Intelligent reply method for online consultation factoring service

Abstract

The invention relates to the technical field of data analysis, is suitable for the field of enterprise finance, and provides an intelligent reply method for online consultation of factoring business, and the method comprises the steps: obtaining the real-time consultation information of a client for the factoring business; based on the real-time consultation information and a preset effective input information obtaining mode, effective input information corresponding to the real-time consultation information is determined; updating a user portrait of the client based on the effective input information; in a preset factoring service knowledge graph, factoring service association information associated with the effective input information and the updated user portrait is searched; and determining multivariate structured data consisting of the effective input information, the updated user portrait and factoring service association information as model input information of a large language model, and outputting real-time feedback information corresponding to the model input information and aiming at the real-time consultation information through the large language model. According to the technical scheme, the real-time response accuracy of factoring business consultation can be improved.

Description

Technical field

[0001] The present application relates to the field of data analysis technology, is applicable to the field of corporate finance, and particularly relates to an intelligent reply method for online consultation on factoring business. [Background Technology]

[0002] Currently, online factoring consultation services in the corporate finance sector are generally challenged by low response efficiency and insufficient service standardization. Traditional manual customer service models, limited by processing power, struggle to parse multimodal customer inquiries—text, images, audio, and more—in real time, leading to delayed responses and inconsistent interpretation. Furthermore, static customer profiles fail to dynamically reflect evolving user needs, resulting in inaccurate recommendations. Furthermore, existing systems rely on manually constructed business knowledge bases, making it difficult to quickly associate complex business rules and resulting in low knowledge reuse.

[0003] Therefore, how to achieve real-time and accurate response to factoring business consultation through intelligent technology has become a technical problem that needs to be solved urgently. [Summary of the invention]

[0004] The embodiment of the present application provides an intelligent reply method for online consultation on factoring business, which aims to solve the technical problems in related technologies such as low efficiency of manual response, insufficient multimodal consultation information processing capabilities, delayed customer portrait updates, and weak relevance of business knowledge, which lead to insufficient accuracy of real-time responses to factoring business consultations.

[0005] In a first aspect, an embodiment of the present application provides an intelligent reply method for online consultation on factoring business, comprising:

[0006] Obtain real-time consulting information from clients regarding factoring business;

[0007] Determining valid input information corresponding to the real-time consultation information based on the real-time consultation information and a preset valid input information acquisition method;

[0008] Based on the valid input information, updating the user profile of the client;

[0009] Searching for factoring business-related information associated with the valid input information and the updated user profile in a preset factoring business knowledge graph;

[0010] Determine that the multivariate structured data consisting of the valid input information, the updated user portrait and the factoring business related information is the model input information of the large language model, and output the real-time feedback information corresponding to the model input information for the real-time consultation information through the large language model.

[0011] In one embodiment of the present application, optionally, determining the valid input information corresponding to the real-time consultation information based on the real-time consultation information and a preset valid input information acquisition method includes:

[0012] If the real-time consultation information includes text information, determining a semantic text of the text information based on an intention recognition model, and adding the semantic text to the valid input information;

[0013] If the real-time consultation information includes picture information, parsing the content text of the picture information based on the CLIP model, and adding the content text to the valid input information;

[0014] If the real-time consultation information includes audio information, the audio information is transcribed into audio text based on the Whisper model, and the audio text is added to the valid input information.

[0015] In one embodiment of the present application, optionally, before obtaining the real-time consulting information of the client regarding the factoring business, the method further includes:

[0016] Extract factoring business entities and inter-entity relationships from a factoring business information database using a BERT+BiLSTM-CRF hybrid neural network model. The factoring business entities include: factoring type, business recommendation language, risk control indicators, and legal terms.

[0017] The factoring business knowledge graph is constructed with the factoring business entities as nodes, the relationships between the entities as edges between the nodes, and the inter-node correlation reflected by the relationships between the entities as weights of the edges.

[0018] In one embodiment of the present application, optionally, the method further includes:

[0019] Performing sentiment analysis on the valid input information to determine real-time user sentiment information of the client;

[0020] Before the large language model outputs the real-time feedback information corresponding to the model input information and for the real-time consultation information, the method further includes:

[0021] The real-time user emotion information is added to the model input information.

[0022] In one embodiment of the present application, optionally, the method further includes:

[0023] Performing sentiment analysis on the valid input information to determine real-time user sentiment information of the client;

[0024] After updating the user profile of the client based on the valid input information and before searching for factoring business-related information associated with the valid input information and the updated user profile, the method further includes:

[0025] Based on the real-time user emotion information, an emotion tag is added to the updated user portrait.

[0026] In one embodiment of the present application, optionally, the method further includes:

[0027] Determining the inquiry type of the real-time consultation information based on the valid input information, wherein the inquiry type includes a poster requirement type and a text requirement type;

[0028] The real-time feedback information is displayed in an information display manner corresponding to the inquiry type, wherein:

[0029] If the inquiry type is the poster requirement type, based on the valid input information, determining a poster template from a poster template set that is adapted to the current factoring business content inquired about by the real-time consultation information;

[0030] Filling the real-time feedback information into the poster template to generate a real-time feedback poster, and displaying the real-time feedback poster on the client;

[0031] If the inquiry type is the text requirement type, the real-time feedback information is displayed on the client.

