Intelligent financial consultation system and method based on large language model

Abstract

The invention discloses an intelligent financial consultation system and method based on a large language model, and the system comprises the steps: receiving a multi-mode financial consultation request of a customer, analyzing the input of the customer through the natural language processing and computer vision technology, extracting key information and intention, and dynamically constructing and updating a customer portrait. And specific requirements of clients can be accurately identified. And after the identified demands are stored in a problem database, the system performs real-time analysis on historical data, market dynamics and customer behaviors by using a machine learning algorithm, thereby automatically identifying and predicting potential demands and market trends of customers. Afterwards, on the basis of the prediction results, the system uses a large language model to generate personalized and multi-dimensional financial suggestions, including different risk levels and profit expectations, so as to help customers to make better decisions.

Description

Technical Field

[0001] The present invention belongs to the field of financial consulting, and in particular, relates to an intelligent financial consulting system and method based on a large language model. Background Art

[0002] In the current financial consulting field, traditional financial consulting systems usually rely on static customer data and simple rule-driven algorithms for analysis and advice. These conventional systems generally only support a single form of customer input, such as text, and have difficulty effectively processing multi-modal inputs such as voice and images. The limitations of this input form make it impossible for the system to fully capture the true intentions and needs of customers.

[0003] In addition, traditional financial consulting systems are often based on pre-defined rule bases or limited historical data, making it difficult to dynamically track customer behavior and demand changes. Once a customer profile is established, it is rarely updated, resulting in insufficient personalization of consulting advice and difficulty in flexibly responding to customer changes or new needs. This static analysis model often results in the advice received by customers not accurately matching their current needs, affecting the accuracy and effectiveness of decision-making.

[0004] Existing systems also have significant deficiencies in identifying and predicting potential customer needs and market trends. Conventional technologies mainly provide simple rule-based predictions by reviewing historical data and market conditions. They lack intelligent prediction algorithms and cannot achieve real-time and forward-looking demand judgments. In addition, such systems lack the ability to conduct in-depth sentiment analysis of customer input and cannot make personalized adjustments based on customer emotions and preferences, resulting in poor user experience.

[0005] In view of this, the present invention is proposed. Summary of the invention

[0006] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide an intelligent financial consulting system and method based on a large language model, which solves the problems raised in the above-mentioned background technology.

[0007] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:

[0008] An intelligent financial consulting system and method based on a large language model, comprising the following steps:

[0009] Receive financial consultation requests from customers, then parse the customer input, extract key information and intent, build customer profiles, and identify specific customer needs;

[0010] The identified customer needs will be stored in the question database and machine learning algorithms will be used to automatically identify and predict customers' potential needs and trends through real-time analysis of historical data, market dynamics and customer behavior;

[0011] After completing data storage and predictive analysis, the large language model is used to generate personalized financial advice related to customer consultations, provide multi-dimensional solutions, and output corresponding consultation answers;

[0012] After answering the questions, monitor market changes and user feedback in real time and adjust consulting strategies based on this information.

[0013] In this embodiment, the steps of receiving a financial consultation request input by a customer are:

[0014] First, confirm the input channel selected by the customer. The input channels include text input, voice input or image / video input. After the customer submits the request, the system verifies the format of the input content. Voice input needs to be converted into text through the voice recognition module, and image / video input needs to determine the clarity and content recognizability.

[0015] In this embodiment, when parsing the customer's input, first, the customer's text input, voice input, and image input are received through a multimodal input interface and preprocessed respectively. For text data, first, the text data is segmented into words, and then the part of speech of each word in the word is identified. Subsequently, important entities in the text are identified, key information is extracted, and then the sentiment tendency of the text is analyzed. The text in the image is identified and converted into an editable text format. For image / video input, first, computer vision algorithms are used to identify elements in the image and extract valuable information. The results from NLP and visual analysis are integrated to form a comprehensive understanding of the customer input. Finally, the analysis results of NLP and computer vision are combined to extract key information in customer consultation, and the consultation topic and related data are identified.

