Intelligent Question Answering System and Methods

Abstract

A smart question-answering system includes an application interface server, a communication server, and an artificial intelligence server electrically connected to the application interface server. The artificial intelligence server inputs a question text into a text vectorization model to obtain a digital text. The application interface server searches a clause text knowledge base based on a vectorized keyword contained in the digital text to obtain a similar clause text. The artificial intelligence server inputs a question-answer text containing the question text, the similar clause text, and a prompt word corresponding to the similar clause text into a first model or a second model indicated by the similar clause text to generate an answer message, and transmits the answer message to a question-answering interface of the communication server via the application interface server.

Description

Technical Field

[0001] This invention relates to an intelligent question-and-answer system, specifically an intelligent question-and-answer system that can generate answers that fit the question without pre-setting knowledge point correspondences. Prior Technology

[0002] Currently, insurance companies provide their sales staff with a question-and-answer assistant (e.g., a chatbot) installed on their smartphones or tablets to help them respond to customer inquiries instantly. However, these assistants typically provide one-to-one knowledge points. If a customer's question doesn't have a corresponding pre-set knowledge point, the system either switches to a live customer service representative or provides irrelevant answers, severely impacting the sales staff's professional image in front of customers. Therefore, providing a system that can generate relevant answers without requiring pre-set knowledge point relationships has become a worthy research topic. Summary of the Invention

[0003] Therefore, the purpose of this invention is to provide an intelligent question-and-answer system that can generate answers that fit the question without the need for pre-setting knowledge point correspondences.

[0004] Therefore, the present invention provides an intelligent question-and-answer system applicable to a terminal device, and includes a communication server, an application interface server and an artificial intelligence server.

[0005] The communication server establishes a connection with the terminal device and provides a question-and-answer interface for the terminal device to input a question text. The application interface server is electrically connected to the communication server and stores a clause text knowledge base. The clause text knowledge base contains multiple clause texts and multiple preset prompts corresponding to each clause text. The artificial intelligence server is electrically connected to the application interface server and stores a text vectorization model, a first model, and a second model. Either the first model or the second model is a generative model. Each clause text indicates one of the first model or the second model.

[0006] The application interface server transmits the question text from the communication server to the artificial intelligence server. The artificial intelligence server inputs the question text into the text vectorization model to obtain a digital text related to the question text, and then sends the digital text back to the application interface server.

[0007] The application interface server searches the term text knowledge base based on a vectorized keyword contained in the digital text to obtain an approximate term text.

[0008] The application interface server generates a question-and-answer text containing the question text, the approximate clause text, and the prompt word corresponding to the approximate clause text, and sends it to the artificial intelligence server.

[0009] The AI ​​server inputs the question-and-answer text into the first or second model indicated by the similar terms text to generate an answer message related to the question-and-answer text, and transmits the answer message to the question-and-answer interface of the communication server via the application interface server.

[0010] In some implementations, if the communication server determines that the content of the question text contains sensitive personal information, it will not send the question text to the application interface server, and will generate and display a prompt message related to re-entering the question on the question-and-answer interface, until the communication server determines that the content of the question text entered by the terminal device does not contain sensitive personal information, at which point the question text will be sent to the application interface server.

[0011] In some implementations, the definition of the approximate clause text is the clause text among those clause texts that has the shortest Euclidean distance to the vectorized keyword.

[0012] In some implementations, either the first model or the second model is a large language model.

[0013] In some implementations, each of these clauses relates to one of three types of coverage: medical coverage, accident coverage, and general coverage.

[0014] Another objective of this invention is to provide a smart question-and-answer method that can generate answers that fit the question without the need for pre-setting knowledge point correspondences.

[0015] Therefore, the present invention provides a smart question-answering method applicable to a terminal device, comprising the following steps: a communication server provides a question-answering interface for the terminal device to input a question text; an application interface server transmits the question text from the communication server to an artificial intelligence server; the artificial intelligence server inputs the question text into a text vectorization model stored therein to obtain a digital text related to the question text, and sends the digital text back to the application interface server; the application interface server searches a clause text knowledge base stored therein based on a vectorized keyword contained in the digital text to obtain an approximate clause text; wherein, the clause text knowledge base contains multiple... The application interface server generates a question-and-answer text containing the question text, the similar clause text, and the prompts corresponding to the similar clause text, and transmits it to the artificial intelligence server; the artificial intelligence server inputs the question-and-answer text into a first model or a second model stored therein, indicated by the similar clause text, to generate an answer message related to the question-and-answer text, and transmits the answer message to the question-and-answer interface of the communication server via the application interface server, wherein either the first model or the second model is a generative model, and each clause text indicates one of the first model or the second model.

