Service recommendation method and device, computer equipment and storage medium

Abstract

The invention belongs to the technical field of artificial intelligence, and relates to a service recommendation method, which comprises the following steps: receiving consultation voice of a target customer, and converting the consultation voice into a consultation text by using a pre-trained voice recognition model; then, performing intention analysis on the text by using the pre-trained large model to obtain client intention information; meanwhile, the language features of the text and the acoustic features of the voice are extracted respectively, so that first and second emotional state information of the customer is determined, and the first and second emotional state information are fused to obtain comprehensive emotional state information. And obtaining a target customer portrait. And finally, based on the intention information, the comprehensive emotional state information and the customer portrait, a pre-trained deep learning recommendation model is adopted to provide service recommendation for the target customer. The invention further provides a service recommendation device, computer equipment and a storage medium. The method can be applied to business management program systems such as financial insurance, and personalized services can be provided for clients.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology and is applied to online processing business scenarios such as finance and insurance. In particular, it relates to a service recommendation method, device, computer device, and storage medium. Background Art

[0002] In the insurance industry, customer service, as a key link in maintaining customer relationships and enhancing corporate image, is of self-evident importance. Traditional customer service systems mainly rely on manual phone calls to handle diverse affairs such as customer consultations and claim applications. However, with the continuous expansion of insurance business and the increasing complexity of customer needs, this traditional model has gradually exposed many drawbacks.

[0003] First of all, the processing efficiency of manual customer service is low. Due to the need to answer calls one by one, the processing speed is slow. Especially during peak business periods, customers often need to wait for a long time, which not only reduces customer satisfaction but also affects the service efficiency of the enterprise. Secondly, the service quality is unstable. The professional levels and emotional states of different customer service staff vary greatly, which directly leads to uneven service quality and makes it difficult to guarantee the customer experience. Finally, the traditional customer service system lacks personalized services. Due to the difficulty in comprehensively and deeply understanding the personalized needs of customers, traditional customer service can only provide standardized services and cannot meet the customized needs of customers.

[0004] In summary, the traditional insurance customer service system has problems such as low efficiency, unstable service quality, and lack of personalized services. Summary of the Invention

[0005] The purpose of the embodiments of this application is to propose a service recommendation method, device, computer device, and storage medium to solve the problems of low efficiency, unstable service quality, and lack of personalized services in the traditional insurance customer service system.

[0006] In the first aspect, a service recommendation method is provided, which adopts the following technical solutions:

[0007] Receive the consultation voice of the target customer, and use a pre-trained speech recognition model to convert the consultation voice into a consultation text; based on the consultation text, use a pre-trained large model to perform intent analysis to obtain the intent information of the target customer; extract the language features of the consultation text, and determine the first emotional state information of the target customer based on the language features; extract the acoustic features of the consultation voice, and determine the second emotional state information of the target customer based on the acoustic features; fuse the first emotional state information and the second emotional state information to determine the comprehensive emotional state information of the target customer; obtain the customer portrait of the target customer, and based on the intent information, comprehensive emotional state information, and customer portrait, use a pre-trained deep learning recommendation model to recommend services to the target customer.

[0008] In a second aspect, a service recommendation device is provided, which adopts the following technical solutions:

[0009] A receiving module, configured to receive the consultation voice of a target customer, and convert the consultation voice into a consultation text by using a pre-trained speech recognition model;

[0010] An analysis module, configured to perform intention analysis on the consultation text by using a pre-trained large model to obtain the intention information of the target customer;

[0011] A first extraction module, configured to extract the language features of the consultation text, and determine the first emotional state information of the target customer based on the language features;

[0012] A second extraction module, configured to extract the acoustic features of the consultation voice, and determine the second emotional state information of the target customer based on the acoustic features;

[0013] A fusion module, configured to fuse the first emotional state information and the second emotional state information to determine the comprehensive emotional state information of the target customer;

[0014] A recommendation module, configured to obtain the customer profile of the target customer, and perform service recommendation for the target customer by using a pre-trained deep learning recommendation model based on the intention information, the comprehensive emotional state information, and the customer profile.

[0015] In a third aspect, a computer device is provided, which adopts the following technical solutions:

[0016] Receive the consultation voice of a target customer, and convert the consultation voice into a consultation text by using a pre-trained speech recognition model;

[0017] Perform intention analysis on the consultation text by using a pre-trained large model to obtain the intention information of the target customer;

[0018] Extract the language features of the consultation text, and determine the first emotional state information of the target customer based on the language features;

[0019] Extract the acoustic features of the consultation voice, and determine the second emotional state information of the target customer based on the acoustic features;

[0020] Fuse the first emotional state information and the second emotional state information to determine the comprehensive emotional state information of the target customer;

[0021] Obtain the customer profile of the target customer, and perform service recommendation for the target customer by using a pre-trained deep learning recommendation model based on the intention information, the comprehensive emotional state information, and the customer profile.

[0022] In a fourth aspect, a computer-readable storage medium is provided, which adopts the following technical solutions:

[0023] Receive the consultation voice of the target customer, and use a pre-trained speech recognition model to convert the consultation voice into consultation text;

[0024] Based on the consultation text, use a pre-trained large model to perform intent analysis to obtain the intent information of the target customer;

[0025] Extract the language features of the consultation text, and determine the first emotional state information of the target customer based on the language features;

[0026] Extract the acoustic features of the consultation voice, and determine the second emotional state information of the target customer based on the acoustic features;

[0027] Fuse the first emotional state information and the second emotional state information to determine the comprehensive emotional state information of the target customer;

[0028] Obtain the customer portrait of the target customer, and based on the intent information, the comprehensive emotional state information, and the customer portrait, use a pre-trained deep learning recommendation model to recommend services to the target customer.

[0029] Compared with the prior art, the embodiments of the present application mainly have the following beneficial effects: By receiving the consultation voice of the target customer and accurately converting it into consultation text with the help of a pre-trained speech recognition model, the problems of slow processing speed and low efficiency of traditional artificial customer service are effectively solved. On this basis, using a pre-trained large model to perform intent analysis on the consultation text can quickly capture the real needs of customers and provide an accurate orientation for subsequent services. At the same time, by extracting the language features of the consultation text and the acoustic features of the consultation voice, the first and second emotional state information of the customer are respectively determined, and the comprehensive emotional state information is obtained through fusion, so that the emotional factors of the customer can be more comprehensively considered in the service process, and the stability of service quality can be improved. In addition, combined with the customer portrait, a pre-trained deep learning recommendation model is used to recommend services, realizing the customization of personalized services and meeting the diverse needs of customers. Brief Description of the Drawings

[0030] In order to more clearly illustrate the solutions in the present application, the following will briefly introduce the drawings required for the description of the embodiments of the present application. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0031] Figure 1 is an exemplary system architecture diagram to which the present application can be applied;

[0032] Figure 2 A flowchart according to an embodiment of the service recommendation method of the present application;

[0033] Figure 3 FIG. 2 is a schematic structural diagram of an embodiment of a service recommendation device according to the present application;

[0034] Figure 4 FIG. 3 is a schematic structural diagram of an embodiment of a computer device according to the present application. Detailed Embodiments

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this application belongs; the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above drawings are intended to cover non-exclusive inclusion. The terms "first", "second", etc. in the specification and claims of this application or the above drawings are used to distinguish different objects and not to describe a specific order.