[0032] In one embodiment of the present application, optionally, the method further includes:

[0033] Obtaining a score of the real-time feedback information by the client as the accuracy of the real-time feedback information;

[0034] If the accuracy is lower than a predetermined accuracy threshold, the current intelligent reply mode is switched to the manual reply mode, and a manual customer service representative whose capability attribute information matches the real-time consultation information is assigned to the client.

[0035] In one embodiment of the present application, optionally, the method further includes:

[0036] Determining the accuracy of the real-time feedback information based on a preset feedback information accuracy evaluation rule and the real-time feedback information;

[0037] If the accuracy is lower than a predetermined accuracy threshold, the current intelligent reply mode is switched to the manual reply mode, and a manual customer service representative whose capability attribute information matches the real-time consultation information is assigned to the client.

[0038] In a second aspect, an embodiment of the present application provides an intelligent reply device for online consultation on factoring business, comprising:

[0039] A real-time consulting information acquisition unit, used to acquire real-time consulting information from the client regarding factoring business;

[0040] an effective input information determining unit, configured to determine effective input information corresponding to the real-time consultation information based on the real-time consultation information and a preset effective input information obtaining method;

[0041] A user portrait updating unit, configured to update the user portrait of the client based on the valid input information;

[0042] A knowledge graph search unit, configured to search a preset factoring business knowledge graph for factoring business-related information associated with the valid input information and the updated user profile;

[0043] The large language model processing unit is used to determine that the multivariate structured data composed of the valid input information, the updated user portrait and the factoring business related information is the model input information of the large language model, and output the real-time feedback information corresponding to the model input information for the real-time consultation information through the large language model.

[0044] In one embodiment of the present application, optionally, the valid input information determining unit includes:

[0045] an intention recognition unit, configured to, if the real-time consultation information includes text information, determine a semantic text of the text information based on an intention recognition model, and add the semantic text to the valid input information;

[0046] An image parsing unit, configured to parse the content text of the image information based on a CLIP model if the real-time consultation information includes image information, and add the content text to the valid input information;

[0047] An audio transcription unit is used to transcribe the audio information into audio text based on a Whisper model if the real-time consultation information includes audio information, and to add the audio text to the valid input information.

[0048] In one embodiment of the present application, optionally, the device further includes:

[0049] an entity and inter-entity relationship extraction unit, configured to extract factoring business entities and inter-entity relationships from a factoring business information database based on a BERT+BiLSTM-CRF hybrid neural network model before the real-time consultation information acquisition unit acquires the real-time consultation information, wherein the factoring business entities include: factoring type, business recommendation language, risk control indicators, and legal terms;

[0050] The knowledge graph construction unit is used to construct the factoring business knowledge graph with the factoring business entities as nodes, the relationships between the entities as edges between the nodes, and the correlation between the nodes reflected by the relationships between the entities as the weights of the edges.

[0051] In one embodiment of the present application, optionally, the device further includes:

[0052] An emotion recognition unit, configured to perform emotion analysis on the valid input information to determine real-time user emotion information of the client;

[0053] The model input information updating unit is used to add the real-time user emotion information to the model input information before the large language model processing unit outputs the real-time feedback information.

[0054] In one embodiment of the present application, optionally, the device further includes:

[0055] An emotion recognition unit, configured to perform emotion analysis on the valid input information to determine real-time user emotion information of the client;

[0056] The user portrait re-updating unit is used to add an emotion tag to the updated user portrait based on the real-time user emotion information after the user portrait updating unit updates the user portrait of the client and before the knowledge graph search unit searches for factoring business related information.

[0057] In one embodiment of the present application, optionally, the device further includes:

[0058] an inquiry type determining unit, configured to determine an inquiry type of the real-time consultation information based on the valid input information, wherein the inquiry type includes a poster requirement type and a text requirement type;

[0059] An information display unit is configured to display the real-time feedback information in an information display manner corresponding to the inquiry type, wherein the information display unit includes:

[0060] a poster template extraction unit configured to, if the inquiry type is the poster requirement type, determine, based on the valid input information, a poster template from a poster template set that is adapted to the current factoring business content consulted by the real-time consultation information;

[0061] a poster generating unit, configured to fill the real-time feedback information into the poster template, generate a real-time feedback poster, and display the real-time feedback poster on the client;

[0062] A text display unit is used to display the real-time feedback information on the client if the inquiry type is the text requirement type.

[0063] In one embodiment of the present application, optionally, the device further includes:

[0064] a client evaluation unit, configured to obtain a score of the client on the real-time feedback information as the accuracy of the real-time feedback information;

[0065] The reply mode switching unit is used to switch from the current intelligent reply mode to the manual reply mode if the accuracy is lower than a predetermined accuracy threshold, and to assign a manual customer service representative whose capability attribute information matches the real-time consultation information to the client.

[0066] In one embodiment of the present application, optionally, the device further includes:

[0067] an automatic evaluation unit, configured to determine the accuracy of the real-time feedback information based on a preset feedback information accuracy evaluation rule and the real-time feedback information;

[0068] The reply mode switching unit is used to switch from the current intelligent reply mode to the manual reply mode if the accuracy is lower than a predetermined accuracy threshold, and to assign a manual customer service representative whose capability attribute information matches the real-time consultation information to the client.

[0069] In a third aspect, an embodiment of the present application provides a computer device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are configured to execute the method described in the first aspect above.

