Large model intelligent customer service method and system based on retrieval enhancement

Abstract

The invention discloses a retrieval enhancement-based large-model intelligent customer service method and system. The method comprises the following steps: receiving a query message; performing intention recognition on the query message, and sending a query message to the target object when it is determined that the intention of the query message is to obtain the recommended financial product; receiving a preference weight input by the target object, and determining a target user portrait corresponding to the target object in a user portrait library according to the identification information of the target object; analyzing the target user portrait, the query message and the preference weight by using a vertical domain intention recognition model; calculating the total investment and the preference weight in the query message by using a multi-objective optimization algorithm to obtain a second recommended financial product combination; and sending the first recommended financial product combination and the second recommended financial product combination to the target object at the same time. The technical problem that in intelligent customer service in the financial field, financial products suitable for the user cannot be accurately recommended to the user based on user requirements is solved.

Description

Technical Field

[0001] This application relates to the technical field of data analysis. Specifically, it relates to a large model intelligent customer service method and system based on retrieval enhancement. Background Art

[0002] In the actual application scenarios of intelligent customer service in the financial field, there are often some unsatisfactory situations. One of the more prominent points is the inability to accurately recommend financial products suitable for users based on the real needs of users. Specifically, when users consult intelligent customer service with various factors such as their specific financial situation, investment goals, and risk tolerance, the intelligent customer service may, due to limitations of algorithms, incomplete data collection, or misunderstandings of user expressions, fail to accurately grasp the core needs of users. For example, a user may be looking for a wealth management product suitable for short-term capital turnover and with low risk, but the intelligent customer service recommends an investment product with a longer term and higher risk; or the user expects the recommended product to match their existing financial asset allocation to achieve the effect of risk diversification and portfolio optimization, but the intelligent customer service fails to consider this and randomly recommends some seemingly popular products that do not fit the user's asset layout, resulting in a large deviation between the recommendation result and the actual needs of the user, affecting the user's satisfaction with the intelligent customer service and the accuracy of product selection in finance.

[0003] For the above problems, no effective solutions have been proposed yet. Summary of the Invention

[0004] This application provides a large model intelligent customer service method and system based on retrieval enhancement to at least solve the technical problem that in the intelligent customer service in the financial field, it is impossible to accurately recommend financial products suitable for users based on user needs.

[0005] According to one aspect of the present application, there is provided a large model intelligent customer service method based on retrieval enhancement, including: receiving a query message input by a target object; performing intent recognition on the query message, and when it is determined that the intent of the query message is to obtain recommended financial products, sending a follow-up message to the target object, where the follow-up message is used to prompt the target object to input preference weights for returns, risks, and terms; receiving the preference weights input by the target object, and determining the target user profile corresponding to the target object in the user profile library according to the identification information of the target object; using a vertical domain intent recognition model to analyze the target user profile, the query message, and the preference weights to obtain a first recommended financial product portfolio output by the vertical domain intent recognition model; using a multi-objective optimization algorithm to calculate the total investment amount and preference weights in the query message to obtain a second recommended financial product portfolio; sending the first recommended financial product portfolio and the second recommended financial product portfolio to the target object at the same time, receiving the selection response message of the target object for the first recommended financial product portfolio and the second recommended financial product portfolio, and updating the vertical domain intent recognition model and the target user profile based on the selection response message.

[0006] Optionally, using a vertical domain intent recognition model to analyze the target user profile, the query message, and the preference weights to obtain a first recommended financial product portfolio output by the vertical domain intent recognition model includes: extracting target values corresponding to preset keywords in the query message, jointly encoding the preset keywords and the target values into numerical features to obtain first numerical features; encoding the target user profile into numerical features to obtain second numerical features, and encoding the preference weights into numerical features to obtain third numerical features; performing weighted splicing on the first numerical features, the second numerical features, and the third numerical features to obtain fused features; determining the matching degree scores between the fused features and candidate products through a neural network ranking model, and determining the top n financial products with the highest matching degree scores as the first recommended financial product portfolio, where the neural network ranking model is a two-tower structure, including: a first feature encoding tower and a second feature encoding tower, where the first feature encoding tower is used to receive the fused features and output a user representation vector, and the second feature encoding tower is used to receive the attribute features of the financial products and output a financial product representation vector; n is a positive integer greater than 1.

[0007] Optionally, a multi-objective optimization algorithm is used to calculate the total investment amount and preference weights in the query message to obtain a second recommended financial product portfolio, including: Step S1, determining a target number of financial products to be recommended; Step S2, generating an initial population, where each individual in the initial population represents a different combination of financial products to be recommended, and each combination includes n financial products to be recommended; Step S3, calculating the objective function of each individual in the current population, where the objective function includes: maximizing a first function value representing revenue, minimizing a second function value representing risk, and minimizing a third function value representing the term; Step S4, performing non-dominated sorting on the individuals, and dividing the current population into different non-dominated levels according to the sorting results; calculating the crowding distance for the individuals in each non-dominated level, and selecting individuals from the current population to enter the breeding pool according to the crowding distance calculation results; performing crossover and mutation operations on the individuals in the breeding pool to generate an offspring population; Step S5, merging the initial population and the offspring population into a target population; performing non-dominated sorting and crowding distance calculation on the target population, and retaining the individuals whose non-dominated rank and crowding degree distance meet the preset requirements to form a new population; repeating Steps S3 to S4 until the maximum number of iterations is reached to obtain the final population; Step S6, determining the target individuals with the highest rank in the non-dominated sorting in the final population, and determining the investment ratio of each financial product to be recommended corresponding to the target individuals as the second recommended financial product portfolio.

