user consultation relevance and product improvement relevance of the dialogue texts based on a second large language model, and further extract a user consultation dialogue text set and a product improvement dialogue text set; obtain a first intelligent customer service training material by processing the user consultation dialogue text set based on a third large language model, and obtain a first product function improvement list by processing the product improvement dialogue text set; An information verification unit configured to process the first intelligent customer service training material and the first product function improvement list respectively by using a regular expression to obtain a second intelligent customer service training material and a second product function improvement list; A result output unit configured to output the second intelligent customer service training material to an intelligent customer service system, and output the second product function improvement list to a product iteration system.

10. An information processing system, characterized by comprising: The system comprises the information processing device of claim 9, the intelligent customer service system and the product iteration system.

11. An electronic device, comprising: Comprise: One or more processors; A storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1-8.

12. A computer readable medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the method of any one of claims 1-8.

Images