[0016] In this embodiment, after obtaining the consulting topic and related data, it is necessary to analyze and extract key information by building an intent recognition model, identify the customer's specific intention and use the context and sentiment analysis results, and then dynamically update the customer portrait based on the analysis and intent recognition results to identify the customer's interest areas and consulting mode. Finally, the key information, intent recognition and customer portrait are integrated to identify the customer's specific needs. The expression is: D=j(K, I, P′, M), where D is the identified customer's specific needs, j is the comprehensive analysis function, M is the identified consulting mode, and I is the identified customer intention.

[0017] In this embodiment, the steps of storing the identified customer needs in the question database are:

[0018] Build a problem database, then store the identified customer needs and related information in the database, then extract and integrate data from different sources to form a unified data set, then extract relevant features from the integrated data and use the Lasso regression algorithm to evaluate the importance of the features, then divide the integrated data set into a training set and a test set, then use the training set data to train the support vector machine model, and then output the prediction results generated by the machine learning model to the system interface for the relevant financial advisor.

[0019] In this embodiment, when using the Lasso regression algorithm to evaluate the importance of features, a new combined feature is first created to capture the potential relationship, feature encoding is performed, and the categorical features are converted into numerical form. Then, the formula Calculate feature X k The correlation between the target variable Y and the target variable Y is used to identify the features with high correlation with the target variable. Then, the integrated data set is randomly divided into training set and test set for subsequent model training and evaluation. Subsequently, the Lasso regression model is created and the initial regularization parameter is set. The expression is: Among them, L(β) represents the loss function of Lasso regression, It means to sum all samples, i is the index of the sample, m is the total number of samples, Y i represents the true value or target variable of the i-th sample, represents the predicted value of the i-th sample, calculated by the model, represents the square of the prediction error of the i-th sample, α is the regularization parameter, which controls the weight of the L1 regularization term, represents the sum of all features, j is the index of the feature, n is the total number of features, |β j | represents the absolute value of the regression coefficient corresponding to the jth feature, This is the variable X k Pearson correlation coefficient between X and Y, Cov(X k ,Y) This is X k The covariance of Y and Y measures the direction and degree of common variation of two variables. This is X k The standard deviation of X k The degree of dispersion of the distribution, σ Y This is the standard deviation of Y, which indicates the dispersion of the distribution of Y and measures the fluctuation range of Y;

[0020] After the creation is completed, the Lasso regression model is trained on the training set to learn the relationship between the features and the target variable. Finally, by analyzing the coefficients of each feature in the Lasso model, it is identified which features have a greater impact on the prediction results.

[0021] In this embodiment, after completing data storage and predictive analysis, the steps of using a large language model to generate personalized financial advice related to customer consultation, provide multi-dimensional solutions, and output corresponding consultation answers are as follows:

[0022] Build a big language model and input the sorted information into the big language model. Then, use the big language model to generate personalized financial advice. The big language model will analyze the input information, combine its knowledge base, and output multi-dimensional solutions related to customer consultations. When generating advice, the model is required to provide multiple alternatives, including different risk levels and expected returns. Then, adjust the tone and style of the output content according to the customer's emotional state and preferences. Then, organize the generated personalized advice into an easy-to-understand format and send it to the customer.

[0023] In this embodiment, the steps of constructing a large language model are:

[0024] According to the Transformer architecture, the number of layers, the number of units in each layer, and the activation function of the model are set. Then, the learning rate, batch size, and number of training rounds are set. Then, the performance of the model is evaluated on the validation set. The perplexity and accuracy are calculated using formulas (1) and (2) for quantitative evaluation. Then, the hyperparameters or training strategies of the model are adjusted based on the evaluation results. Finally, the trained model is deployed to the production environment and the data is further analyzed. In the process of quantitative evaluation, the expression is:

[0025]