[0008] Optionally, determining a target number of financial products to be recommended includes: selecting the top m financial products with the highest matching scores in the neural network sorting model, and determining the top m financial products with the highest matching scores as the financial products to be recommended, where m is the target number and m is a positive integer greater than n.

[0009] Optionally, the method further includes: determining the first function value through the following formula : where is the revenue expectation weight, is the i-th combination of financial products to be recommended, k is the number of combinations of financial products to be recommended, k = , is the expected rate of return of product i, and i is a positive integer not greater than k; determining the second function value through the following formula : where is the risk tolerance weight, is the risk coefficient of product i; determining the third function value through the following formula : where is the risk coefficient of product i, is the term for product i, and D is the user's expected term; the objective function also includes a constraint penalty term, where the constraint penalty term includes: *C is less than the preset minimum investment amount for product i, where C is the total investment amount.

[0010] Optionally, intent recognition of the query message includes: using an intent recognition model to perform intent recognition on the query message to obtain the recognition result output by the intent recognition model, where the recognition result includes: the intent of the query message is to obtain a recommended financial product, and the intent of the query message is not to obtain a recommended financial product; the intent recognition model is trained by the following method: obtaining a user query text data set, annotating the user query text data in the user query text data set to generate a binary classification label including the intent of recommended financial products and the intent of non-recommended financial products, and obtaining the annotated user query text data; generating a feature vector corresponding to the annotated user query text data, where the feature vector includes at least one of the following: a term frequency-inverse document frequency vector generated by the term frequency-inverse document frequency algorithm, a context semantic embedding vector generated based on a pre-trained language model; inputting the feature vector into a neural network classifier including a fully connected layer, using an activation function to output a prediction result, where the prediction result is the probability value of the intent of recommended financial products; judging the relationship between the probability value and a preset threshold, and determining it as the intent of recommended financial products when the probability value is greater than the preset threshold, and determining it as the intent of non-recommended financial products when the probability value is less than the preset threshold; updating the model parameters of the neural network classifier through the backpropagation algorithm according to the error between the prediction result output by the neural network classifier and the annotated label to obtain the intent recognition model.

[0011] Optionally, the user portrait library is constructed by the following method: extracting a user behavior feature set from transaction records, log traces, and historical session records, where the user behavior feature set at least includes: numerical features, time series features, categorical features; performing standardization and normalization processing on the numerical features through the principal component analysis algorithm, extracting the latent vector representation of the time series features through a long short-term memory autoencoder, and mapping the categorical features to a continuous vector space through one-hot encoding to obtain a target feature set; performing clustering processing on the features in the target feature set based on density peaks to obtain a clustering result, and constructing a user portrait library based on the clustering result, where the clustering result carries user group labels.

[0012] According to another aspect of the present application, there is also provided a large model intelligent customer service system based on retrieval enhancement, including: a receiving module, configured to receive a query message input by a target object; an identification module, configured to perform intent identification on the query message, and when it is determined that the intent of the query message is to obtain a recommended financial product, send a follow-up message to the target object, where the follow-up message is used to prompt the target object to input preference weights for returns, risks, and terms; a determination module, configured to receive the preference weights input by the target object, and determine a target user profile corresponding to the target object in the user profile library according to the identification information of the target object; an analysis module, configured to analyze the target user profile, the query message, and the preference weights by using a vertical domain intent recognition model to obtain a first recommended financial product portfolio output by the vertical domain intent recognition model; a calculation module, configured to calculate the total investment amount and preference weights in the query message by using a multi-objective optimization algorithm to obtain a second recommended financial product portfolio; an optimization module, configured to send the first recommended financial product portfolio and the second recommended financial product portfolio to the target object at the same time, receive the selection response message of the target object for the first recommended financial product portfolio and the second recommended financial product portfolio, and update the vertical domain intent recognition model and the target user profile based on the selection response message.

[0013] According to another aspect of the present application, there is also provided a non-volatile storage medium, where the storage medium includes a stored program, and when the program runs, it controls the device where the storage medium is located to execute the above-mentioned large model intelligent customer service method based on retrieval enhancement.

[0014] According to another aspect of the present application, there is also provided an electronic device, including: a memory and a processor, where the processor is configured to run the program stored in the memory, and when the program runs, it executes the above-mentioned large model intelligent customer service method based on retrieval enhancement.

[0015] According to another aspect of the present application, there is also provided a computer program, where when the computer program is executed by a processor, it implements the above-mentioned large model intelligent customer service method based on retrieval enhancement.

[0016] According to another aspect of the present application, there is also provided a computer program product, where the computer program product includes a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the above-mentioned large model intelligent customer service method based on retrieval enhancement.

[0017] In this application, a query message input by a target object is received; the intent of the query message is recognized. When it is determined that the intent of the query message is to obtain recommended financial products, a follow-up message is sent to the target object, where the follow-up message is used to prompt the target object to input preference weights for returns, risks, and terms; the preference weights input by the target object are received, and according to the identification information of the target object, the target user profile corresponding to the target object is determined in the user profile library; a vertical domain intent recognition model is used to analyze the target user profile, the query message, and the preference weights to obtain a first recommended financial product portfolio output by the vertical domain intent recognition model; a multi-objective optimization algorithm is used to calculate the total investment amount and preference weights in the query message to obtain a second recommended financial product portfolio; the first recommended financial product portfolio and the second recommended financial product portfolio are sent to the target object at the same time, the selection response message of the target object for the first recommended financial product portfolio and the second recommended financial product portfolio is received, and based on the selection response message, the vertical domain intent recognition model and the target user profile are updated, achieving the purpose of accurately recommending financial products suitable for users based on user needs, thus realizing the technical effect of enhancing user stickiness, and further solving the technical problem that in intelligent customer service in the financial field, it is impossible to accurately recommend financial products suitable for users based on user needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0019] Figure 1 is a flowchart of a method for intelligent customer service of a large model based on retrieval enhancement according to an embodiment of the present application;

[0020] Figure 2 is a structural diagram of a system for intelligent customer service of a large model based on retrieval enhancement according to an embodiment of the present application;

[0021] Figure 3 is a hardware structure block diagram of a computer terminal for a method for intelligent customer service of a large model based on retrieval enhancement according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0023] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of this application are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application described here can be implemented in an order different from those illustrated or described here. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0024] According to an embodiment of the present application, a method embodiment of an intelligent customer service method based on retrieval enhancement for a large model is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here.