[0070] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the method described in the first aspect above.

[0071] The above technical solution addresses the technical challenges of inaccurate real-time responses to factoring consultations, such as inefficient manual responses, insufficient multimodal consultation information processing capabilities, delayed customer profile updates, and weak business knowledge relevance. This solution enables intelligent response processing throughout the entire factoring consultation process. Specifically, it captures multimodal consultation information submitted by customers in real time and uses large language models, cross-modal understanding, and speech recognition technologies to accurately analyze customer intent and construct a complete and valid input. Simultaneously, it dynamically updates user profiles based on this information to understand evolving customer needs in real time. It also leverages pre-built expertise graphs to quickly link relevant business rules and terms, forming a structured knowledge network. Ultimately, it integrates customer needs, real-time profiles, and expertise, generating professional and personalized responses using large language models. This end-to-end intelligent processing mechanism significantly improves service efficiency and quality, ensuring that responses are both compliant with business regulations and relevant to customers' actual situations, providing customers with a precise and efficient professional consulting service experience.

Brief Description of the Drawings

[0072] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0073] Figure 1 A flowchart of an intelligent reply method for online consultation on factoring business according to one embodiment of the present application is shown;

[0074] Figure 2 A flowchart of an intelligent reply method for online consultation on factoring business according to another embodiment of the present application is shown;

[0075] Figure 3 A block diagram of a computer device according to an embodiment of the present application is shown;

[0076] Figure 4 A block diagram of a computer device according to another embodiment of the present application is shown. [Specific implementation method]

[0077] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0078] Figure 1A flow chart of an intelligent reply method for online consultation on factoring business according to an embodiment of the present application is shown.

[0079] like Figure 1 As shown, according to one embodiment of the present application, an intelligent reply method for online consultation on factoring business includes:

[0080] Step 102: Obtain real-time consulting information from the client regarding factoring business.

[0081] The user manually enters or selects real-time consultation information for their desired factoring business within the client's real-time dialog box. The client then uploads this real-time consultation information to the server's intelligent response system for online factoring consultations. This real-time consultation information reflects the client's need for information on the factoring business they are seeking.

[0082] By acquiring factoring business inquiries submitted by users in real time on the client side, the system can instantly capture the customer's specific needs and provide accurate data sources for subsequent intelligent responses. Users enter or select inquiries through a dialog box, ensuring intuitive and convenient expression of needs. The system's reception and processing of real-time inquiries ensures a seamless connection between customer needs and intelligent services.

[0083] Step 104 : Based on the real-time consultation information and a preset method for obtaining valid input information, determine the valid input information corresponding to the real-time consultation information.

[0084] Because real-time consultation information reflects the client's need for information about the factoring business they desire, valid input information that more accurately reflects the client's user consultation needs can be extracted from the real-time consultation information to serve as the basis for generating responses that meet the user's consultation needs. Specifically, various preset methods for obtaining valid input information can be provided for different types of real-time consultation information.

[0085] This allows us to accurately identify key content reflecting the customer's core needs from real-time consultation information through pre-defined and effective information extraction methods. This step ensures targeted follow-up processing, filtering out irrelevant information and focusing on the customer's true inquiry intent, laying the foundation for generating high-quality responses. Multiple pre-defined information acquisition methods allow for flexible response to different types of consultation content, enhancing the system's adaptability.

[0086] In one possible design, if the real-time consultation information includes text information, the semantic context of the text information is determined based on the intent recognition model, and the semantic context is added to the valid input information. Optionally, the intent recognition model is a large language model, optionally a GPT-3.5-turbo model.

[0087] If the real-time consultation information includes image information, the content text of the image information is parsed based on the CLIP model and the content text is added to the valid input information. The CLIP model (Contrastive Language–Image Pretraining) is a multimodal pretraining model that maps images and text into the same semantic space through contrastive learning, enabling cross-modal semantic understanding and retrieval.

[0088] If the real-time consultation information includes audio information, the audio information is transcribed into audio text based on the Whisper model, and the audio text is added to the valid input information. The Whisper model is an automatic speech recognition model used to convert audio information into text. It supports multiple languages ​​and can handle speech recognition and translation tasks simultaneously.

[0089] The above design achieves a comprehensive analysis of customer consultation information through multimodal intelligent processing technology. When the consultation information is text, a large language model can be used to accurately identify the text semantics, accurately capture customer intent, and ensure that the deeper meaning of the text information is fully extracted. For image-based consultation information, an advanced cross-modal understanding model is used to convert visual content into semantic text, breaking through the limitations of traditional systems' ability to process image information. For audio input, high-precision speech recognition technology is used to achieve real-time conversion from speech to text, eliminating voice communication barriers. These three processing methods work together to jointly construct complete and valid input information, providing a comprehensive and accurate data foundation for subsequent intelligent responses, significantly improving the system's adaptability to complex consultation scenarios. The multimodal fusion processing mechanism ensures both efficient analysis of various types of information and consistency in semantic understanding, providing customers with a more accurate and natural interactive experience.

[0090] Step 106: Update the user profile of the client based on the valid input information.