[0026] Among them, Perplexity is an indicator to measure the performance of the language model, which indicates the model's ability to predict the test data. exp is a natural exponential function, indicating that the perplexity is obtained by taking the exponent of the average negative log-likelihood. N is the number of samples in the test set or the length of the prediction sequence. The sum symbol indicates the accumulation of all samples, i represents the index of each sample, logP(y i |x ＜i ) represents the i-th target y i Input x in the given preamble ＜i The logarithm of the predicted probability under the condition of i |x ＜i ) is the probability distribution of the model output, Accuracy is the accuracy rate, which is an indicator to measure the prediction accuracy of the classification model, N is the total number of samples in the test set, is the summation symbol, which means the accuracy of all samples is accumulated. Indicator function, when the predicted value y i Equal to the true value y i If true, it returns 1; otherwise, it returns 0.

[0027] In this embodiment, after answering the question, the steps of monitoring market changes and user feedback in real time and adjusting the consulting strategy according to the information are as follows:

[0028] Obtain market dynamics and provide a feedback mechanism after customers receive consulting advice, allowing customers to rate, comment on or raise questions about the advice. Then, use data analysis tools to analyze real-time market data, identify potential market trends, abnormal fluctuations or risk signals, and classify and perform sentiment analysis on the collected user feedback to understand customers' satisfaction, recognition and demand changes for consulting advice. Then, integrate market data and user feedback to form a comprehensive analysis view. Finally, set key performance indicators and compare current indicators with historical data to determine whether there is a significant downward trend and whether it is still suitable for the current market environment and customer needs. Then, apply the adjusted consulting strategy again to customer consultation, collect new user feedback, and form a feedback loop.

[0029] An intelligent financial consulting system based on a large language model, comprising:

[0030] The customer input module is used to receive financial consulting requests from customers, and then parse the customer input, extract key information and intent, build a customer profile, and identify the customer's specific needs;

[0031] Analysis and demand identification module: The identified customer needs will be stored in the problem database and the machine learning algorithm will be used to automatically identify and predict the customer's potential needs and trends through real-time analysis of historical data, market dynamics and customer behavior;

[0032] Demand storage and machine learning analysis module: After completing data storage and predictive analysis, the module uses a large language model to generate personalized financial advice related to customer consultations, provide multi-dimensional solutions, and output corresponding consultation answers;

[0033] The large language model generation module is used to monitor market changes and user feedback in real time after answering the questions, and adjust the consulting strategy based on this information.

[0034] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all the advantages described below at the same time:

[0035] The present invention receives multimodal financial consulting requests from customers, uses natural language processing and computer vision technology to parse customer input, extract key information and intentions, and dynamically builds and updates customer portraits to accurately identify customers' specific needs. After storing the identified needs in the problem database, the system uses machine learning algorithms to perform real-time analysis of historical data, market dynamics, and customer behavior, thereby automatically identifying and predicting customers' potential needs and market trends. Afterwards, based on these prediction results, the system uses a large language model to generate personalized, multi-dimensional financial advice, including different risk levels and return expectations, to help customers make better decisions. After completing the consulting service, the system will continue to monitor market changes and customer feedback, and adjust consulting strategies based on real-time data, thereby providing more accurate consulting services that dynamically respond to market changes, greatly improving customer experience and decision-making efficiency.

[0036] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings described below are only some embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0038] In the figure:

[0039] Figure 1 A flowchart for the financial consulting method;

[0040] Figure 2 This is a block diagram of the financial consulting system.

[0041] It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but are intended to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0042] The present invention will now be described in further detail with reference to the accompanying drawings.

[0043] See also Figure 1-2 As shown, in this embodiment, an intelligent financial consulting system and method based on a large language model are provided, including the following steps:

[0044] Receive financial consultation requests from customers, then parse the customer input, extract key information and intent, build a customer profile, and identify the customer's specific needs; by parsing multiple customer inputs (text, voice, image), the system can accurately and quickly extract key information and customer intent to ensure that the customer's specific needs can be fully understood. This multimodal input processing method improves the flexibility and accuracy of customer consultation.