[0025] Figure 1 is a flowchart of an intelligent customer service method based on retrieval enhancement for a large model according to an embodiment of the present application, as Figure 1 shown, the method includes the following steps:

[0026] Step S101, receiving a query message input by a target object.

[0027] Among them, the query message may include information such as the total investment amount and investment type.

[0028] Step S102, performing intent recognition on the query message. When it is determined that the intent of the query message is to obtain a recommended financial product, a follow-up message is sent to the target object, where the follow-up message is used to prompt the target object to input the preference weights for returns, risks and terms.

[0029] The follow-up message can adopt a structured template, such as the following template:

[0030] "To accurately recommend for you, please inform the weight distribution of the following preferences (the total is 100%):

[0031] 1. Return requirement: ____%

[0032] 2. Risk tolerance: ____%

[0033] 3. Investment term: ____%

[0034] Please enter numbers (e.g., 50, 30, 20).

[0035] Specifically, it can be implemented through the following simplified code:

[0036] if intent == "Recommend financial products":

[0037] send_followup_message(template_id = 3) # Trigger the preset follow-up question template

[0038] log_user_state("awaiting_preferences") # Mark the user state as waiting for input.

[0039] Step S103: Receive the preference weights input by the target object, and determine the target user profile corresponding to the target object in the user profile library according to the identification information of the target object.

[0040] Among them, the user profile library includes but is not limited to the following content: Personal identity information: including the user's name, gender, age, ID number, contact information, etc. Occupational information: the user's occupational type (such as white-collar, blue-collar, freelancer, entrepreneur, etc.), the nature of the work unit. Income level: including monthly income, annual salary and other information, as well as the sources of income (salary, bonus, investment income, rental income, etc.). Asset situation: covering the valuation and distribution of the user's real estate, vehicles, financial assets (such as stocks, bonds, funds, bank deposits, etc.). Liability situation: mainly including the amount of the user's loans (mortgage loans, car loans, consumer loans, etc.) and the repayment progress. Investment experience: including the types of financial product investments the user has participated in the past (such as the investment years of stocks, the duration of fund regular investment, etc.), investment success and failure cases, etc. Risk preference: for example, it is divided into conservative, stable, aggressive, etc. Investment goal: clarify whether the user is making short-term investments (such as for buying electronic products, traveling, etc.) or long-term investments. Online behavior data: browsing records on financial institution websites or APPs (such as which financial product pages have been browsed, the types of investment information followed, etc.), search keywords (such as searching for "wealth management products with high annualized yields", etc.). Transaction behavior data: including the financial product transaction records of the user in financial institutions (transaction amount, transaction frequency, transaction channels, etc.). Product preference: record the user's preference degree for different financial products (such as funds, insurance, futures, etc.), including the categories of financial products actively collected, consulted or purchased by the user. Product feedback: including the user's evaluation of the purchased financial products (such as satisfaction, improvement suggestions, etc.).

[0041] Step S104: Analyze the target user profile, query message and preference weights using the vertical domain intent recognition model to obtain the first recommended financial product portfolio output by the vertical domain intent recognition model.

[0042] In step S104, the user profile includes, for example: Static attributes: age, occupation, asset size, risk tolerance rating (such as R1 - R5). Dynamic behaviors: historical investment records (product type, holding period, return performance), and frequency of financial consultation. The query message text can be, for example, "Recommend stable funds" or "High - yield short - term wealth management". The preference weights can be, for example: Return: 50%, Risk: 30%, Term: 20%.

[0043] Use the BERT model fine - tuned in the financial domain to process the query message and output a text embedding vector; after concatenating the numerical features (such as age, assets) and categorical features (such as occupation code) of the user profile, map them to a profile vector through a fully - connected layer; after normalizing the return, risk, and term weight vectors, concatenate them with the text embedding vector and the profile vector and input them into the cross - attention mechanism to generate a joint representation; the fully - connected layer maps the joint representation to a matching score vector in the financial product library; generate a probability distribution through Softmax and select the Top - K products to form the first recommended portfolio.

[0044] In step S105, use a multi - objective optimization algorithm to calculate the total investment amount and preference weights in the query message to obtain the second recommended financial product portfolio.

[0045] In step S106, send the first recommended financial product portfolio and the second recommended financial product portfolio to the target object simultaneously, receive the selection response message of the target object for the first recommended financial product portfolio and the second recommended financial product portfolio, and update the vertical - domain intent recognition model and the target user profile based on the selection response message.

[0046] Among them, the selection response message is used to indicate whether the target object has selected the first recommended financial product portfolio and the second recommended financial product portfolio.

[0047] According to the above steps, receive the query message input by the target object; perform intent recognition on the query message, and when it is determined that the intent of the query message is to obtain recommended financial products, send a follow-up message to the target object, where the follow-up message is used to prompt the target object to input the preference weights for returns, risks, and terms; receive the preference weights input by the target object, and determine the target user profile corresponding to the target object in the user profile library according to the identification information of the target object; use the vertical domain intent recognition model to analyze the target user profile, query message, and preference weights to obtain the first recommended financial product portfolio output by the vertical domain intent recognition model; use the multi-objective optimization algorithm to calculate the total investment amount and preference weights in the query message to obtain the second recommended financial product portfolio; send the first recommended financial product portfolio and the second recommended financial product portfolio to the target object at the same time, receive the selection response message of the target object for the first recommended financial product portfolio and the second recommended financial product portfolio, and update the vertical domain intent recognition model and the target user profile based on the selection response message, achieving the purpose of accurately recommending financial products suitable for users based on user needs, thereby realizing the technical effect of enhancing user stickiness.