[0036] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in connection with the embodiments can be included in at least one embodiment of this application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0037] In order to enable those skilled in the technical field to better understand the solution of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0038] As Figure 1 shown, the system architecture 100 may include a terminal device 101, a network 102, and a server 103. The terminal device 101 may be a laptop computer 1011, a tablet computer 1012, or a mobile phone 1013. The network 102 is a medium for providing a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0039] A user may use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications may be installed on the terminal device 101, such as a web browser application, a shopping application, a search application, an instant messaging tool, an email client, a social platform software, etc.

[0040] The terminal device 101 can be various electronic devices with a display screen and supporting web browsing. In addition to the laptop 1011, the tablet computer 1012, or the mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV) player, a laptop computer, a desktop computer, and the like.

[0041] The server 103 can be a server that provides various services, such as a background server that provides support for the pages displayed on the terminal device 101.

[0042] It should be noted that the service recommendation method provided by the embodiments of the present application is generally executed by the server / terminal device. Correspondingly, the service recommendation device is generally set in the server / terminal device.

[0043] It should be understood that Figure 1 the numbers of the terminal devices, the network, and the server in

[0044] are merely illustrative. According to the implementation requirements, there can be any number of terminal devices, networks, and servers. Figure 2 , which shows a flowchart of an embodiment of the service recommendation method according to the present application. The service recommendation method includes the following steps:

[0045] Step S201, receive the consultation voice of the target customer, and use a pre-trained speech recognition model to convert the consultation voice into a consultation text.

[0046] In this embodiment, the electronic device (such as Figure 1 the server / terminal device shown) on which the service recommendation method runs can receive the consultation voice of the target customer through a wired connection method or a wireless connection method. It should be noted that the above wireless connection methods can include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra wideband) connections, and other currently known or future-developed wireless connection methods.

[0047] In this embodiment, the target customer refers to a specific user group that needs to receive services in the insurance business, and interacts with the system through behaviors such as voice consultation and business handling. For example, in the scenario of vehicle insurance renewal, the target customer may be the owner who has no accident record in the previous year, and the system needs to recommend a customized insurance plan according to their driving habits.

[0048] In this embodiment, the consultation voice is the original audio data of insurance-related questions raised by the target customer in the form of voice, including multi-dimensional information such as language content, intonation, and speech rate. Its sources include channels such as telephones and voice assistants, and quality improvement requires preprocessing such as noise reduction and segmentation. For example, when the customer asks "What is the reimbursement scope of medical insurance?", this voice serves as the input data for system analysis and is used for subsequent text conversion and intent recognition.

[0049] In this embodiment, the speech recognition model is a pre-trained model constructed based on deep learning technology and is used to convert the consultation voice into text.

[0050] In this embodiment, the consultation text is the structured text result output by the speech recognition model, representing the core content of the customer's consultation. It serves as the input for subsequent tasks such as intent analysis and emotion recognition. For example, the text "Accident insurance claim process" converted from the consultation voice "How is the claim process for accident insurance?" is the basis for analysis and is used to extract the customer's intent and language features.

[0051] Step S202: Based on the consultation text, use a pre-trained large model to perform intent analysis to obtain the intent information of the target customer.

[0052] In this embodiment, the large model refers to a deep learning model pre-trained based on a large amount of data and has powerful language understanding and generation capabilities. In the insurance scenario, the large model analyzes the semantics and context relationships of the consultation text to extract the customer's intent information. For example, when the input is "I want to purchase critical illness insurance for my family", the large model can recognize the intent of "Family critical illness insurance purchase" and generate the corresponding business label.

[0053] In this embodiment, the intent information refers to the customer's core needs or behavioral goals parsed from the consultation text, representing the motivation for the customer to interact with the system. It is generated through the semantic analysis of the large model and is used to guide service recommendations. For example, when the customer consults "How to modify the policy beneficiary", the intent information is "Policy beneficiary change", and the system triggers the change process guide or recommends relevant services accordingly.

[0054] Step S203: Extract the language features of the consultation text and determine the first emotional state information of the target customer based on the language features.

[0055] In this embodiment, the language features refer to elements such as words and sentence patterns in the consultation text that reflect the customer's expression style and emotional tendency. For example, frequent use of words such as "urgently needed" and "as soon as possible" may indicate urgency, while words such as "worried" and "doubtful" imply anxiety. By analyzing the language features, the system can initially judge the first emotional state information of the customer, such as "The customer uses negative words multiple times, which may express dissatisfaction".

[0056] In this embodiment, the first emotional state information refers to the customer's emotional state obtained through the analysis of the language features of the consultation text, representing the emotional tendency of the customer at the text expression level. For example, if negative words (such as "trouble" and "dissatisfaction") frequently appear in the customer's text, the system can determine that the first emotional state is "negative", which is used for subsequent service strategy adjustment, such as preferentially allocating a human customer service representative.

[0057] Step S204: Extract the acoustic features of the consultation voice, and determine the second emotional state information of the target customer based on the acoustic features.

[0058] In this embodiment, the acoustic features refer to the physical attributes in the consultation voice that reflect the customer's emotional state, including pitch, volume, speech rate, etc. For example, a high pitch and a fast speech rate may indicate excitement or anxiety, while a low and slow speech may imply hesitation or dissatisfaction. By extracting the acoustic features, the system can further analyze the second emotional state information of the customer, such as "frequent pauses in the customer's voice may indicate hesitation".

[0059] In this embodiment, the second emotional state information refers to the customer's emotional state obtained through the analysis of the acoustic features of the consultation voice, representing the emotional tendency of the customer at the voice expression level. For example, if there is a sudden change in the volume or an increase in the intonation in the customer's voice, the system can determine that the second emotional state is "angry" or "excited", which is used for comprehensive emotional assessment and service priority ranking.

[0060] Step S205: Integrate the first emotional state information and the second emotional state information to determine the comprehensive emotional state information of the target customer.

[0061] In this embodiment, the comprehensive emotional state information refers to the panoramic view of the customer's emotions generated by integrating the first emotional state information and the second emotional state information. For example, if the first emotion is "negative" and the second emotion is "angry", the comprehensive emotion may be determined as "strong dissatisfaction", and the system will provide soothing services or trigger the complaint handling process accordingly.