[0091] Valid input information is extracted from the client's real-time consultation information and accurately reflects the client's real-time consultation intentions and needs for factoring services. Therefore, valid input information reflects, to a certain extent, the client's current user profile, which differs from the client's historical user profile in the system based on the real-time consultation intentions and needs. For example, valid input information can reflect the urgency of the client's funding needs, the role of the current inquiring party in the client's enterprise's decision-making chain, and the correlation between the client's real-time consultation intentions and needs and their sensitivity to industry policies. All of this information reflected in valid input information reflects the client's current characteristics and, compared to historical user profiles, better reflects the client's real-time capability level and characteristic performance. Therefore, the client's user profile can be updated based on the valid input information, making the current user profile more accurately and reliably reflect the client's real-time capability level and characteristic performance. Since factoring services for different clients have different processing standards, and these processing standards dynamically change based on the urgency of the user's current needs, the real-time updated user profile can be used together with the valid input information as the basis for providing corresponding answers to real-time consultation information.

[0092] Real-time updates to user profiles based on valid input provide a dynamic understanding of customers' latest characteristics and evolving needs. By analyzing key dimensions such as the urgency of funding needs and roles within the decision-making chain, a more accurate profile of the customer's current status can be constructed, providing a basis for personalized service. This real-time update mechanism ensures the system's sensitivity to changing customer needs, significantly improving the timeliness and accuracy of service delivery.

[0093] Step 108: Search the preset factoring business knowledge graph for factoring business related information that is associated with the valid input information and the updated user profile.

[0094] The preset factoring business knowledge graph includes various entities involved in factoring business and the relationships between entities. Based on valid input information and updated user portraits, entities and relationships between entities associated with real-time consulting information can be searched in the preset factoring business knowledge graph.

[0095] By leveraging a pre-defined factoring business knowledge graph, we intelligently link customer needs and real-time profiles with specialized knowledge. By searching for matching entities and relationships, we can quickly locate specialized knowledge nodes relevant to customer inquiries, providing sufficient knowledge support for generating professional responses. This graph-based retrieval method ensures professionalism while improving information acquisition efficiency.

[0096] Of course, before step 102, a factoring business knowledge graph needs to be constructed in advance, and its construction method includes: extracting factoring business entities and inter-entity relationships in the factoring business information database based on the BERT+BiLSTM-CRF hybrid neural network model, wherein the factoring business entities include: factoring type, business recommendation language, risk control indicators and legal terms; taking the factoring business entities as nodes, the inter-entity relationships as edges between the nodes, and the inter-node correlation reflected by the inter-entity relationships as the weights of the edges, to construct the factoring business knowledge graph.

[0097] By constructing a knowledge graph for factoring business, a deep structural processing of professional knowledge is achieved. Using advanced hybrid neural network technology, key business elements are accurately identified from massive amounts of business data, including the specific characteristics of various factoring products, standard business processing scripts, core risk management indicators, and relevant legal provisions. By converting these business entities into knowledge nodes, using the logical relationships between entities as connecting links, and introducing association strength weights, the system establishes a hierarchical and closely related professional knowledge network. This knowledge organization method not only fully preserves the inherent connections between business rules, but also intuitively displays the importance of different business elements through quantified weight values, providing a solid knowledge foundation for subsequent intelligent consulting services. The construction of the knowledge graph enables the system to quickly locate business key points, accurately understand the connections between complex business rules, and significantly improve the professionalism and accuracy of intelligent responses.

[0098] Step 110: Determine that the multivariate structured data consisting of the valid input information, the updated user portrait, and the factoring business related information is the model input information of the large language model, and output the real-time feedback information corresponding to the model input information for the real-time consultation information through the large language model.

[0099] This structured, multi-faceted data is fed into a large language model to generate professional responses to customer inquiries. By integrating customer needs, real-time profiles, and specialized knowledge, the system can deliver responses that are both compliant with business regulations and relevant to the customer's specific situation. This end-to-end intelligent processing fully automates the entire process from customer inquiry to professional response, significantly improving service efficiency and quality.

[0100] Specifically, the multi-structured data is structured with valid input information as the vertex, the updated user profile as the first branch of the vertex, and factoring business-related information as the second branch shared by the vertex and the first branch. This complex tree-like data structure enables intelligent integration and deep correlation of consulting information. With the customer's core needs as the data vertex, the real-time updated user profile is first used as the first-level branch to accurately reflect the customer's current characteristics and changing needs. On this basis, matching professional factoring business knowledge is further added as the second-level branch to form a complete knowledge-related network. This hierarchical data structure not only maintains the centrality of customer needs but also organically integrates user characteristics and professional knowledge. This data structure improves the logical organization of information and retrieval efficiency. When faced with complex consulting scenarios, it can quickly construct the optimal solution path, significantly enhancing the pertinence and practicality of intelligent responses.

[0101] It should be noted that during the knowledge graph construction process, edge weights can be calculated by analyzing the co-occurrence frequency, semantic similarity, and logical association strength defined by business rules between entities in business data. Specifically, when two business entities frequently co-occur in a large number of business documents, their first association weight will be increased accordingly. This means that the frequency of co-occurrence of the two business entities in the business can be set proportional to their first association weight. If there is a clear process dependency between the entities, a corresponding second association weight can be set based on the depth of the process dependency. Simultaneously, the semantic analysis model is combined to determine the relevance of the entities in the context, and this relevance is used as the third association weight. Finally, the weighted sum of the first, second, and third association weights is used as the edge weight. The sum of the first, second, and third association weights represents their respective contributions to historical consultation.