[0048] The following gives an exemplary illustration and explanation of Figure 1 the steps shown.

[0049] According to some optional embodiments of the present application, using the vertical domain intent recognition model to analyze the target user profile, query message, and preference weights to obtain the first recommended financial product portfolio output by the vertical domain intent recognition model can be achieved by the following method: extract the target value corresponding to the preset keyword in the query message, encode the preset keyword and the target value into numerical features to obtain the first numerical feature; encode the target user profile into numerical features to obtain the second numerical feature, and encode the preference weights into numerical features to obtain the third numerical feature; perform weighted concatenation on the first numerical feature, the second numerical feature, and the third numerical feature to obtain a fused feature; determine the matching degree score between the fused feature and the candidate products through a neural network ranking model, and determine the financial products with the top n highest matching degree scores as the first recommended financial product portfolio, where the neural network ranking model is a two-tower structure, including: a first feature encoding tower and a second feature encoding tower, where the first feature encoding tower is used to receive the fused feature and output a user representation vector, and the second feature encoding tower is used to receive the attribute features of the financial products and output a financial product representation vector; n is a positive integer greater than 1.

[0050] According to some other alternative embodiments of the present application, the total investment amount and preference weights in the query message are calculated using a multi-objective optimization algorithm to obtain a second recommended financial product portfolio, which can be achieved through the following method: Step S1, determine a target number of financial products to be recommended; Step S2, generate an initial population, where each individual in the initial population represents a different combination of financial products to be recommended, and each combination includes n financial products to be recommended; Step S3, calculate the objective function of each individual in the current population, where the objective function includes: maximizing a first function value representing the return, minimizing a second function value representing the risk, and minimizing a third function value representing the term; Step S4, perform non-dominated sorting on the individuals, and according to the sorting result, divide the current population into different non-dominated levels; calculate the crowding distance for the individuals in each non-dominated level, and according to the crowding distance calculation result, select individuals from the current population to enter the breeding pool; perform crossover and mutation operations on the individuals in the breeding pool to generate an offspring population; Step S5, merge the initial population and the offspring population into a target population; perform non-dominated sorting and crowding distance calculation on the target population, and retain the individuals whose non-dominated rank and crowding degree distance meet the preset requirements to form a new population; repeat Steps S3 to S4 until the maximum number of iterations is reached to obtain the final population; Step S6, determine the target individual with the highest rank in the non-dominated sorting in the final population, and determine the investment ratio of each financial product to be recommended corresponding to the target individual as the second recommended financial product portfolio.

[0051] Among them, Step S4 can be achieved through the following method: Use the Pareto dominance relationship to divide the levels. The non-dominated solutions (not comprehensively dominated by other solutions) are listed in the first layer, the sub-optimal solutions are in the second layer, and so on. Within the same non-dominated level, calculate the distribution density of each individual in the objective space, and preferentially retain the individuals with sparse distribution to maintain diversity. Fill from the highest non-dominated level to the lowest until the population capacity is reached; randomly select the cutting points and exchange part of the products of the parent individuals; replace a certain product with other products in the candidate pool with a preset probability.

[0052] Step S5 can be achieved through the following method: Merge the parent generation (e.g., 200 individuals) and the offspring generation (e.g., 200 individuals) into 400 individuals, and re-perform non-dominated sorting and crowding degree calculation. Select the top 200 individuals from the mixed population in the order of non-dominated level first and crowding degree from high to low as the new generation population. Stop when the maximum number of iterations (e.g., 100 generations) is reached or the Pareto front change rate < threshold (e.g., 0.1%).

[0053] In Step S6, the individual with the largest crowding degree (taking into account both diversity and superiority) can be selected from the first non-dominated layer of the final population.

[0054] The above steps are implemented by an adaptively improved NSGA-II algorithm to accurately determine the effect of the second recommended financial product portfolio.

[0055] Specifically, the target number of financial products to be recommended can be determined by the following method: select the top m financial products with the highest matching scores in the neural network ranking model, and determine the top m financial products with the highest matching scores as the financial products to be recommended, where m is the target number and m is a positive integer greater than n.

[0056] It should be noted that in the above embodiment, it is assumed that the neural network ranking model determines the top 5 financial products with the highest matching scores as the first recommended financial product portfolio. In the multi-objective optimization algorithm, select the top 10 financial products with the highest matching scores. Among these 10 financial products, use the multi-objective optimization algorithm based on risk, return, and term to select the optimal 5 financial products ( ), as the second recommended financial product portfolio. The purpose is to verify the accuracy of the vertical domain intent recognition model in the direction of financial product recommendation to select more financial products that meet the customer's investment preferences. In addition, selecting the top m financial products with the highest matching scores in the neural network ranking model and determining the top m financial products with the highest matching scores as the starting parameters in the multi-objective optimization algorithm can simplify the calculation process of the multi-objective optimization algorithm and improve the calculation efficiency.

[0057] Specifically, the first function value can be determined by the following formula : where is the expected return weight, is the i-th combination of the financial products to be recommended, k is the number of combinations of the financial products to be recommended, k = , is the expected return rate of product i, and i is a positive integer not greater than k; the second function value is determined by the following formula : where is the risk tolerance weight, is the risk coefficient of product i; the third function value is determined by the following formula : where is the risk coefficient of product i, is the term of product i, and D is the user's expected term; the objective function also includes a constraint penalty term, where the constraint penalty term includes: *C is less than the preset minimum investment amount of product i, and C is the total investment amount.