[0062] Step S206: Obtain the customer profile of the target customer, and based on the intent information, the comprehensive emotional state information, and the customer profile, use a pre-trained deep learning recommendation model to recommend services to the target customer.

[0063] In this embodiment, the customer profile refers to a structured tag system constructed based on multi-dimensional data such as the customer's historical behavior, preferences, and demographic characteristics. For example, it includes information such as age, occupation, insurance claim record, and consultation history, which is used to depict the overall picture of the customer. Through the customer profile, the system can achieve personalized service recommendations, such as recommending value-added services or exclusive claim channels for high-frequency claim customers.

[0064] In this embodiment, the deep learning recommendation model is a prediction model trained based on multi-source data such as customer portraits, intent information, and emotional states, and is used to generate a service recommendation list. It learns the complex associations between data through a neural network. For example, by combining the customer's "health insurance consultation" intent with an "anxious" emotion, it recommends a health insurance product that includes a fast claim settlement service and attaches health management suggestions.

[0065] In this embodiment, service recommendation refers to customized service suggestions generated by the system based on customer portraits, intent information, and comprehensive emotional states through the deep learning recommendation model. For example, for young customers who consult about "travel insurance" and are in a positive mood, it recommends a short-term travel insurance product that includes high-risk sports protection and attaches emergency rescue services. For customers with a negative mood, it preferentially provides an entry for artificial customer service intervention or complaint handling.

[0066] This application effectively solves the problems of slow processing speed and low efficiency of traditional artificial customer service by receiving the consultation voice of the target customer and accurately converting it into a consultation text with the help of a pre-trained speech recognition model. On this basis, by using a pre-trained large model to analyze the intent of the consultation text, it can quickly capture the real needs of the customer and provide an accurate orientation for subsequent services. At the same time, by extracting the language features of the consultation text and the acoustic features of the consultation voice, the first and second emotional state information of the customer are respectively determined, and the comprehensive emotional state information is fused, so that the service process can more comprehensively consider the emotional factors of the customer and improve the stability of service quality. In addition, combined with the customer portrait, a pre-trained deep learning recommendation model is used for service recommendation, realizing the customization of personalized services and meeting the diverse needs of customers.

[0067] In some optional implementation manners of this embodiment, in step 201, using a pre-trained speech recognition model to convert the consultation voice into a consultation text specifically includes the following steps: F

[0068] Perform language recognition on the consultation voice to determine the language type of the consultation voice; preprocess the consultation voice to obtain the preprocessed consultation voice, and based on the language type, obtain the pre-trained speech recognition model; input the preprocessed consultation voice into the speech recognition model for speech conversion processing to obtain the consultation text corresponding to the consultation voice.

[0069] In this embodiment, language recognition refers to the process of determining the language category to which it belongs by analyzing the language features (such as phonemes, syllable structures, grammatical patterns, etc.) in the audio signal through an algorithm. This technology can be based on a multi-language acoustic model and a language model, and realizes classification by comparing the matching degree between the input voice and the predefined language template.

[0070] In this embodiment, the language type refers to the natural language category to which the consultation voice belongs, such as Chinese, English, Japanese, etc.

[0071] In this embodiment, preprocessing refers to the steps of performing operations such as noise reduction, segmentation, and normalization on the original consultation voice to improve the accuracy of speech recognition. Specifically, background noise can be eliminated and silent segments can be removed through signal processing techniques (such as spectral subtraction for noise reduction and endpoint detection for segmentation), and the voice can be segmented into segments suitable for model input.

[0072] In one example, when a customer dials a financial insurance consultation hotline, the system first receives the consultation voice. Using a language recognition algorithm, feature extraction is performed on the voice signal, such as Mel-frequency cepstral coefficients, etc. By comparing with a pre-constructed language model library, the language type of the consultation voice is determined, such as Chinese, English, etc. Preprocessing is performed on the consultation voice, including operations such as noise reduction and endpoint detection. Spectral subtraction or a deep learning-based noise reduction algorithm is used for noise reduction to remove background noise interference. Endpoint detection determines the start and end positions of the voice through an energy threshold method or an endpoint detection model based on a neural network. If the language type is Chinese, a pre-trained Chinese speech recognition model (i.e., the speech recognition model) is obtained from the model library based on the determined language type. This Chinese speech recognition model is a model trained on a large amount of financial insurance voice data. The preprocessed consultation voice is input into this Chinese speech recognition model. The model performs feature extraction and pattern matching on the voice signal and converts the voice signal into a corresponding text sequence. For example, when a customer consults "I want to know about the coverage of the A type of critical illness insurance", the model accurately converts it into text.

[0073] In another example, when a patient initiates a voice consultation through a medical consultation platform, the system receives the voice signal. Using a language recognition algorithm, spectral analysis and feature extraction are performed on the voice signal, and in combination with a multilingual model library, the language type is determined. Preprocessing is performed on the consultation voice, including removing interfering noises such as breathing sounds and coughing sounds, using an adaptive filtering algorithm. Endpoint detection uses a convolutional neural network model based on deep learning to improve the detection accuracy. If the language type is Chinese, a pre-trained Chinese medical speech recognition model (i.e., the speech recognition model) is obtained based on the determined language type. This Chinese medical speech recognition model is a model trained on a large amount of medical voice data and professional medical texts. The preprocessed consultation voice is input into the Chinese medical speech recognition model. The model uses an attention mechanism to focus on the key parts of the voice signal and accurately recognizes medical terms and professional vocabulary. For example, when a patient consults "I've been feeling dizzy lately and my blood pressure is a bit high. What should I do", the model accurately converts it into text and correctly recognizes key information such as "dizzy" and "high blood pressure".

[0074] Embodiments of the present application can determine the language type by performing speech recognition on the consultation speech, and can accurately adapt to the subsequent processing flow. Different languages have their unique speech characteristics and grammar rules. Accurately identifying the language type provides a basis for obtaining a suitable pre-trained speech recognition model, and avoids recognition errors caused by language mismatch. Preprocessing the consultation speech can effectively remove noise, standardize the speech signal, and improve the speech quality. For example, removing background noise can make the speech characteristics clearer, and endpoint detection can accurately locate the effective speech segment, laying a good foundation for subsequent speech conversion processing. Obtaining a pre-trained speech recognition model based on the language type and inputting the preprocessed consultation speech into it for conversion can make full use of the language patterns and characteristics learned by the model from large-scale data. This greatly improves the accuracy of speech conversion, can quickly and accurately convert the consultation speech into consultation text, solves the problem of low processing efficiency of traditional human customer service, provides reliable data support for subsequent steps such as text-based intent analysis and emotion state determination, and further improves the overall customer service quality and efficiency.

[0075] In some alternative implementation manners, in step 203, extracting the language features of the consultation text and determining the first emotion state information of the target customer based on the language features specifically include the following steps:

[0076] Performing language feature extraction on the consultation text to obtain the language features of the consultation text; inputting the language features into a pre-trained language emotion classifier to obtain the first emotion state information of the target customer.