[0102] The technical solution of the present application is further explained below through examples in actual scenarios. Of course, the following are only examples. The technical solution of the present application can be applied to multiple scenarios of factoring business and even any other online consulting business, without being limited by the following examples.

[0103] Manager Zhang, a finance manager at a small and medium-sized enterprise, inquired about factoring through WeChat Work. He uploaded a picture of a purchase contract and left a voice message: "We have a 5 million yuan account receivable for medical devices. The buyer is a tertiary hospital. Can we do non-recourse factoring? What's the approximate interest rate?"

[0104] For the above real-time consultation information, the system receives dual-modal input of images and voice, and parses the procurement contract image through the CLIP model to extract key information such as "Purchaser: XX City First Hospital (Grade 3)", "Contract Amount: 5 million yuan", and "Payment Period: 180 days". At the same time, the voice is transcribed into text through the Whisper model, and the GPT-3.5-turbo large language model is used to identify the two core intentions in the real-time consultation information: 1. Confirm the feasibility of non-recourse factoring; 2. Inquire about the interest rate range.

[0105] Next, the system extracts fields such as "Grade 3 Hospital" and "Medical Devices" as valid input information, automatically marking the customer as a "High-Quality Core Enterprise Supply Chain." Based on the 5 million yuan amount and 180-day payment period, the system determines the urgency of the funding need to be "Medium." Based on this, the user profile tags are updated to: "Industry: Healthcare," "Buyer Credit Rating: AAA," and "Payment Period Preference: Medium to Short-Term."

[0106] Furthermore, relevant nodes with valid input information are searched in the factoring knowledge graph, including but not limited to the following examples:

[0107] Entity node: "Non-recourse Factoring" → associated with "Medical Industry Applicable Terms";

[0108] Relationship edge: The weight coefficient between "Grade 3 Hospitals" and "Factoring Rate" is 0.92 (strong correlation);

[0109] At the same time, the association rules were retrieved: medical equipment accounts receivable can be factored without recourse; the benchmark interest rate for buyers of tertiary hospitals is LPR+1.2%~1.8%.

[0110] Based on the above content and structured data, the system generates the following response using a large language model:

[0111] Dear customer, based on the information you provided:

[0112] Meet the conditions for non-recourse factoring (the buyer is a tertiary hospital with high-quality credit);

[0113] The current factoring interest rate range for the medical industry is: 4.35% to 4.95% (refer to LPR floating);

[0114] Recommended solution: You can apply for a credit line of RMB 6 million. It is recommended to provide: ① Hospital purchase confirmation letter ② Accounts receivable details.

[0115] In summary, this technical solution addresses the technical problems in related technologies such as low manual response efficiency, insufficient multimodal consulting information processing capabilities, delayed customer profile updates, and weak business knowledge relevance, which lead to insufficient accuracy in real-time responses to factoring business consultations. It achieves full-process intelligent response processing for factoring business consulting services. Specifically, it obtains multimodal consulting information submitted by customers in real time, uses large language models, cross-modal understanding, and speech recognition technologies to accurately analyze customer intentions and construct complete and effective input information. At the same time, based on this information, it dynamically updates user profiles to grasp changes in customer needs in real time. It can also quickly associate relevant business rules and terms through pre-built professional knowledge maps to form a structured knowledge network. Ultimately, it integrates customer needs, real-time profiles, and professional knowledge to generate professional and personalized responses through large language models.

[0116] This end-to-end intelligent processing mechanism achieves a fully intelligent upgrade of factoring consulting services through the deep integration of three core technology modules: multimodal information processing, dynamic user profile updates, and professional knowledge graph retrieval. In terms of service efficiency, it can shorten the processing cycle of traditional manual consultations from several hours to a response level of seconds. Through a parallel computing architecture, it can simultaneously handle thousands of concurrent consultation requests, significantly improving business processing capabilities. Regarding service quality, fine-grained knowledge graphs can be used to ensure that responses comply with the latest business regulations. Real-time profile analysis can also be used to provide personalized recommendations, enhancing the user experience.

[0117] Furthermore, the multimodal fusion technology employed transcends the single interaction model of traditional consulting services, supporting multiple information input methods such as text, voice, and images, thereby expanding coverage of business consulting scenarios. The dynamic profiling engine updates customer profile data in a short cycle, ensuring that the system always provides recommendations based on the latest needs. This intelligent processing mechanism balances business compliance with user experience. A rigorous knowledge graph verification mechanism ensures that each recommendation complies with regulatory requirements. Furthermore, the natural language generation capabilities of a large language model are leveraged to translate specialized terminology into expressions that are easily understood by customers.

[0118] Figure 2 A flow chart of an intelligent reply method for online consultation on factoring business according to another embodiment of the present application is shown.

[0119] like Figure 2 As shown, according to another embodiment of the present application, an intelligent reply method for online consultation factoring business includes:

[0120] Step 202: Obtain real-time consulting information from the client regarding factoring business.

[0121] Step 204: Based on the real-time consultation information and a preset method for obtaining valid input information, determine the valid input information corresponding to the real-time consultation information.

[0122] Step 206: Update the user profile of the client based on the valid input information.

[0123] Step 208: Search the preset factoring business knowledge graph for factoring business related information that is associated with the valid input information and the updated user profile.