[0058] As some alternative embodiments of the present application, the intent recognition of the query message can be achieved through the following method: Use an intent recognition model to perform intent recognition on the query message to obtain the recognition result output by the intent recognition model. The recognition result includes: the intent of the query message is to obtain a recommended financial product, and the intent of the query message is not to obtain a recommended financial product; the intent recognition model is trained through the following method: Obtain a user query text data set, annotate the user query text data in the user query text data set to generate a binary classification label including the intent of recommending financial products and the intent of not recommending financial products, and obtain the annotated user query text data; Generate a feature vector corresponding to the annotated user query text data. The feature vector includes at least one of the following: a term frequency-inverse document frequency vector generated by the term frequency-inverse document frequency algorithm, and a context semantic embedding vector generated based on a pre-trained language model; Input the feature vector into a neural network classifier including a fully connected layer, and use an activation function to output a prediction result, where the prediction result is the probability value of the intent of recommending financial products; Determine the relationship between the probability value and a preset threshold. When the probability value is greater than the preset threshold, it is determined as the intent of recommending financial products, and when the probability value is less than the preset threshold, it is determined as the intent of not recommending financial products; According to the error between the prediction result output by the neural network classifier and the annotated label, update the model parameters of the neural network classifier through the backpropagation algorithm to obtain the intent recognition model.

[0059] In some alternative embodiments of the present application, the user portrait library is constructed through the following method: Extract a set of user behavior characteristics from transaction records, log traces, and historical session records. The set of user behavior characteristics includes at least: numerical features, time series features, and categorical features; Standardize and normalize the numerical features through the principal component analysis algorithm, extract the latent vector representation of the time series features through a long short-term memory autoencoder, and map the categorical features to a continuous vector space through one-hot encoding to obtain a target feature set; Perform clustering processing on the features in the target feature set based on density peaks to obtain a clustering result, and based on the clustering result, construct a user portrait library, where the clustering result carries user group labels.

[0060] Specifically, when constructing the user portrait library, business labels can be defined according to the common characteristics of features within the cluster, such as: High-net-worth active type: High transaction frequency, large-scale consumption, VIP level. Risk-sensitive type: Frequently querying balance, sessions containing keywords such as "security" and "risk control". Long-term planning type: Regular fixed investment, paying attention to pension financial products. Further, write the user ID, feature vector, and group label into the database to support real-time query.

[0061] Figure 2 is a structural diagram of a large model intelligent customer service system based on retrieval enhancement according to an embodiment of the present application, as Figure 2As shown in the figure, the system includes:

[0062] A receiving module 21, configured to receive a query message input by a target object.

[0063] An identification module 22, configured to perform intent identification on the query message. When it is determined that the intent of the query message is to obtain recommended financial products, a follow-up message is sent to the target object, where the follow-up message is used to prompt the target object to input preference weights for returns, risks, and terms.

[0064] A determination module 23, configured to receive the preference weights input by the target object, and determine the target user profile corresponding to the target object in the user profile library according to the identification information of the target object.

[0065] An analysis module 24, configured to analyze the target user profile, the query message, and the preference weights by using a vertical domain intent recognition model to obtain a first recommended financial product portfolio output by the vertical domain intent recognition model.

[0066] A calculation module 25, configured to calculate the total investment amount and preference weights in the query message by using a multi-objective optimization algorithm to obtain a second recommended financial product portfolio.

[0067] An optimization module 26, configured to send the first recommended financial product portfolio and the second recommended financial product portfolio to the target object at the same time, receive the selection response message of the target object for the first recommended financial product portfolio and the second recommended financial product portfolio, and update the vertical domain intent recognition model and the target user profile based on the selection response message.

[0068] Optionally, analyzing the target user profile, the query message, and the preference weights by using a vertical domain intent recognition model to obtain a first recommended financial product portfolio output by the vertical domain intent recognition model includes the following steps: extracting target values corresponding to preset keywords in the query message, jointly encoding the preset keywords and the target values into numerical features to obtain first numerical features; encoding the target user profile into numerical features to obtain second numerical features, and encoding the preference weights into numerical features to obtain third numerical features; performing weighted concatenation on the first numerical features, the second numerical features, and the third numerical features to obtain fused features; determining the matching degree scores between the fused features and candidate products through a neural network ranking model, and determining the financial products with the top n high matching degree scores as the first recommended financial product portfolio, where the neural network ranking model is a two-tower structure, including: a first feature encoding tower and a second feature encoding tower, where the first feature encoding tower is configured to receive the fused features and output a user representation vector, and the second feature encoding tower is configured to receive the attribute features of the financial products and output a financial product representation vector; n is a positive integer greater than 1.

[0069] Optionally, a multi-objective optimization algorithm is used to calculate the total investment amount and preference weights in the query message to obtain a second recommended financial product portfolio, including the following steps: Step S1, determine the target number of financial products to be recommended; Step S2, generate an initial population, where each individual in the initial population represents a different combination of financial products to be recommended, and each combination includes n financial products to be recommended; Step S3, calculate the objective function of each individual in the current population, where the objective function includes: maximizing the first function value representing the return, minimizing the second function value representing the risk, and minimizing the third function value representing the term; Step S4, perform non-dominated sorting on the individuals, and divide the current population into different non-dominated levels according to the sorting results; calculate the crowding distance for the individuals in each non-dominated level, and select individuals from the current population to enter the breeding pool according to the crowding distance calculation results; perform crossover and mutation operations on the individuals in the breeding pool to generate an offspring population; Step S5, merge the initial population and the offspring population into a target population; perform non-dominated sorting and crowding distance calculation on the target population, and retain the individuals whose non-dominated rank and crowding degree distance meet the preset requirements to form a new population; repeat steps S3 to S4 until the maximum number of iterations is reached to obtain the final population; Step S6, determine the target individual with the highest rank in the non-dominated sorting in the final population, and determine the investment ratio of each financial product to be recommended corresponding to the target individual as the second recommended financial product portfolio.