[0077] In this embodiment, the language emotion classifier is a pre-trained model based on deep learning, which is used to analyze the language features in the text (such as word choice, sentence structure, frequency of use of negative words, etc.) to judge the emotion state of the speaker.

[0078] In one example, the system receives the consultation voice submitted by the target customer through an online channel (such as an insurance APP, the customer service entrance of the official website, etc.), and uses a pre-trained speech recognition model to convert it into a consultation text. For example, when the customer consults "The claim conditions of the critical illness insurance I bought are too harsh. Is this reasonable?", the system accurately recognizes and converts it into the corresponding text. Language features of the consultation text are extracted, and techniques such as lexical analysis and syntactic analysis are used. Lexical analysis identifies the words in the text, such as nouns (insurance product names), verbs (claims), adjectives (harsh), etc. Syntactic analysis determines the sentence structure, such as the subject-predicate-object relationship. At the same time, the emotional polarity of the words is calculated, such as the occurrence frequencies of positive words (satisfied) and negative words (dissatisfied). The words are converted into vector representations through a word vector model, and the language feature vector of the text is comprehensively obtained. The extracted language feature vector is input into a pre-trained language emotion classifier. This classifier is trained based on a large amount of labeled data in the financial insurance field and uses deep learning algorithms. The classifier analyzes the language features to judge the customer's emotional state, such as anger, anxiety, confusion, etc. For the above consultation text, the classifier judges that the customer is in a "dissatisfied" emotional state and outputs the first emotional state information.

[0079] In the embodiment of the present application, by extracting the language features of the consultation text, the key information contained in the text can be deeply mined. Through analysis techniques such as lexical and syntactic analysis, features such as words, sentence structures, and the emotional polarity of words are accurately identified, and the text is converted into a feature vector form that can be processed by a computer. This process comprehensively captures the semantic information in the customer's consultation and provides a rich data basis for subsequent emotion analysis. The extracted language features are input into a pre-trained language emotion classifier, and with the help of the emotion patterns learned from a large amount of data, the emotional state of the customer can be quickly and accurately judged. Compared with traditional manual judgment, it avoids the judgment deviation caused by the subjective factors of customer service staff. For example, it can accurately identify whether the customer is in a state of anger, anxiety, or confusion, etc., and obtain the first emotional state information. This provides a key basis for subsequent determination of the comprehensive emotional state and service recommendation, helps to improve the pertinence and personalization of services, and effectively solves the problems of unstable service quality and lack of personalized services in traditional customer service systems.

[0080] In some optional implementation manners, in step 204, the acoustic features of the consultation voice are extracted, and the second emotional state information of the target customer is determined based on the acoustic features, which specifically includes the following steps:

[0081] The acoustic features of the consultation voice are extracted to obtain the acoustic features of the consultation voice; the acoustic features are input into a pre-trained acoustic emotion classifier to obtain the second emotional state information of the target customer.

[0082] In this embodiment, the acoustic emotion classifier is a pre-trained model based on audio signal processing, which infers the emotional state of the speaker by analyzing the acoustic features of speech (such as pitch, volume, speech rate, pause frequency, etc.).

[0083] In one example, the system receives the consultation speech submitted by the target customer through channels such as telephone and online customer service. The consultation speech is preprocessed, including noise reduction processing, using an adaptive filtering algorithm to remove background noise, such as environmental noise and equipment interference noise. Endpoint detection, using a double-threshold method based on energy and zero-crossing rate to determine the start and end positions of the speech, and removing invalid speech segments. Acoustic feature extraction is performed on the preprocessed consultation speech. The extracted features include fundamental frequency (reflecting the pitch of the speech), energy (reflecting the intensity of the speech), duration (the duration of the speech signal), and Mel-frequency cepstral coefficients (used to describe the spectral characteristics of the speech), etc. Professional signal processing algorithms and tools are used to accurately calculate these acoustic feature parameters and combine them into an acoustic feature vector. The extracted acoustic feature vector is input into the pre-trained acoustic emotion classifier. The classifier is constructed based on deep learning algorithms and is trained on labeled speech data in the large-scale financial insurance field. The classifier analyzes the acoustic features and judges the customer's emotional state, such as excitement, frustration, calmness, etc., and outputs the second emotional state information.

[0084] The embodiment of the present application can deeply mine the key information in the speech signal other than semantics by extracting the acoustic features of the consultation speech. By extracting acoustic features such as fundamental frequency, energy, duration, and Mel-frequency cepstral coefficients, the physiological and behavioral characteristics of the customer during speech expression are comprehensively captured, and these features are closely related to the customer's emotional state. For example, the change in fundamental frequency can reflect the intonation fluctuation of the customer, and the energy level can reflect the strength and emotional intensity of the customer's speech. By inputting the extracted acoustic features into the pre-trained acoustic emotion classifier and relying on the emotional patterns learned from the large-scale data, the second emotional state information of the customer can be quickly and accurately judged. Compared with traditional manual judgment, it avoids the judgment deviation caused by subjective factors. This process makes up for the deficiency of only relying on consultation text for emotion analysis, provides strong support for judging the customer's emotional state from the speech level, helps to understand the customer's emotion more comprehensively and accurately, provides a key basis for subsequent comprehensive emotional state determination and service recommendation, and effectively improves the quality and personalization degree of insurance customer service.

[0085] In some alternative implementation manners, in step 206, obtaining the customer portrait of the target customer specifically includes the following steps:

[0086] Obtain the historical data of the target customer; preprocess the historical data to obtain the preprocessed historical data, and extract the key features of the preprocessed historical data; establish customer tags for the target customer based on the key features; integrate the key features and customer tags to generate a customer profile of the target customer.

[0087] In this embodiment, the historical data refers to structured or unstructured records generated during the target customer's past interactions with insurance services, including purchase records (such as policy type, insured amount, payment period), consultation content (such as voice text, consultation time), claim records (such as claim amount, reason), etc.

[0088] In this embodiment, the key features refer to the core attributes extracted from the preprocessed historical data that can significantly characterize the customer's behavior pattern or demand preference. They can be selected through statistical analysis (such as frequency statistics, correlation analysis) or machine learning algorithms (such as principal component analysis, feature importance assessment).

[0089] In this embodiment, the customer tags refer to semantic identifiers generated based on the key features and used to describe the customer's specific attributes or behavior patterns.

[0090] In this embodiment, integration refers to the process of associating and fusing the key features with the customer tags to generate a structured customer profile.