[0045] The identified customer needs will be stored in the question database and machine learning algorithms will be used to automatically identify and predict the customer's potential needs and trends through real-time analysis of historical data, market dynamics and customer behavior; the system will dynamically update the customer profile based on the customer's input and historical behavior, and identify the customer's areas of interest and consulting patterns. Through this dynamic update, the system can better predict customer needs and provide highly personalized consulting services. This approach greatly improves the customer experience, allowing each consultation to accurately match the customer's current needs.

[0046] After completing data storage and predictive analysis, the system uses a large language model to generate personalized financial advice related to customer consultations, provide multi-dimensional solutions, and output corresponding consultation answers; using machine learning algorithms, the system can analyze customer needs in real time, and automatically identify potential needs and market trends in combination with historical data and market dynamics. Through this intelligent analysis, the system can not only solve customers' immediate problems, but also provide customers with forward-looking advice to help them foresee future market opportunities and risks. After predictive analysis, the system generates personalized financial advice related to customers through a large language model, covering multiple dimensions. By providing multiple alternatives, customers can choose the solution that best suits their needs and preferences, which not only enhances customer participation, but also increases the flexibility of financial decision-making.

[0047] After answering the questions, the system monitors market changes and user feedback in real time, and adjusts the consulting strategy based on this information. With the real-time monitoring function of market changes and customer feedback, it can continuously track market trends and user feedback after customers receive advice, and dynamically adjust the consulting strategy based on this information. This adaptive strategy ensures that the system can maintain an efficient service level in a rapidly changing market environment and optimize in real time according to the latest market trends and customer needs.

[0048] In this embodiment, the steps of receiving a financial consultation request input by a customer are:

[0049] First, confirm the input channel selected by the customer. The input channels include text input, voice input or image / video input. After the customer submits the request, the system verifies the format of the input content. Among them, voice input needs to be converted into text through the voice recognition module, and image / video input needs to determine the clarity and content recognizability. Through multimodal input channels (text, voice, image / video), the system can flexibly adapt to the different needs and input methods of customers, improving the customer experience. Format verification and processing of input content ensures the accuracy of the data, making subsequent analysis more precise. The parsing and preprocessing mechanisms of text, voice, and images enable the system to efficiently extract important information and avoid information loss. At the same time, these processes provide high-quality data input for subsequent natural language processing and machine learning, ensuring the stability and performance of the system.

[0050] In this embodiment, when parsing the customer's input, first, the customer's text input, voice input, and image input are received through a multimodal input interface and the text input, voice input, and image input are preprocessed respectively. For text data, first, the text data is divided into words, then the part of speech of each word is identified, then the important entities in the text are identified, key information is extracted, and then the sentiment tendency of the text is analyzed, the text in the image is identified, and it is converted into an editable text format. For image / video input, first, the elements in the image are identified using a computer vision algorithm, and valuable information is extracted. The results from NLP and visual analysis are integrated to form a comprehensive understanding of customer input. Finally, the analysis results of NLP and computer vision are combined to extract key information in customer consultation, and the consultation topic and related data are identified. Through natural language processing (NLP) and computer vision technology, the system can efficiently parse customer input, extract key information and identify customer intentions. Establishing a dynamic customer portrait helps to better understand customer behavior and preferences, and thus provide more accurate personalized suggestions for subsequent consultations. The system can adjust the suggestions according to different input modes and judge the customer's emotions through sentiment analysis, which makes the consulting service more intelligent and humane, greatly improving customer satisfaction.

[0051] In this embodiment, after obtaining the consulting topic and related data, it is necessary to construct an intention recognition model to analyze and extract key information, identify the customer's specific intentions and use the context and sentiment analysis results, and then dynamically update the customer portrait to identify the customer's interest areas and consulting mode based on the analysis and intention recognition results. Finally, the key information, intention recognition and customer portrait are integrated to identify the customer's specific needs. The expression is: D = j (K, I, P', M) where D is the identified customer's specific needs, j is a comprehensive analysis function, M is the identified consulting mode, and I is the identified customer intention. After storing customer needs and related information in the database, the system can use machine learning algorithms for real-time analysis to help financial advisors identify customers' potential needs and market trends. This approach not only improves the accuracy of consulting, but also provides customers with forward-looking solutions. Through automated demand forecasting and trend analysis, the system can help customers better cope with future market changes and risks, while reducing manual intervention and improving the efficiency of financial services.