[0016] In some implementations, if the communication server determines that the content of the question text contains sensitive personal information, it will not send the question text to the application interface server, and will generate and display a prompt message related to re-entering the question on the question-and-answer interface, until the communication server determines that the content of the question text entered by the terminal device does not contain sensitive personal information, at which point the question text will be sent to the application interface server.

[0017] In some implementations, the definition of the approximate clause text is the clause text among those clause texts that has the shortest Euclidean distance to the vectorized keyword.

[0018] In some implementations, either the first model or the second model is a large language model.

[0019] In some implementations, each of these clauses relates to one of three types of coverage: medical coverage, accident coverage, and general coverage.

[0020] The advantages of this invention are as follows: the AI ​​server inputs the question text into the text vectorization model to obtain the digital text; the application interface server searches the clause text knowledge base based on the vectorized keywords contained in the digital text to obtain the approximate clause text; the AI ​​server inputs the question-and-answer text containing the question text, the approximate clause text, and the prompt word corresponding to the approximate clause text into the first or second model indicated by the prompt word to generate the answer message related to the question-and-answer text, and transmits the answer message to the question-and-answer interface of the communication server via the application interface server, so that the answer message that fits the question text can be generated even without a preset knowledge point correspondence. Simple Explanation of the Diagram

[0021] Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, wherein: Figure 1 is a block diagram illustrating an embodiment of the intelligent question-answering system of the present invention; and Figure 2 is a flowchart illustrating an embodiment of how the intelligent question-answering system of Figure 1 executes the intelligent question-answering method of the present invention. Implementation

[0022] Before the present invention is described in detail, it should be noted that similar elements are represented by the same reference numerals in the following description. Unless otherwise defined, "electrical connection" as used in this patent specification refers to a "wired electrical connection" in which multiple electronic devices / devices / components are interconnected through conductive materials, and a "radio connection" in which one-way / two-way wireless signal transmission is performed through wireless communication technology. On the other hand, "electrical connection" as used in this patent specification also refers to a "direct electrical connection" formed by multiple electronic devices / devices / components being directly interconnected, and an "indirect electrical connection" formed by multiple electronic devices / devices / components being indirectly interconnected through other electronic devices / devices / components.

[0023] Referring to Figure 1, an embodiment of the intelligent question-and-answer system of the present invention is applicable to a terminal device 1 and includes a communication server 2, an application interface server 3, and an artificial intelligence server 4. In this embodiment, the terminal device 1 is, for example, but not limited to, a computer device such as a smartphone or tablet computer capable of executing an insurance application, and is operated and used by a business person of an insurance institution.

[0024] The communication server 2 provides an environment for the insurance application to run, enabling the terminal device 1 to establish a connection with the communication server 2 via the Internet or a local area network by executing the insurance application. In this embodiment, the communication server 2 is, for example, one or more computer devices set up by the insurance institution.

[0025] The application interface server 3 is electrically connected to the communication server 2 and stores a policy text knowledge base. This policy text knowledge base contains multiple policy texts and multiple preset prompts corresponding to each policy text. More specifically, each policy text relates to one of a medical insurance policy, an accident insurance policy, and a general insurance policy; each prompt is either a basic prompt or a special prompt. In detail, the policy text knowledge base is a vector database, and each policy text may contain, for example, a summary version, a full version, and regulations (e.g., derived from the full version through paragraph segmentation) of the insurance product policy, as well as national health insurance medical service reimbursement items and payment standards, a common-sounding table of disease names, etc., but is not limited thereto. In this embodiment, the application interface server 3 is, for example, one or more computer devices set up by the insurance institution; the policy text knowledge base is, for example, but not limited to, Chroma DB.