[0091] In one example, the system can obtain historical data of target customers from multiple data sources such as the customer relationship management system and business database of an insurance company, including purchase records (such as the types of insurance products purchased, insured amounts, premiums, purchase times, etc.), consultation content (such as insurance terms consulted, claims settlement processes, investment products, etc.), claims settlement records (reasons for claims settlement, claim amounts, claim times, etc.), and customer feedback (results of satisfaction surveys, complaints and suggestions, etc.). Preprocess the obtained historical data, including data cleaning to remove duplicate, incorrect, and incomplete data; data conversion to uniformly convert data in different formats into a format suitable for analysis; data normalization to scale numerical data to a specific interval and eliminate the influence of dimensions. For example, perform normalization processing on the premium data in the purchase records so that its value is within the interval [0, 1]. Adopt feature engineering methods to extract key features from the preprocessed historical data. For purchase records, extract features such as insurance product type preferences, purchase frequencies, and insured amount distributions. For consultation content, extract features such as the insurance fields of concern and problem types. For claims settlement records, extract features such as the distribution of reasons for claims settlement and the range of claim amounts. Use dimensionality reduction algorithms such as principal component analysis to reduce the feature dimensions and retain the main information. Based on the extracted key features, establish customer tags for the target customers. For example, according to the insurance product type preferences, label customers with "preference for critical illness insurance", "preference for accident insurance", etc.; according to the purchase frequency, label them with "high-frequency purchase customers", "low-frequency purchase customers", etc.; according to the reasons for claims settlement, label them with "customers with disease claims", "customers with accident claims", etc. Integrate the key features and customer tags to generate a customer portrait of the target customers. Adopt a structured data representation method to store the customer portrait as a data structure containing multiple fields, and each field corresponds to a key feature or a customer tag.

[0092] The embodiments of this application can comprehensively collect the interaction information between customers and insurance companies by obtaining historical data of target customers, such as purchase records and consultation content. Preprocess these historical data to remove noise and invalid information, obtain clean and usable data, and lay a foundation for subsequent analysis. Extract key features from the preprocessed historical data to accurately mine important information such as the behavior patterns and preferences of customers. Establish customer tags based on the key features, such as "preference for high insured amounts" and "customers who frequently consult about claims settlement", which can present customer characteristics in an intuitive way. Integrate the key features and customer tags to generate a customer portrait, which can form a comprehensive and in-depth understanding of customers.

[0093] In some optional implementation manners, in step 206, based on the intent information, the comprehensive emotional state information, and the customer portrait, a pre-trained deep learning recommendation model is used to perform service recommendations for the target customers, which specifically includes the following steps:

[0094] Integrate the intention information, comprehensive emotional state information, and customer portrait to obtain the customer feature vector of the target customer; input the customer feature vector into a pre-trained deep learning recommendation model to obtain a service recommendation list; screen multiple services in the service recommendation list based on the intention information to determine the target service; and recommend services to the target customer based on the target service.

[0095] In this embodiment, the customer feature vector refers to the structured data representation generated by integrating the intention information, comprehensive emotional state information, and customer portrait of the target customer, and is used to comprehensively represent the customer's current needs, emotional state, and historical behavior patterns.

[0096] In this embodiment, the service recommendation list refers to the candidate service set generated after inputting the customer feature vector into a pre-trained deep learning recommendation model, and contains multiple service items.

[0097] In this embodiment, the target service refers to the final recommendation result selected from the service recommendation list and most in line with the current intention and needs of the target customer.

[0098] In one example, the system receives the intention information of the target customer (such as consulting the critical illness insurance claim process indicating a need for critical illness insurance), the comprehensive emotional state information (such as anxiety, satisfaction, etc.), and the customer portrait (including customer age, occupation, purchase record, risk preference, etc.). Using a data fusion algorithm, these heterogeneous data are integrated to construct the customer feature vector of the target customer. For example, the intention information is encoded into a vector representation, which is concatenated with the structured data in the customer portrait and the vector of the comprehensive emotional state information to form a multi-dimensional customer feature vector. The customer feature vector is input into a pre-trained deep learning recommendation model. This model is trained based on a large amount of data in the financial insurance field and has learned the complex mapping relationship between customer features and insurance services. The model outputs a service recommendation list according to the customer feature vector. The list contains various possible insurance services, such as different types of insurance product recommendations, claim assistance services, health management services, etc., and gives a recommendation score for each service. Screen multiple services in the service recommendation list based on the intention information. For example, if the customer's intention is to consult the critical illness insurance claim process, then services related to critical illness insurance claims, such as professional claim guidance services and claim progress query services, are preferentially selected. Determine the target service according to the recommendation score and intention matching degree. Recommend services to the target customer based on the target service. Through the online channels of the insurance company (such as APP message push, SMS notification, etc.) or offline channels (such as customer service staff phone communication), introduce the content, advantages, and handling methods of the target service to the customer in detail.

[0099] In the embodiments of the present application, the intention information, the comprehensive emotional state information, and the customer portrait can be integrated to form a customer feature vector, comprehensively integrating multi-dimensional information such as the semantic needs, emotional expressions, and historical behaviors of the customer during the consultation process. The intention information accurately reflects the current insurance-related needs of the customer, the comprehensive emotional state information captures the emotional tendencies of the customer during the consultation, and the customer portrait depicts the long-term characteristics and preferences of the customer. Inputting the customer feature vector into a pre-trained deep learning recommendation model, the model can generate a service recommendation list that fits the actual situation of the customer based on the complex patterns learned from a large amount of data. This list covers a variety of possible insurance services, providing the customer with rich choices. Screening the service recommendation list based on the intention information further focuses on the core needs of the customer, ensuring that the recommended services are more targeted and practical. The finally determined target service can accurately meet the personalized needs of the customer, effectively improving the customer's satisfaction and acceptance of insurance services, and solving the problem of the lack of personalized services in traditional customer service systems.

[0100] In some alternative implementation manners, the step of "performing service recommendation for the target customer based on the target service" specifically includes the following steps:

[0101] Obtain service auxiliary information from a preset knowledge base based on the target service; perform service recommendation for the target customer based on the target service and the service auxiliary information.

[0102] In this embodiment, the knowledge base refers to a structured or semi-structured data set stored in the system, including information such as professional knowledge related to insurance business, service rules, product terms, common question answers, and historical cases.

[0103] In this embodiment, the service auxiliary information refers to the supporting content extracted from the knowledge base based on the target service and used to enhance the service recommendation effect, including product details, operation guides, risk warnings, case references, etc.