[0052] In this embodiment, the steps of storing the identified customer needs in the question database are:

[0053] Build a database of questions, then store the identified customer needs and related information in the database, then extract and integrate data from different sources to form a unified data set, then extract relevant features from the integrated data and use the Lasso regression algorithm to evaluate the importance of the features, then divide the integrated data set into a training set and a test set, then use the training set data to train the support vector machine model, and then output the prediction results generated by the machine learning model to the system interface to the relevant financial advisor. By creating combined features and using the Lasso regression algorithm to evaluate their importance, the system can identify the most important parts for customer demand prediction from a large number of features. This not only improves the accuracy of the model, but also effectively reduces irrelevant or useless features, thereby simplifying the computational complexity and increasing the training speed of the model. The regularization function of Lasso regression helps to avoid overfitting and enables the model to have stronger generalization capabilities when facing new data.

[0054] In this embodiment, when using the Lasso regression algorithm to evaluate the importance of features, a new combined feature is first created to capture the potential relationship, feature encoding is performed, and the categorical features are converted into numerical form. Then, the formula Calculate feature X k The correlation between the target variable Y and the target variable Y is used to identify the features with high correlation with the target variable. Then, the integrated data set is randomly divided into training set and test set for subsequent model training and evaluation. Subsequently, the Lasso regression model is created and the initial regularization parameter is set. The expression is: Among them, L(β) represents the loss function of Lasso regression, It means to sum all samples, i is the index of the sample, m is the total number of samples, Y i represents the true value or target variable of the i-th sample, represents the predicted value of the i-th sample, calculated by the model, represents the square of the prediction error of the i-th sample, α is the regularization parameter, which controls the weight of the L1 regularization term, represents the sum of all features, j is the index of the feature, n is the total number of features, |βj | represents the absolute value of the regression coefficient corresponding to the jth feature, This is the variable X k Pearson correlation coefficient between X and Y, Cov(X k ,Y) This is X k The covariance of Y and Y measures the direction and degree of common variation of two variables. This is X k The standard deviation of X k The degree of dispersion of the distribution, σ Y This is the standard deviation of Y, which indicates the dispersion of the distribution of Y and measures the fluctuation range of Y;

[0055] After the creation is completed, the Lasso regression model is trained on the training set to learn the relationship between the features and the target variable. Finally, by analyzing the coefficients of each feature in the Lasso model, it is identified which features have a greater impact on the prediction results.

[0056] In this embodiment, after completing data storage and predictive analysis, the steps of using a large language model to generate personalized financial advice related to customer consultation, provide multi-dimensional solutions, and output corresponding consultation answers are as follows:

[0057] Build a large language model and input the sorted information into the large language model. Then, use the large language model to generate personalized financial advice. The large language model will analyze the input information, combine its knowledge base, and output multi-dimensional solutions related to customer consultation. When generating advice, the model is required to provide multiple alternatives, including different risk levels and expected returns. Then, according to the customer's emotional state and preferences, adjust the tone and style of the output content, and then organize the generated personalized advice into an easy-to-understand format and send it to the customer. Through the large language model, the system can generate personalized financial advice for customers and provide multi-dimensional solutions (such as risk level, expected return, etc.) according to different needs. This not only allows customers to choose the right solution according to their own situation, but also enhances customers' trust in the system. By combining sentiment analysis and customer preferences, the system can adjust the style of advice according to the customer's state, provide more humane services, and improve customer satisfaction.