[0124] Step 210: Determine the multivariate structured data consisting of the valid input information, the updated user profile, and the factoring business related information as model input information for the large language model.

[0125] Step 212: Perform sentiment analysis on the valid input information to determine the real-time user sentiment information of the client, and add the real-time user sentiment information to the model input information.

[0126] Sentiment analysis can accurately identify a customer's emotional state during a consultation, such as anxiety, eagerness, or satisfaction, and incorporate this emotional label as a key dimension into the model input. For example, if a customer expresses "high anxiety" regarding a consultation regarding "factoring interest rates," the system automatically assigns a sentiment weight of 0.8 (on a scale of 0-1). This data directly influences the response strategy generated.

[0127] Step 214: outputting, through the large language model, real-time feedback information corresponding to the model input information and for the real-time consultation information.

[0128] During the response generation phase, the large language model dynamically adjusts the output based on sentiment tags. The adjustments include:

[0129] Optimize the tone of voice: Use more reassuring language for anxious customers, such as "Please rest assured, the risk factor for buyers of tertiary hospitals is only 0.02";

[0130] Re-prioritize information: In an emergency, prioritize core solutions rather than complete processes;

[0131] Risk warning reinforcement: When sentiment analysis shows that customers are overly optimistic, the risk clause reminders are automatically strengthened.

[0132] Therefore, the client user's emotion is added as an additional element of the model input to its input data tuple, which facilitates the real-time optimization of the answer content based on the user's emotional information. It not only increases the empathy performance of the intelligent customer service system, provides emotion-aware value-added services for high-net-worth customers, and improves the user experience, but also effectively prevents customer churn caused by ignoring emotional judgment, and discovers potential service improvement points through emotional big data analysis.

[0133] In another possible design, after determining the client's real-time user emotion information, an emotion tag can be added to the updated user profile based on the real-time user emotion information, thereby adding the real-time user emotion information as a new dimension of the user profile. In this way, in the step of determining the multivariate structured data consisting of the valid input information, the updated user profile, and the factoring business-related information as the model input information for the large language model, the user profile used is the secondarily updated user profile with the added emotion tag.

[0134] In other words, user emotion information can be used as part of the user profile and updated to the user profile twice. Therefore, rather than directly adding user emotion information as a dimension of model input, it can be added as branch information under the user profile dimension in multi-dimensional structured data. This, to a certain extent, avoids over-complication of model input information, helps the large language model more accurately and conveniently understand the multi-dimensional structured data input, and improves the response speed of the large language model, enabling faster responses to client users and improving the timeliness of automatic query responses.

[0135] exist Figure 1 and Figure 2 On the basis of the illustrated embodiment, the inquiry type of the real-time consultation information may be determined based on the valid input information, and the real-time feedback information may be displayed in an information display manner corresponding to the inquiry type.

[0136] The inquiry types include poster request types and text request types.

[0137] If the inquiry type is the poster requirement type, based on the valid input information, determine a poster template in the poster template set that is adapted to the current factoring business content consulted by the real-time consultation information; fill the real-time feedback information into the poster template, generate a real-time feedback poster, and display the real-time feedback poster on the client.

[0138] For example, if the client user is a corporate user, when he inquires about n business within the factoring business, if n business has activities related to the client's real-time consulting information, the system can determine that his inquiry type is the poster demand type, and then produce a poster with the activity content and feedback it to the client.

[0139] For another example, if the client user is a business handler for factoring business on the system side, he or she may ask "make a poster for the promotion of business n". In this way, the system can determine that the inquiry type is the poster demand type, and directly produce a poster with the relevant marketing content of business n and feedback it to the client.

[0140] In addition, if the inquiry type is the text demand type, it means that the client has no demand for obtaining the poster, and the real-time feedback information is displayed on the client.

[0141] In this way, real-time feedback information can be displayed to client users in an answer display method that meets the actual needs of the client, thereby improving the user experience.

[0142] In addition, in a possible design, after real-time feedback information is generated, the client's score for the real-time feedback information can be obtained as the accuracy of the real-time feedback information; if the accuracy is lower than a predetermined accuracy threshold, the current intelligent reply mode is switched to the manual reply mode, and a manual customer service representative whose capability attribute information matches the real-time consultation information is assigned to the client.

[0143] In another possible design, after generating real-time feedback information, the accuracy of the real-time feedback information can be determined based on preset feedback information accuracy evaluation rules and the real-time feedback information; if the accuracy is lower than a predetermined accuracy threshold, the current intelligent reply mode is switched to the manual reply mode, and a manual customer service representative whose capability attribute information matches the real-time consultation information is assigned to the client.

[0144] This allows users to determine whether the current smart response is sufficiently accurate and meets their actual needs based on real-time user evaluation or automatic system judgment, and to promptly redirect to human customer service when the accuracy falls below a predetermined accuracy threshold. The predetermined accuracy threshold is the minimum accuracy required for real-time feedback to clearly and accurately respond to users' real-time inquiries.