[0104] In one example, the system receives the consultation voice of the target customer. After steps such as speech recognition, intention analysis, determination of emotional state, and integration of customer portraits, a customer feature vector is obtained. The customer feature vector is input into a pre-trained deep learning recommendation model to generate a service recommendation list, and the services in the list are filtered based on the intention information to determine the target service. For example, if the customer consults about the claim settlement process of a critical illness insurance, the system determines that the target service is "critical illness insurance claim settlement assistance service". Based on the determined target service, service auxiliary information is obtained from a preset knowledge base. The preset knowledge base contains details such as detailed introductions, handling processes, required materials, and answers to common questions of various insurance services. Taking the "critical illness insurance claim settlement assistance service" as an example, the system extracts the specific process of this service from the knowledge base, such as the claim settlement materials that the customer needs to prepare (diagnosis certificates, medical records, expense lists, etc.), the submission methods of claim settlement applications (online or offline), and the estimated claim settlement processing time. The target service and the obtained service auxiliary information are integrated and recommended to the target customer through the online channels (such as APP message push, email, etc.) or offline channels (such as phone communication with customer service staff) of the insurance company. The recommended content not only includes the service name but also elaborates in detail on the service auxiliary information to help the customer comprehensively understand the service content and handling methods.

[0105] In the embodiment of this application, service auxiliary information is obtained from a preset knowledge base based on the target service. This process makes full use of the rich resources of the preset knowledge base. The preset knowledge base covers detailed information, handling processes, answers to common questions, etc. of various insurance services, and can provide comprehensive and accurate auxiliary content for the target service. Combining the target service to obtain auxiliary information makes the service recommendation no longer limited to a simple service name but includes in-depth knowledge related to the service. Recommending services to the target customer based on the target service and service auxiliary information can provide clearer and more specific service guidance for the customer. While understanding the target service, the customer can also obtain specific information required for handling the service, such as required materials and handling steps. This not only improves the customer's awareness and acceptance of the service but also reduces the confusion and misunderstanding caused by insufficient information. In this way, the problem of the lack of personalized services in the traditional customer service system is effectively solved, and the customer service quality and customer satisfaction are improved.

[0106] It should be emphasized that to further ensure the privacy and security of the above-mentioned consultation voice, consultation text, intention information, first emotional state information, second emotional state information, and comprehensive emotional state information, the above-mentioned consultation voice, consultation text, intention information, first emotional state information, second emotional state information, and comprehensive emotional state information can also be stored in a node of a blockchain.

[0107] The blockchain referred to in this application is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithms. A blockchain, essentially a decentralized database, is a series of data blocks generated by using cryptographic methods. Each data block contains information about a batch of network transactions, which is used to verify the validity of the information (anti-counterfeiting) and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer, an application service layer, etc.

[0108] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Among them, Artificial Intelligence (AI) uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, sense the environment, acquire knowledge, and use knowledge to obtain the best results in theory, methods, technologies, and application systems.

[0109] Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation / interaction systems, and mechatronics. Artificial intelligence software technologies mainly include several major directions such as computer vision technology, robotics, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0110] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, an optical disk, a Read-Only Memory (ROM), or a Random Access Memory (RAM), etc.

[0111] It should be understood that although the steps in the flowchart of the accompanying drawings are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings can include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. Their execution order is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or sub-steps or stages of other steps.

[0112] For further reference Figure 3, as an implementation of the method described above Figure 2 An embodiment of a service recommendation device is provided in this application. This device embodiment corresponds to Figure 2 the method embodiment shown, and this device can be specifically applied to various electronic devices.

[0113] As Figure 4 shown, the service recommendation device 400 in this embodiment includes: a receiving module 401, an analysis module 402, a first extraction module 403, a second extraction module 404, a fusion module 405, and a recommendation module 406. Among them:

[0114] The receiving module 401 is configured to receive the consultation voice of the target customer, and use a pre-trained speech recognition model to convert the consultation voice into a consultation text;

[0115] The analysis module 402 is configured to perform intent analysis on the consultation text using a pre-trained large model to obtain the intent information of the target customer;

[0116] The first extraction module 403 is configured to extract the language features of the consultation text and determine the first emotional state information of the target customer based on the language features;

[0117] The second extraction module 404 is configured to extract the acoustic features of the consultation voice and determine the second emotional state information of the target customer based on the acoustic features;

[0118] The fusion module 405 is configured to fuse the first emotional state information and the second emotional state information to determine the comprehensive emotional state information of the target customer;

[0119] The recommendation module 406 is configured to obtain the customer portrait of the target customer, and based on the intent information, the comprehensive emotional state information, and the customer portrait, use a pre-trained deep learning recommendation model to perform service recommendation for the target customer.

[0120] In this embodiment, by receiving the consultation voice of the target customer and accurately converting it into a consultation text with the help of a pre-trained speech recognition model, the problems of slow processing speed and low efficiency of traditional artificial customer service are effectively solved. On this basis, using a pre-trained large model to perform intent analysis on the consultation text can quickly capture the real needs of customers and provide accurate guidance for subsequent services. At the same time, by extracting the language features of the consultation text and the acoustic features of the consultation voice, the first and second emotional state information of the customer are respectively determined, and the comprehensive emotional state information is fused, so that the service process can more comprehensively consider the emotional factors of customers and improve the stability of service quality. In addition, combined with the customer portrait, a pre-trained deep learning recommendation model is used for service recommendation, realizing the customization of personalized services and meeting the diverse needs of customers.

[0121] In one embodiment, the receiving module 402 includes:

[0122] A language recognition sub-module for performing language recognition on the consultation voice to determine the language type of the consultation voice;

[0123] A voice preprocessing sub-module for preprocessing the consultation voice to obtain the preprocessed consultation voice, and based on the language type, obtaining a pre-trained voice recognition model;

[0124] A conversion sub-module for inputting the preprocessed consultation voice into the voice recognition model for voice conversion processing to obtain the consultation text corresponding to the consultation voice.

[0125] In the embodiment of the present application, the language type can be determined by performing language recognition on the consultation voice, which can accurately adapt to the subsequent processing flow. Different languages have their unique voice characteristics and grammar rules. Accurately recognizing the language type provides a basis for obtaining a suitable pre-trained voice recognition model, avoiding recognition errors caused by language mismatch. Preprocessing the consultation voice can effectively remove noise and standardize the voice signal, improving the voice quality. For example, removing background noise can make the voice characteristics clearer, and endpoint detection can accurately locate the effective voice segment, laying a good foundation for subsequent voice conversion processing. Obtaining a pre-trained voice recognition model based on the language type and inputting the preprocessed consultation voice into it for conversion can make full use of the language patterns and features learned by the model from large-scale data. This greatly improves the accuracy of voice conversion, can quickly and accurately convert the consultation voice into the consultation text, solves the problem of low processing efficiency of traditional human customer service, provides reliable data support for subsequent steps such as text-based intent analysis and emotion state determination, and further improves the overall customer service quality and efficiency.

[0126] In one embodiment, the first extraction module 403 includes:

[0127] A first extraction sub-module for extracting language features from the consultation text to obtain the language features of the consultation text;

[0128] A first input sub-module for inputting the language features into a pre-trained language emotion classifier to obtain the first emotion state information of the target customer.