[0058] In this embodiment, the steps of constructing a large language model are:

[0059] According to the Transformer architecture, the number of layers, the number of units in each layer, and the activation function of the model are set. Then, the learning rate, batch size, and number of training rounds are set. Then, the performance of the model is evaluated on the validation set. The perplexity and accuracy are calculated using formulas (1) and (2) for quantitative evaluation. Then, the hyperparameters or training strategies of the model are adjusted based on the evaluation results. Finally, the trained model is deployed to the production environment and the data is further analyzed. In the process of quantitative evaluation, the expression is:

[0060]

[0061] Among them, Perplexity is an indicator to measure the performance of the language model, which indicates the model's ability to predict the test data. exp is a natural exponential function, indicating that the perplexity is obtained by taking the exponent of the average negative log-likelihood. N is the number of samples in the test set or the length of the prediction sequence. The sum symbol indicates the accumulation of all samples, i represents the index of each sample, logP(y i |x ＜i ) represents the i-th target y i Input x in the given preamble ＜i The logarithm of the predicted probability under the condition of i |x ＜i ) is the probability distribution of the model output, Accuracy is the accuracy rate, which is an indicator to measure the prediction accuracy of the classification model, N is the total number of samples in the test set, is the summation symbol, which means the accuracy of all samples is accumulated. Indicator function, when the predicted value y i Equal to the true value y i , returns 1; otherwise returns 0. Based on the Transformer architecture, a large language model is built, and after multiple rounds of tuning, the system can provide customers with more accurate financial advice. By setting the number of layers, units, activation functions, and training parameters of the model, the system can achieve optimal performance in different tasks. Through the evaluation of accuracy and perplexity, the system can dynamically adjust hyperparameters to optimize the output of the model. After deployment, the system can continuously improve its performance by analyzing new data to ensure that customers get the latest and best solutions for their inquiries.

[0062] In this embodiment, after answering the question, the steps of monitoring market changes and user feedback in real time and adjusting the consulting strategy according to the information are as follows:

[0063] Obtain market dynamics. After customers receive consulting advice, provide a feedback mechanism to allow customers to rate, comment or raise questions about the advice. Then, use data analysis tools to analyze real-time market data, identify potential market trends, abnormal fluctuations or risk signals, and classify and sentimentally analyze the collected user feedback to understand customer satisfaction, recognition and demand changes for consulting advice. Then integrate market data and user feedback to form a comprehensive analysis view, and finally set key performance indicators. Compare current indicators with historical data to determine whether there is a significant downward trend and whether it is still suitable for the current market environment and customer needs. Then, apply the adjusted consulting strategy to customer consultation again, collect new user feedback, and form a feedback loop. By monitoring market changes in real time and collecting user feedback, the system can adjust consulting strategies in a timely manner to adapt to the current market environment and customer needs. This not only improves the flexibility of financial services, but also ensures that advice is always based on the latest market dynamics, helping customers make more reasonable decisions. Establishing a feedback loop mechanism ensures continuous optimization of the system, and measures the effectiveness of strategy adjustments through key performance indicators, thereby improving customer experience and business value.

[0064] An intelligent financial consulting system based on a large language model, comprising:

[0065] The customer input module is used to receive financial consulting requests from customers, and then parse the customer input, extract key information and intent, build a customer profile, and identify the customer's specific needs;

[0066] Analysis and demand identification module: The identified customer needs will be stored in the problem database and the machine learning algorithm will be used to automatically identify and predict the customer's potential needs and trends through real-time analysis of historical data, market dynamics and customer behavior;

[0067] Demand storage and machine learning analysis module: After completing data storage and predictive analysis, the module uses a large language model to generate personalized financial advice related to customer consultations, provide multi-dimensional solutions, and output corresponding consultation answers;

[0068] The large language model generation module is used to monitor market changes and user feedback in real time after answering the questions, and adjust the consulting strategy based on this information.

[0069] The present invention is not limited to the above-mentioned embodiments. Anyone should be aware that any structural changes made under the enlightenment of the present invention, and any technical solutions that are the same or similar to the present invention, fall within the protection scope of the present invention. The technology, shape, and structural parts not described in detail in the present invention are all well-known technologies.