[0145] The embodiment of the present application provides an intelligent reply device for online consultation on factoring business, including:

[0146] A real-time consulting information acquisition unit, used to acquire real-time consulting information from the client regarding factoring business;

[0147] an effective input information determining unit, configured to determine effective input information corresponding to the real-time consultation information based on the real-time consultation information and a preset effective input information obtaining method;

[0148] A user portrait updating unit, configured to update the user portrait of the client based on the valid input information;

[0149] A knowledge graph search unit, configured to search a preset factoring business knowledge graph for factoring business-related information associated with the valid input information and the updated user profile;

[0150] The large language model processing unit is used to determine that the multivariate structured data composed of the valid input information, the updated user portrait and the factoring business related information is the model input information of the large language model, and output the real-time feedback information corresponding to the model input information for the real-time consultation information through the large language model.

[0151] In one embodiment of the present application, optionally, the valid input information determining unit includes:

[0152] an intention recognition unit, configured to, if the real-time consultation information includes text information, determine a semantic text of the text information based on an intention recognition model, and add the semantic text to the valid input information;

[0153] An image parsing unit, configured to parse the content text of the image information based on a CLIP model if the real-time consultation information includes image information, and add the content text to the valid input information;

[0154] An audio transcription unit is used to transcribe the audio information into audio text based on a Whisper model if the real-time consultation information includes audio information, and to add the audio text to the valid input information.

[0155] In one embodiment of the present application, optionally, the device further includes:

[0156] an entity and inter-entity relationship extraction unit, configured to extract factoring business entities and inter-entity relationships from a factoring business information database based on a BERT+BiLSTM-CRF hybrid neural network model before the real-time consultation information acquisition unit acquires the real-time consultation information, wherein the factoring business entities include: factoring type, business recommendation language, risk control indicators, and legal terms;

[0157] The knowledge graph construction unit is used to construct the factoring business knowledge graph with the factoring business entities as nodes, the relationships between the entities as edges between the nodes, and the correlation between the nodes reflected by the relationships between the entities as the weights of the edges.

[0158] In one embodiment of the present application, optionally, the device further includes:

[0159] An emotion recognition unit, configured to perform emotion analysis on the valid input information to determine real-time user emotion information of the client;

[0160] The model input information updating unit is used to add the real-time user emotion information to the model input information before the large language model processing unit outputs the real-time feedback information.

[0161] In one embodiment of the present application, optionally, the device further includes:

[0162] An emotion recognition unit, configured to perform emotion analysis on the valid input information to determine real-time user emotion information of the client;

[0163] The user portrait re-updating unit is used to add an emotion tag to the updated user portrait based on the real-time user emotion information after the user portrait updating unit updates the user portrait of the client and before the knowledge graph search unit searches for factoring business related information.

[0164] In one embodiment of the present application, optionally, the device further includes:

[0165] an inquiry type determining unit, configured to determine an inquiry type of the real-time consultation information based on the valid input information, wherein the inquiry type includes a poster requirement type and a text requirement type;

[0166] An information display unit is configured to display the real-time feedback information in an information display manner corresponding to the inquiry type, wherein the information display unit includes:

[0167] a poster template extraction unit configured to, if the inquiry type is the poster requirement type, determine, based on the valid input information, a poster template from a poster template set that is adapted to the current factoring business content consulted by the real-time consultation information;

[0168] a poster generating unit, configured to fill the real-time feedback information into the poster template, generate a real-time feedback poster, and display the real-time feedback poster on the client;

[0169] A text display unit is used to display the real-time feedback information on the client if the inquiry type is the text requirement type.

[0170] In one embodiment of the present application, optionally, the device further includes:

[0171] a client evaluation unit, configured to obtain a score of the client on the real-time feedback information as the accuracy of the real-time feedback information;

[0172] The reply mode switching unit is used to switch from the current intelligent reply mode to the manual reply mode if the accuracy is lower than a predetermined accuracy threshold, and to assign a manual customer service representative whose capability attribute information matches the real-time consultation information to the client.

[0173] In one embodiment of the present application, optionally, the device further includes:

[0174] an automatic evaluation unit, configured to determine the accuracy of the real-time feedback information based on a preset feedback information accuracy evaluation rule and the real-time feedback information;

[0175] The reply mode switching unit is used to switch from the current intelligent reply mode to the manual reply mode if the accuracy is lower than a predetermined accuracy threshold, and to assign a manual customer service representative whose capability attribute information matches the real-time consultation information to the client.

[0176] The device uses any one of the solutions in the above embodiments, and therefore has all the above technical effects, which will not be described in detail here.

[0177] In addition, in one embodiment, the present application provides a computer device, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown. The computer device includes a processor, memory, network interface and database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it can implement the method described in any of the above embodiments.

[0178] In one embodiment, the present application further provides a computer device, which may be a client, and its internal structure diagram may be as follows: Figure 4As shown. The computer device includes a processor, memory, a network interface, a display screen, and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When executed by the processor, the computer program can implement the method described in any of the above embodiments.

[0179] Any of the aforementioned computer devices in the embodiments of the present application may exist in various forms, including but not limited to:

[0180] (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and are primarily designed to provide voice and data communications. These terminals include smartphones (e.g., iPhones), multimedia phones, feature phones, and low-end phones.

[0181] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers, have computing and processing capabilities, and generally also have mobile Internet access. These terminals include PDAs, MIDs, and UMPCs, such as the iPad.

[0182] (3) Portable entertainment devices: These devices can display and play multimedia content. These devices include audio and video players (such as iPods), handheld game consoles, e-books, as well as smart toys, wearable devices, and portable car navigation devices.