[0129] Embodiments of the present application can extract language features from consultation texts, enabling in-depth mining of the key information contained in the texts. Through lexical, syntactic, and other analysis techniques, features such as vocabulary, sentence structure, and the sentiment polarity of words are accurately identified, and the text is transformed into a feature vector form that can be processed by a computer. This process comprehensively captures the semantic information in customer consultations, providing a rich data basis for subsequent sentiment analysis. The extracted language features are input into a pre-trained language sentiment classifier, and by leveraging the sentiment patterns learned from large-scale data, the emotional state of the customer can be quickly and accurately determined. Compared with traditional manual judgment, it avoids judgment biases caused by subjective factors of customer service staff. For example, it can accurately identify whether the customer is in an angry, anxious, or confused mood, obtaining the first emotional state information. This provides a key basis for subsequent determination of the comprehensive emotional state and service recommendation, helping to improve the pertinence and personalization of services, and effectively solving the problems of unstable service quality and lack of personalized services in traditional customer service systems.

[0130] In one embodiment, the second extraction module 404 includes:

[0131] A second extraction sub-module for extracting acoustic features from the consultation voice to obtain the acoustic features of the consultation voice;

[0132] A second input sub-module for inputting the acoustic features into a pre-trained acoustic sentiment classifier to obtain the second emotional state information of the target customer.

[0133] Embodiments of the present application can extract acoustic features from consultation voices, enabling in-depth mining of the key information other than semantics in the voice signals. By extracting acoustic features such as fundamental frequency, energy, duration, and Mel-frequency cepstral coefficients, the physiological and behavioral characteristics of the customer during speech expression are comprehensively captured, and these characteristics are closely related to the customer's emotional state. For example, changes in the fundamental frequency can reflect the intonation fluctuations of the customer, and the energy level can reflect the intensity and emotional strength of the customer's speech. The extracted acoustic features are input into a pre-trained acoustic sentiment classifier, and by leveraging the sentiment patterns learned from large-scale data, the second emotional state information of the customer can be quickly and accurately determined. Compared with traditional manual judgment, it avoids judgment biases caused by subjective factors. This process makes up for the deficiency of relying solely on consultation texts for sentiment analysis, provides strong support for judging the customer's emotional state from the voice level, helps to understand the customer's emotions more comprehensively and accurately, provides a key basis for subsequent determination of the comprehensive emotional state and service recommendation, and effectively improves the quality and personalization of insurance customer service.

[0134] In one embodiment, the recommendation module 406 includes:

[0135] An acquisition sub-module for acquiring the historical data of the target customer;

[0136] A data preprocessing sub-module for preprocessing historical data to obtain preprocessed historical data and extracting key features of the preprocessed historical data;

[0137] A building sub-module for building customer tags of target customers based on the key features;

[0138] A first integration sub-module for integrating the key features and customer tags to generate a customer portrait of the target customer.

[0139] In the embodiment of the present application, by obtaining the historical data of the target customer, such as purchase records, consultation content, etc., the interaction information between the customer and the insurance company can be comprehensively collected. Preprocessing these historical data to remove noise and invalid information, obtaining clean and available data, laying a foundation for subsequent analysis. Extracting the key features of the preprocessed historical data can accurately mine important information such as the customer's behavior pattern and preference. Building customer tags based on the key features, such as "high insurance amount preference", "frequent consultation on claims customers", etc., can present the customer features in an intuitive way. Integrating the key features and customer tags to generate a customer portrait can form a comprehensive and in-depth understanding of the customer.

[0140] In one embodiment, the recommendation module 406 includes:

[0141] A second integration sub-module for integrating the intent information, comprehensive emotional state information, and customer portrait to obtain a customer feature vector of the target customer;

[0142] An input third sub-module for inputting the customer feature vector into a pre-trained deep learning recommendation model to obtain a service recommendation list;

[0143] A screening sub-module for screening multiple services in the service recommendation list based on the intent information to determine the target service;

[0144] A recommendation sub-module for recommending services to the target customer based on the target service.

[0145] Embodiments of the present application can integrate intent information, comprehensive emotional state information, and customer portraits to form customer feature vectors, comprehensively integrating multi-dimensional information such as customers' semantic needs, emotional expressions, and historical behaviors during the consultation process. The intent information accurately reflects the current insurance-related needs of customers, the comprehensive emotional state information captures the emotional tendencies of customers during the consultation, and the customer portrait depicts the long-term characteristics and preferences of customers. Inputting the customer feature vector into a pre-trained deep learning recommendation model, the model can generate a service recommendation list that fits the actual situation of the customer based on the complex patterns learned from a large amount of data. This list covers various possible insurance services, providing customers with rich choices. Screening the service recommendation list based on the intent information further focuses on the core needs of customers, ensuring that the recommended services are more targeted and practical. The finally determined target service can accurately meet the personalized needs of customers, effectively improving customers' satisfaction and acceptance of insurance services, and solving the problem of the lack of personalized services in traditional customer service systems.

[0146] In one embodiment, the recommendation sub-module is further configured to obtain service auxiliary information from a preset knowledge base based on the target service; and perform service recommendation for the target customer based on the target service and the service auxiliary information.

[0147] Embodiments of the present application obtain service auxiliary information from a preset knowledge base based on the target service, and this process makes full use of the rich resources of the preset knowledge base. The preset knowledge base covers detailed information, handling processes, common question answers, etc. of various insurance services, and can provide comprehensive and accurate auxiliary content for the target service. Combining the target service to obtain auxiliary information makes the service recommendation no longer limited to simple service names, but includes in-depth knowledge related to the service. Performing service recommendation for the target customer based on the target service and the service auxiliary information can provide customers with clearer and more specific service guidance. While customers understand the target service, they can also obtain specific information required for handling the service, such as required materials, handling steps, etc. This not only improves customers' awareness and acceptance of the service, but also reduces the confusion and misunderstanding caused by insufficient information. In this way, the problem of the lack of personalized services in traditional customer service systems is effectively solved, and the quality of customer service and customer satisfaction are improved.

[0148] To solve the above technical problems, embodiments of the present application also provide a computer device. For details, please refer to Figure 4 , Figure 4 which is the basic structural block diagram of the computer device in this embodiment.

[0149] The computer device 4 includes a memory 61, a processor 62, and a network interface 63 that communicate with each other through a system bus. It should be noted that only the computer device 6 with a memory 61, a processor 62, and a network interface 63 is shown in the figure. However, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Among them, those skilled in the art of the present technology can understand that the computer device here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0150] The computer device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The computer device can perform human-computer interaction with the user through a keyboard, a mouse, a remote control, a touchpad, a voice control device, etc.

[0151] The memory 61 includes at least one type of readable storage medium. The readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 61 can be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 can also be an external storage device of the computer device 6, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device 6. Of course, the memory 61 can also include both the internal storage unit and the external storage device of the computer device 6. In this embodiment, the memory 61 is generally used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions of the service recommendation method. In addition, the memory 61 can also be used to temporarily store various data that have been output or will be output.