Claims

1. An intelligent financial consulting system and method based on a large language model, characterized in that: The following steps are involved: Receive financial consultation requests from customers, then parse the customer input, extract key information and intent, build customer profiles, and identify specific customer needs; The identified customer needs will be stored in the question database and machine learning algorithms will be used to automatically identify and predict customers' potential needs and trends through real-time analysis of historical data, market dynamics and customer behavior; After completing data storage and predictive analysis, the large language model is used to generate personalized financial advice related to customer consultations, provide multi-dimensional solutions, and output corresponding consultation answers; After answering the questions, monitor market changes and user feedback in real time and adjust consulting strategies based on this information.

2. The intelligent financial consulting system and method based on a large language model according to claim 1, characterized in that: The steps for receiving a financial consultation request entered by a client are: First, confirm the input channel selected by the customer. The input channels include text input, voice input or image / video input. After the customer submits the request, the system verifies the format of the input content. Voice input needs to be converted into text through the voice recognition module, and image / video input needs to determine the clarity and content recognizability.

3. The intelligent financial consulting system and method based on a large language model according to claim 2, characterized in that: When parsing customer input, first, receive the customer's text input, voice input, and image input through the multimodal input interface and pre-process the text input, voice input, and image input respectively. For text data, first, segment the text data into words, then identify the part of speech of each word in the word, then identify the important entities in the text, extract key information, and then analyze the sentiment tendency of the text, recognize the text in the image, and convert it into an editable text format. For image / video input, first, use computer vision algorithms to identify elements in the image and extract valuable information. Integrate the results from NLP and visual parsing to form a comprehensive understanding of customer input. Finally, combine the parsing results of NLP and computer vision to extract key information in customer consultations, identify consultation topics and related data.

4. The intelligent financial consulting system and method based on a large language model according to claim 3, characterized in that: After obtaining the consulting topic and related data, it is necessary to analyze the extracted key information by building an intent recognition model, identify the customer's specific intentions and use the context and sentiment analysis results. Then, based on the analysis and intent recognition results, dynamically update the customer portrait to identify the customer's interest areas and consulting patterns. Finally, combine the key information, intent recognition and customer portrait to identify the customer's specific needs. The expression is: D = j(K, I, P′, M), where D is the identified customer's specific needs, j is the comprehensive analysis function, M is the identified consulting pattern, and I is the identified customer intention.

5. The intelligent financial consulting system and method based on a large language model according to claim 4, characterized in that: The steps to identify customer needs and store them in the problem database are: Build a problem database, then store the identified customer needs and related information in the database, then extract and integrate data from different sources to form a unified data set, then extract relevant features from the integrated data and use the Lasso regression algorithm to evaluate the importance of the features, then divide the integrated data set into a training set and a test set, then use the training set data to train the support vector machine model, and then output the prediction results generated by the machine learning model to the system interface for the relevant financial advisor.

6. The intelligent financial consulting system and method based on a large language model according to claim 5, characterized in that: When using the Lasso regression algorithm to evaluate the importance of features, we first create new combined features to capture potential relationships, perform feature encoding, convert categorical features into numerical form, and then use the formula Calculate feature X k The correlation between the target variable Y and the target variable Y is used to identify the features with high correlation with the target variable. Then, the integrated data set is randomly divided into training set and test set for subsequent model training and evaluation. Subsequently, the Lasso regression model is created and the initial regularization parameter is set. The expression is: Among them, L(β) represents the loss function of Lasso regression, It means to sum all samples, i is the index of the sample, m is the total number of samples, Y i represents the true value or target variable of the i-th sample, shows the predicted value of the ith sample, calculated by the model, represents the square of the prediction error of the i-th sample, α is the regularization parameter, which controls the weight of the L1 regularization term, represents the sum of all features, j is the index of the feature, n is the total number of features, |β j | represents the absolute value of the regression coefficient corresponding to the jth feature, ρ This is the variable X k Pearson correlation coefficient between X and Y, Cov(X k ,Y) This is X k The covariance of Y and Y measures the direction and degree of common variation of two variables. This is X k The standard deviation of X k The degree of dispersion of the distribution, σ Y This is the standard deviation of Y, which indicates the dispersion of the distribution of Y and measures the fluctuation range of Y; After the creation is completed, the Lasso regression model is trained on the training set to learn the relationship between the features and the target variable. Finally, by analyzing the coefficients of each feature in the Lasso model, it is identified which features have a greater impact on the prediction results.