[0183] (4) Server: A device that provides computing services. The server consists of a processor, hard disk, memory, system bus, etc. The server is similar to a general computer architecture, but because it needs to provide highly reliable services, it has higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0184] (5) Other electronic devices with data interaction functions.

[0185] In addition, an embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to perform the following steps:

[0186] Obtain real-time consulting information from clients regarding factoring business;

[0187] Determining valid input information corresponding to the real-time consultation information based on the real-time consultation information and a preset valid input information acquisition method;

[0188] Based on the valid input information, updating the user profile of the client;

[0189] Searching for factoring business-related information associated with the valid input information and the updated user profile in a preset factoring business knowledge graph;

[0190] Determine that the multivariate structured data consisting of the valid input information, the updated user portrait and the factoring business related information is the model input information of the large language model, and output the real-time feedback information corresponding to the model input information for the real-time consultation information through the large language model.

[0191] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant description in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0192] The above, in combination with the accompanying drawings, describes in detail the technical solution of this application. Through the technical solution of this application, the full-process intelligent response processing of factoring business consulting services is realized. The end-to-end intelligent processing mechanism significantly improves service efficiency and quality, ensuring that the reply content is in line with business specifications and close to the customer's actual situation, providing customers with accurate, efficient and professional consulting service experience.

[0193] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0194] The terms used in the embodiments of the present application are for the purpose of describing specific embodiments only and are not intended to limit the present application. The singular forms "a", "an", "the" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0195] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some interface, indirect coupling or communication connection of the device or unit, which may be electrical, mechanical or other forms.

[0196] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0197] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0198] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. An intelligent reply method for online consultation on factoring business, characterized in that: include: Obtain real-time consulting information from clients regarding factoring business; Determining valid input information corresponding to the real-time consultation information based on the real-time consultation information and a preset valid input information acquisition method; Based on the valid input information, updating the user profile of the client; Searching for factoring business-related information associated with the valid input information and the updated user profile in a preset factoring business knowledge graph; Determine that the multivariate structured data consisting of the valid input information, the updated user portrait and the factoring business related information is the model input information of the large language model, and output the real-time feedback information corresponding to the model input information for the real-time consultation information through the large language model.

2. The method according to claim 1, characterized in that The determining, based on the real-time consultation information and a preset valid input information acquisition method, the valid input information corresponding to the real-time consultation information includes: If the real-time consultation information includes text information, determining a semantic text of the text information based on an intention recognition model, and adding the semantic text to the valid input information; If the real-time consultation information includes picture information, parsing the content text of the picture information based on the CLIP model, and adding the content text to the valid input information; If the real-time consultation information includes audio information, the audio information is transcribed into audio text based on the Whisper model, and the audio text is added to the valid input information.

3. The method according to claim 2, characterized in that Before obtaining the real-time consulting information of the client regarding the factoring business, the method further includes: Extract factoring business entities and inter-entity relationships from a factoring business information database using a BERT+BiLSTM-CRF hybrid neural network model. The factoring business entities include: factoring type, business recommendation language, risk control indicators, and legal terms. The factoring business knowledge graph is constructed with the factoring business entities as nodes, the relationships between the entities as edges between the nodes, and the inter-node correlation reflected by the relationships between the entities as weights of the edges.

4. The method according to claim 3, characterized in that Also includes: Performing sentiment analysis on the valid input information to determine real-time user sentiment information of the client; Before the large language model outputs the real-time feedback information corresponding to the model input information and for the real-time consultation information, the method further includes: The real-time user emotion information is added to the model input information.

5. The method according to claim 3, characterized in that Also includes: Performing sentiment analysis on the valid input information to determine real-time user sentiment information of the client; After updating the user profile of the client based on the valid input information and before searching for factoring business-related information associated with the valid input information and the updated user profile, the method further includes: Based on the real-time user emotion information, an emotion tag is added to the updated user portrait.

6. The method according to any one of claims 1 to 5, characterized in that Also includes: Determining the inquiry type of the real-time consultation information based on the valid input information, wherein the inquiry type includes a poster requirement type and a text requirement type; The real-time feedback information is displayed in an information display manner corresponding to the inquiry type, wherein: If the inquiry type is the poster requirement type, based on the valid input information, determining a poster template from a poster template set that is adapted to the current factoring business content inquired about by the real-time consultation information; Filling the real-time feedback information into the poster template to generate a real-time feedback poster, and displaying the real-time feedback poster on the client; If the inquiry type is the text requirement type, the real-time feedback information is displayed on the client.

7. The method according to claim 6, characterized in that Also includes: Obtaining a score of the real-time feedback information by the client as the accuracy of the real-time feedback information; If the accuracy is lower than a predetermined accuracy threshold, the current intelligent reply mode is switched to the manual reply mode, and a manual customer service representative whose capability attribute information matches the real-time consultation information is assigned to the client.

8. The method according to claim 6, characterized in that Also includes: Determining the accuracy of the real-time feedback information based on a preset feedback information accuracy evaluation rule and the real-time feedback information; If the accuracy is lower than a predetermined accuracy threshold, the current intelligent reply mode is switched to the manual reply mode, and a manual customer service representative whose capability attribute information matches the real-time consultation information is assigned to the client.

9. A computer device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are configured to enable the processor to execute the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that Computer-executable instructions are stored, and the computer-executable instructions are configured to execute the method according to any one of claims 1 to 8.

Images