[0070] Optionally, determining the target number of financial products to be recommended includes the following steps: Select the top m financial products with the highest matching scores in the neural network ranking model, and determine the top m financial products with the highest matching scores as the financial products to be recommended, where m is the target number and m is a positive integer greater than n.

[0071] Optionally, the retrieval-enhanced large model intelligent customer service system is also used to perform the following steps: Determine the first function value through the following formula : Where is the return expectation weight, is the i-th combination of financial products to be recommended, k is the number of combinations of financial products to be recommended, k = , is the expected return rate of product i, and i is a positive integer not greater than k; determine the second function value through the following formula : Where is the risk tolerance weight, is the risk coefficient of product i; determine the third function value through the following formula : Where is the risk coefficient of product i, For the term of product i, D is the user's expected term; the objective function also includes a constraint penalty term, where the constraint penalty term includes: *C is less than the preset minimum investment amount of product i, and C is the total investment amount.

[0072] Optionally, intent recognition of the query message includes: using an intent recognition model to perform intent recognition on the query message to obtain the recognition result output by the intent recognition model, where the recognition result includes: the intent of the query message is to obtain a recommended financial product, and the intent of the query message is not to obtain a recommended financial product; the intent recognition model is trained by the following method: obtaining a user query text data set, annotating the user query text data in the user query text data set to generate a binary classification label including the intent of recommended financial products and the intent of non-recommended financial products, and obtaining the annotated user query text data; generating a feature vector corresponding to the annotated user query text data, where the feature vector includes at least one of the following: a term frequency-inverse document frequency vector generated by the term frequency-inverse document frequency algorithm, a context semantic embedding vector generated based on a pre-trained language model; inputting the feature vector into a neural network classifier including a fully connected layer, and using an activation function to output a prediction result, where the prediction result is the probability value of the intent of recommended financial products; judging the relationship between the probability value and a preset threshold, and when the probability value is greater than the preset threshold, it is determined as the intent of recommended financial products, and when the probability value is less than the preset threshold, it is determined as the intent of non-recommended financial products; updating the model parameters of the neural network classifier through the backpropagation algorithm according to the error between the prediction result output by the neural network classifier and the annotated label to obtain the intent recognition model.

[0073] Optionally, the user portrait library is constructed by the following method: extracting a set of user behavior characteristics from transaction records, log traces, and historical session records, where the set of user behavior characteristics includes at least: numerical features, time series features, categorical features; performing standardization and normalization processing on the numerical features through the principal component analysis algorithm, extracting the latent vector representation of the time series features through a long short-term memory autoencoder, and mapping the categorical features to a continuous vector space through one-hot encoding to obtain a target feature set; performing clustering processing on the features in the target feature set based on density peaks to obtain a clustering result, and constructing a user portrait library based on the clustering result, where the clustering result carries user group labels.

[0074] It should be noted that the above Figure 2 Each module can be a program module (for example, a set of program instructions that implement a specific function), or a hardware module. For the latter, it can be presented in the following forms, but not limited to: the presentation form of each of the above modules is a processor, or the functions of each of the above modules are implemented by a processor.

[0075] It should be noted that Figure 2 For the preferred implementation manners of the illustrated embodiments, reference may be made to Figure 1 the relevant descriptions of the illustrated embodiments, which will not be elaborated herein.

[0076] Figure 3 The hardware structure block diagram of a computer terminal for implementing a large model intelligent customer service method based on retrieval enhancement is shown. As Figure 3 shown, the computer terminal 30 may include one or more processors 302 (shown as 302a, 302b,..., 302n in the figure) (the processor 302 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 304 for storing data, and a transmission module 306 for communication functions. In addition, it may further include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply, and / or a camera. Those of ordinary skill in the art can understand that Figure 3 the structure shown is only schematic and does not limit the structure of the above-mentioned electronic device. For example, the computer terminal 30 may further include more or fewer components than Figure 3 shown therein, or have a different configuration from Figure 3 shown.

[0077] It should be noted that the above one or more processors 302 and / or other data processing circuits are generally referred to as "data processing circuits" herein. The data processing circuit may be embodied in whole or in part as software, hardware, firmware, or any combination thereof. In addition, the data processing circuit may be a single independent processing module, or be incorporated in whole or in part into any one of the other elements in the computer terminal 30. As involved in the embodiments of the present application, the data processing circuit serves as a processor control.

[0078] The memory 304 can be used to store software programs and modules of application software, such as the program instructions / data storage devices corresponding to the large model intelligent customer service method based on retrieval enhancement in the embodiments of the present application. The processor 302 executes various functional applications and data processing by running the software programs and modules stored in the memory 304, that is, implements the above-mentioned large model intelligent customer service method based on retrieval enhancement. The memory 304 may include a high-speed random access memory, and may further include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 304 may further include a memory remotely disposed relative to the processor 302, and these remote memories may be connected to the computer terminal 30 through a network. Examples of the above network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0079] The transmission module 306 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by the communication provider of the computer terminal 30. In one example, the transmission module 306 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission module 306 can be a Radio Frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0080] The display can be, for example, a touch-screen liquid crystal display (LCD), which enables the user to interact with the user interface of the computer terminal 30.