7. The intelligent financial consulting system and method based on a large language model according to claim 6, characterized in that: After completing data storage and predictive analysis, the steps to use the big language model to generate personalized financial advice related to customer consultation, provide multi-dimensional solutions, and output the corresponding consultation answers are as follows: Build a big language model and input the sorted information into the big language model. Then, use the big language model to generate personalized financial advice. The big language model will analyze the input information, combine its knowledge base, and output multi-dimensional solutions related to customer consultations. When generating advice, the model is required to provide multiple alternatives, including different risk levels and expected returns. Then, adjust the tone and style of the output content according to the customer's emotional state and preferences. Then, organize the generated personalized advice into an easy-to-understand format and send it to the customer.

8. The intelligent financial consulting system and method based on a large language model according to claim 7, characterized in that: The steps to build a large language model are: According to the Transformer architecture, the number of layers, the number of units in each layer, and the activation function of the model are set. Then, the learning rate, batch size, and number of training rounds are set. Then, the performance of the model is evaluated on the validation set. The perplexity and accuracy are calculated using formulas (1) and (2) for quantitative evaluation. Then, the hyperparameters of the model are adjusted based on the evaluation results. Finally, the trained model is deployed to the production environment and the data is further analyzed. During the quantitative evaluation process, the expression is: Among them, Perplexity is an indicator to measure the performance of the language model, which indicates the model's ability to predict test data. exp This is the natural exponential function, indicating that the perplexity is obtained by taking the exponent of the mean negative log-likelihood, N The number of samples in the test set or the length of the prediction sequence, The sum symbol indicates the accumulation of all samples, i represents the index of each sample, logP(y i |x ＜i ) represents the i-th target y i Input x in the given preamble <i The logarithm of the predicted probability under the condition of i |x <i ) is the probability distribution of the model output, Accuracy is the accuracy rate, which is an indicator to measure the prediction accuracy of the classification model, N is the total number of samples in the test set, is the summation symbol, which means the accuracy of all samples is accumulated. Indicator function, when the predicted value Equal to the true value y i If true, it returns 1; otherwise, it returns 0.

9. The intelligent financial consulting system and method based on a large language model according to claim 8, characterized in that: After answering the questions, monitor market changes and user feedback in real time, and adjust the consulting strategy based on this information: Obtain market dynamics and provide a feedback mechanism after customers receive consulting advice, allowing customers to rate, comment on or raise questions about the advice. Then, use data analysis tools to analyze real-time market data, identify potential market trends, abnormal fluctuations, and risk signals, and classify and perform sentiment analysis on the collected user feedback to understand customers' satisfaction, recognition, and demand changes for consulting advice. Then, integrate market data and user feedback to form a comprehensive analysis view. Finally, set key performance indicators and compare current indicators with historical data to determine whether there is a significant downward trend and whether it is still suitable for the current market environment and customer needs. Then, apply the adjusted consulting strategy again to customer consultation, collect new user feedback, and form a feedback loop.

10. An intelligent financial consulting system based on a large language model, characterized in that: include: The customer input module is used to receive financial consulting requests from customers, and then parse the customer input, extract key information and intent, build a customer profile, and identify the customer's specific needs; Analysis and demand identification module: The identified customer needs will be stored in the problem database and the machine learning algorithm will be used to automatically identify and predict the customer's potential needs and trends through real-time analysis of historical data, market dynamics and customer behavior; Demand storage and machine learning analysis module: After completing data storage and predictive analysis, the module uses a large language model to generate personalized financial advice related to customer consultations, provide multi-dimensional solutions, and output corresponding consultation answers; The large language model generation module is used to monitor market changes and user feedback in real time after answering the questions, and adjust the consulting strategy based on this information.

Images