[0152] The processor 62 may be a Central Processing Unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 62 is generally used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to run the computer-readable instructions stored in the memory 61 or process data, such as running the computer-readable instructions of the service recommendation method.

[0153] The network interface 63 may include a wireless network interface or a wired network interface. The network interface 63 is generally used to establish a communication connection between the computer device 6 and other electronic devices.

[0154] In the embodiments of the present application, by receiving the consultation voice of the target customer and accurately converting it into a consultation text with the help of a pre-trained speech recognition model, the problems of slow processing speed and low efficiency of traditional human customer service are effectively solved. On this basis, by using a pre-trained large model to analyze the intent of the consultation text, the true needs of the customer can be quickly captured, providing an accurate orientation for subsequent services. At the same time, by extracting the language features of the consultation text and the acoustic features of the consultation voice, the first and second emotional state information of the customer are respectively determined, and the comprehensive emotional state information is fused, enabling the service process to more comprehensively consider the emotional factors of the customer and improving the stability of service quality. In addition, combined with the customer portrait, a pre-trained deep learning recommendation model is used for service recommendation, realizing the customization of personalized services and meeting the diverse needs of customers.

[0155] The present application also provides another implementation manner, that is, to provide a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor, so that at least one processor executes the steps of the service recommendation method as described above.

[0156] In the embodiments of the present application, by receiving the consultation voice of the target customer and accurately converting it into a consultation text with the help of a pre-trained speech recognition model, the problems of slow processing speed and low efficiency of traditional human customer service are effectively solved. On this basis, by using a pre-trained large model to analyze the intent of the consultation text, the true needs of the customer can be quickly captured, providing an accurate orientation for subsequent services. At the same time, by extracting the language features of the consultation text and the acoustic features of the consultation voice, the first and second emotional state information of the customer are respectively determined, and the comprehensive emotional state information is fused, enabling the service process to more comprehensively consider the emotional factors of the customer and improving the stability of service quality. In addition, combined with the customer portrait, a pre-trained deep learning recommendation model is used for service recommendation, realizing the customization of personalized services and meeting the diverse needs of customers.

[0157] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-described example methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions for causing a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of the present application.

[0158] Obviously, the above-described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The accompanying drawings show the preferred embodiments of the present application, but do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure content of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions described in the foregoing specific embodiments, or perform equivalent replacements on some of the technical features. Any equivalent structure directly or indirectly using the content of the specification and drawings of the present application in other related technical fields is similarly within the scope of patent protection of the present application.

[0159] The non-company software tools or components that appear in the embodiments of the present application are only introduced by way of example and do not represent actual use.

Claims

1. A service recommendation method, characterized in that, Including the following steps: Receiving the consultation voice of the target customer, and using a pre-trained speech recognition model to convert the consultation voice into a consultation text; Based on the consultation text, performing intention analysis using a pre-trained large model to obtain the intention information of the target customer; Extracting the language features of the consultation text, and determining the first emotional state information of the target customer based on the language features; Extracting the acoustic features of the consultation voice, and determining the second emotional state information of the target customer based on the acoustic features; Fusing the first emotional state information and the second emotional state information to determine the comprehensive emotional state information of the target customer; Obtaining the customer profile of the target customer, and based on the intention information, the comprehensive emotional state information, and the customer profile, using a pre-trained deep learning recommendation model to perform service recommendation for the target customer.

2. The method according to claim 1, characterized in that, The step of using a pre-trained speech recognition model to convert the consultation voice into a consultation text specifically includes: Performing language recognition on the consultation voice to determine the language type of the consultation voice; Preprocessing the consultation voice to obtain the preprocessed consultation voice, and based on the language type, obtaining a pre-trained speech recognition model; Inputting the preprocessed consultation voice into the speech recognition model for speech conversion processing to obtain the consultation text corresponding to the consultation voice.

3. The method according to claim 1, wherein The step of extracting the language features of the consultation text and determining the first emotional state information of the target customer based on the language features specifically includes: Performing language feature extraction on the consultation text to obtain the language features of the consultation text; Inputting the language features into a pre-trained language emotion classifier to obtain the first emotional state information of the target customer.

4. The method according to claim 1, wherein The step of extracting the acoustic features of the consultation voice and determining the second emotional state information of the target customer based on the acoustic features specifically includes: Performing acoustic feature extraction on the consultation voice to obtain the acoustic features of the consultation voice; Inputting the acoustic features into a pre-trained acoustic emotion classifier to obtain the second emotional state information of the target customer.

5. The method according to claim 1, characterized in that The step of obtaining the customer profile of the target customer specifically includes: Obtaining the historical data of the target customer; Preprocessing the historical data to obtain the preprocessed historical data, and extracting the key features of the preprocessed historical data; Based on the key features, establishing the customer label of the target customer; Integrating the key features and the customer label to generate the customer profile of the target customer.

6. The method according to claim 1, wherein The step of performing service recommendation for the target customer based on the intention information, the comprehensive emotional state information, and the customer profile using a pre-trained deep learning recommendation model specifically includes: Integrating the intention information, the comprehensive emotional state information, and the customer profile to obtain the customer feature vector of the target customer; Inputting the customer feature vector into a pre-trained deep learning recommendation model to obtain a service recommendation list; Based on the intention information, filter multiple services in the service recommendation list to determine the target service; Based on the target service, make service recommendations to the target customer.

7. The method according to claim 1, characterized in that, The step of making service recommendations to the target customer based on the target service specifically includes: Based on the target service, obtain service auxiliary information from a preset knowledge base; Based on the target service and the service auxiliary information, make service recommendations to the target customer.

8. A service recommendation device, characterized in that, It includes: A receiving module, configured to receive the consultation voice of the target customer, and use a pre-trained speech recognition model to convert the consultation voice into a consultation text; An analysis module, configured to perform intention analysis on the consultation text using a pre-trained large model to obtain the intention information of the target customer; A first extraction module, configured to extract the language features of the consultation text and determine the first emotional state information of the target customer based on the language features; A second extraction module, configured to extract the acoustic features of the consultation voice and determine the second emotional state information of the target customer based on the acoustic features; A fusion module, configured to fuse the first emotional state information and the second emotional state information to determine the comprehensive emotional state information of the target customer; A recommendation module, configured to obtain the customer portrait of the target customer, and based on the intention information, the comprehensive emotional state information, and the customer portrait, use a pre-trained deep learning recommendation model to make service recommendations to the target customer.

9. A computer device, characterized in that, It includes a memory and a processor. Computer-readable instructions are stored in the memory, and when the processor executes the computer-readable instructions, the steps of the service recommendation method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that, Computer-readable instructions are stored on the computer-readable storage medium, and when the computer-readable instructions are executed by the processor, the steps of the service recommendation method according to any one of claims 1 to 7 are implemented.

Images