[0081] It should be noted that, in some alternative embodiments, the above Figure 3 illustrated computer terminal may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware elements and software elements. It should be pointed out that Figure 3 is only an example of a specific specific instance and is intended to illustrate the types of components that may exist in the above computer terminal.

[0082] It should be noted that Figure 3 the illustrated computer terminal is used to execute Figure 1 the retrieval-enhanced large model intelligent customer service method shown above. Therefore, the relevant explanations in the execution method of the above commands also apply to this electronic device, and will not be elaborated here.

[0083] The embodiment of the present application also provides a non-volatile storage medium, which includes a stored program. Wherein, when the program runs, it controls the device where the storage medium is located to execute the above retrieval-enhanced large model intelligent customer service method.

[0084] A program for a non - volatile storage medium to perform the following functions: receiving a query message input by a target object; performing intent recognition on the query message, and when it is determined that the intent of the query message is to obtain a recommended financial product, sending a follow - up message to the target object, where the follow - up message is used to prompt the target object to input preference weights for returns, risks, and terms; receiving the preference weights input by the target object, and determining the target user profile corresponding to the target object in the user profile library according to the identification information of the target object; using a vertical - domain intent recognition model to analyze the target user profile, the query message, and the preference weights to obtain a first recommended financial product portfolio output by the vertical - domain intent recognition model; using a multi - objective optimization algorithm to calculate the total investment amount and preference weights in the query message to obtain a second recommended financial product portfolio; sending the first recommended financial product portfolio and the second recommended financial product portfolio to the target object at the same time, receiving the selection response message of the target object for the first recommended financial product portfolio and the second recommended financial product portfolio, and updating the vertical - domain intent recognition model and the target user profile based on the selection response message.

[0085] An embodiment of the present application also provides an electronic device, including: a memory and a processor, where the processor is used to run a program stored in the memory, and when the program runs, it executes the above - mentioned large - model intelligent customer service method based on retrieval enhancement.

[0086] The processor is used to run a program that performs the following functions: receiving a query message input by a target object; performing intent recognition on the query message, and when it is determined that the intent of the query message is to obtain a recommended financial product, sending a follow - up message to the target object, where the follow - up message is used to prompt the target object to input preference weights for returns, risks, and terms; receiving the preference weights input by the target object, and determining the target user profile corresponding to the target object in the user profile library according to the identification information of the target object; using a vertical - domain intent recognition model to analyze the target user profile, the query message, and the preference weights to obtain a first recommended financial product portfolio output by the vertical - domain intent recognition model; using a multi - objective optimization algorithm to calculate the total investment amount and preference weights in the query message to obtain a second recommended financial product portfolio; sending the first recommended financial product portfolio and the second recommended financial product portfolio to the target object at the same time, receiving the selection response message of the target object for the first recommended financial product portfolio and the second recommended financial product portfolio, and updating the vertical - domain intent recognition model and the target user profile based on the selection response message.

[0087] The above serial numbers of the embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0088] In the above - mentioned embodiments of the present application, the descriptions of each embodiment have their own focuses. For parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0089] In the above embodiments of the present application, the collected information is information and data authorized by the user or fully authorized by all parties. Moreover, for the processing of relevant data such as collection, storage, use, processing, transmission, provision, disclosure, and application, all comply with relevant laws, regulations, and standards, necessary protection measures are taken, it does not violate public order and good customs, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0090] In several embodiments provided by the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the described device embodiments are merely illustrative. For example, the division of the units can be a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0091] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0092] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0093] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the relevant technology, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. And the aforementioned storage medium includes: USB flash drives, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), mobile hard disks, magnetic disks, or optical discs and other various media that can store program codes.

[0094] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. An intelligent customer service method for large models based on retrieval enhancement, characterized in that, Including: Receiving a query message input by a target object; Performing intent recognition on the query message, and when it is determined that the intent of the query message is to obtain a recommended financial product, sending a follow-up message to the target object, where the follow-up message is used to prompt the target object to input preference weights for returns, risks, and terms; Receiving the preference weights input by the target object, and determining a target user profile corresponding to the target object in the user profile library according to the identification information of the target object; Analyzing the target user profile, the query message, and the preference weights by using a vertical domain intent recognition model to obtain a first recommended financial product portfolio output by the vertical domain intent recognition model; Calculating the total investment amount and the preference weights in the query message by using a multi-objective optimization algorithm to obtain a second recommended financial product portfolio; Simultaneously sending the first recommended financial product portfolio and the second recommended financial product portfolio to the target object, receiving a selection response message of the target object for the first recommended financial product portfolio and the second recommended financial product portfolio, and updating the vertical domain intent recognition model and the target user profile based on the selection response message.

2. The method according to claim 1, wherein Analyzing the target user profile, the query message, and the preference weights by using a vertical domain intent recognition model to obtain a first recommended financial product portfolio output by the vertical domain intent recognition model, including: Extracting a target value corresponding to a preset keyword in the query message, and jointly encoding the preset keyword and the target value into a numerical feature to obtain a first numerical feature; Encoding the target user profile into a numerical feature to obtain a second numerical feature, and encoding the preference weights into a numerical feature to obtain a third numerical feature; Performing weighted splicing on the first numerical feature, the second numerical feature, and the third numerical feature to obtain a fusion feature; Determining the matching degree score between the fusion feature and candidate products through a neural network ranking model, and determining the financial products with the top n highest matching degree scores as the first recommended financial product portfolio, where the neural network ranking model is a two-tower structure, including: a first feature encoding tower and a second feature encoding tower, where the first feature encoding tower is used to receive the fusion feature and output a user representation vector, and the second feature encoding tower is used to receive the attribute features of financial products and output a financial product representation vector; n is a positive integer greater than 1.

3. The method according to claim 2, wherein Calculating the total investment amount and the preference weights in the query message by using a multi-objective optimization algorithm to obtain a second recommended financial product portfolio, including: Step S1, determining a target number of financial products to be recommended; Step S2, generating an initial population, where each individual in the initial population represents a different combination of financial products to be recommended, and each combination includes n financial products to be recommended; Step S3: Calculate the objective function for each individual in the current population. The objective function includes: maximizing the first function value representing the return, minimizing the second function value representing the risk, and minimizing the third function value representing the term. Step S4: Perform non-dominated sorting on the individuals, and divide the current population into different non-dominated levels according to the sorting results. Calculate the crowding distance for the individuals in each non-dominated level, and select individuals from the current population to enter the breeding pool according to the crowding distance calculation results. Perform crossover and mutation operations on the individuals in the breeding pool to generate an offspring population. Step S5: Combine the initial population and the offspring population into a target population. Perform non-dominated sorting and crowding distance calculation on the target population, and retain the individuals whose non-dominated rank and crowding distance meet the preset requirements to form a new population. Repeat steps S3 to S4 until the maximum number of iterations is reached to obtain the final population. Step S6: Determine the target individuals with the highest rank in the non-dominated sorting in the final population, and determine the investment ratio of each of the to-be-recommended financial products corresponding to the target individuals as the second recommended financial product portfolio.

4. The method according to claim 3, characterized in that Determine the target number of to-be-recommended financial products, including: selecting the top m financial products with the highest matching scores in the neural network ranking model, and determining the top m financial products with the highest matching scores as the to-be-recommended financial products, where m is the target number and m is a positive integer greater than n.

5. The method according to claim 4, wherein The method further includes: The first function value is determined by the following formula :[[]]END]] ， Among them, is the expected return weight, is the i-th combination of the financial products to be recommended, k is the number of combinations of the financial products to be recommended, k = , is the expected return rate of product i, and i is a positive integer not greater than k; Determine the second function value through the following formula :[[]]END]] ， Among them, is the risk tolerance weight, is the risk coefficient of product i; The third function value is determined by the following formula :[[]]END]] ， Among them, is the risk coefficient of product i, is the term of product i, and D is the user's expected term; The objective function further includes a constraint penalty term, where the constraint penalty term includes: *C is less than the preset minimum investment amount of product i, and C is the total investment amount.

6. The method according to claim 1, characterized in that, Performing intent recognition on the query message, including: Using an intent recognition model to perform intent recognition on the query message to obtain the recognition result output by the intent recognition model, where the recognition result includes: the intent of the query message is to obtain a recommended financial product, and the intent of the query message is not to obtain a recommended financial product. The intent recognition model is trained by the following method: Obtain a user query text dataset, annotate the user query text data in the user query text dataset to generate a binary classification label including the intent of recommended financial products and the intent of non-recommended financial products, and obtain the annotated user query text data. Generate feature vectors corresponding to the annotated user query text data, where the feature vectors include at least one of the following: a term frequency-inverse document frequency vector generated by the term frequency-inverse document frequency algorithm, and a context semantic embedding vector generated based on a pre-trained language model. Input the feature vectors into a neural network classifier including a fully connected layer, and use an activation function to output a prediction result, where the prediction result is the probability value of the intent of recommended financial products. Judge the relationship between the probability value and a preset threshold. When the probability value is greater than the preset threshold, it is determined as the intent of recommended financial products, and when the probability value is less than the preset threshold, it is determined as the intent of non-recommended financial products. Update the model parameters of the neural network classifier through the backpropagation algorithm according to the error between the prediction result output by the neural network classifier and the annotated label to obtain an intent recognition model.

7. The method according to claim 1, characterized in that, The user profile library is constructed by the following method: Extract a user behavior feature set from transaction records, log traces, and historical session records, where the user behavior feature set at least includes: numerical features, time series features, and categorical features; Perform standardization and normalization processing on the numerical features through the principal component analysis algorithm, extract the latent vector representation of the time series features through the long short-term memory autoencoder, and map the categorical features to a continuous vector space through one-hot encoding to obtain a target feature set; Perform clustering processing on the features in the target feature set based on density peaks to obtain a clustering result, and construct the user profile library based on the clustering result, where the clustering result carries user group labels.

8. An intelligent customer service system based on retrieval enhancement for large models, characterized in that, It includes: A receiving module for receiving a query message input by a target object; An identification module for performing intent identification on the query message, and when it is determined that the intent of the query message is to obtain recommended financial products, sending a follow-up message to the target object, where the follow-up message is used to prompt the target object to input preference weights for returns, risks, and terms; A determination module for receiving the preference weights input by the target object and determining the target user profile corresponding to the target object in the user profile library according to the identification information of the target object; An analysis module for analyzing the target user profile, the query message, and the preference weights by using a vertical domain intent recognition model to obtain a first recommended financial product portfolio output by the vertical domain intent recognition model; A calculation module for calculating the total investment amount and the preference weights in the query message by using a multi-objective optimization algorithm to obtain a second recommended financial product portfolio; An optimization module for simultaneously sending the first recommended financial product portfolio and the second recommended financial product portfolio to the target object, receiving a selection response message of the target object for the first recommended financial product portfolio and the second recommended financial product portfolio, and updating the vertical domain intent recognition model and the target user profile based on the selection response message.

9. A non-volatile storage medium, characterized in that, The non-volatile storage medium includes a stored program, where when the program runs, it controls the device where the non-volatile storage medium is located to execute the retrieval-enhanced large model intelligent customer service method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes: A memory and a processor, where the processor is used to run the program stored in the memory, and when the program runs, it executes the retrieval-enhanced large model intelligent customer service method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the retrieval-enhanced large model intelligent customer service method according to any one of claims 1 to 7.

Images