[0026] The AI ​​server 4 is electrically connected to the application interface server 3 and stores a text vectorization model, a first model, and a second model. More specifically, either the first model or the second model is a generative model. More specifically, either the first model or the second model is a large language model. In this embodiment, the AI ​​server 4 is, for example, one or more computer devices; the text vectorization model is, for example, but not limited to, the Microsoft vectorization model; the first model is, for example, but not limited to, GPT-4o; and the second model is, for example, but not limited to, GPT-3.5 Turbo.

[0027] Each of these terms refers to either the first model or the second model. More specifically, each of these terms, when relating to the medical coverage or the accident coverage, refers to the first model; and each of these terms, when relating to the general coverage, refers to the second model.

[0028] Hereinafter, referring to Figures 1 and 2, an embodiment of the intelligent question-answering method of the present invention will be described in detail. The intelligent question-answering method includes the following steps S21 to S28.

[0029] First, the salesperson establishes a connection with the communication server 2 via the Internet or local area network by operating the terminal device 1 to execute the insurance application therein.

[0030] Next, in step S21, the communication server 2 provides a question-and-answer interface for the terminal device 1 to input a question text. For example, the first question text might be "Main payment items of Meixin Chuanfu Foreign Currency Interest Rate Variable Whole Life Insurance (Term Payment Type)"; the second question text might be "Is Meixin Chuanfu Foreign Currency Interest Rate Variable Whole Life Insurance (Term Payment Type) principal-protected?"; and the third question text might be "How long does it take to withdraw funds from Meixin Chuanfu Foreign Currency Interest Rate Variable Whole Life Insurance (Term Payment Type)?". It should be noted that the communication server 2 displays a warning message on the question-and-answer interface, such as "Do not enter any sensitive personal information," to remind the salesperson not to enter any sensitive personal information to prevent the leakage of customers' sensitive personal information.

[0031] Then, in step S22, the communication server 2 determines whether the content of the question text contains sensitive personal information. This sensitive personal information includes, but is not limited to, a customer's (natural person's) name, date of birth, national identity card number, passport number, characteristics, fingerprints, marital status, family background, education, occupation, medical records, medical history, genetic information, sex life, health checkups, criminal record, contact information, financial situation, social activities, and other information that can directly or indirectly identify the individual. If the determination result is negative (that is, if the communication server 2 determines that the content of the question text does not contain the sensitive personal information), the process proceeds to step S23, whereby the communication server 2 transmits the question text to the application interface server 3. Conversely, if the judgment result is positive (that is, when the communication server 2 determines that the content of the question text contains the sensitive personal information), the communication server 2 generates and displays a prompt message related to re-entering the question on the question-and-answer interface, and does not send the question text to the application interface server 3. The process will return to step S22 so that the communication server 2 can repeat step S22 until the communication server 2 determines that the content of the question text entered by the terminal device 1 does not contain the sensitive personal information, and then sends the question text to the application interface server 3 (step S23).

[0032] Next, in step S24, the application interface server 3 transmits the problem text from the communication server 2 to the artificial intelligence server 4.

[0033] Then, in step S25, the AI ​​server 4 inputs the question text into the text vectorization model stored therein to obtain a digital text related to the question text, and sends the digital text back to the application interface server 3.

[0034] Next, in step S26, the application interface server 3 searches the clause text knowledge base based on a vectorized keyword contained in the digital text to obtain a similar clause text. Specifically, the similar clause text is defined as the clause text among those clause texts that has the shortest Euclidean distance to the vectorized keyword. Continuing from the previous example, the first question text corresponds to a similar clause text, for example, "Meixin Chuanfu Foreign Currency Interest Rate Variable Whole Life Insurance (Term Payment Type)"; the corresponding prompt word is, for example, the basic prompt word "Answer provided according to the summary version or full version of the insurance product terms"; the second question text corresponds to a similar clause text, for example, "Meixin Chuanfu Foreign Currency Interest Rate Variable Whole Life Insurance (Term Payment Type)"; the corresponding prompt word is, for example, the special prompt word "If the question text mentions 'principal protection,' do not reply to the question text"; the third question text corresponds to a similar clause text, for example, "Meixin Chuanfu Foreign Currency Interest Rate Variable Whole Life Insurance (Term Payment Type)"; the corresponding prompt word is, for example, the special prompt word "If the question text mentions 'how long it takes to receive the money,' reply 'Hello, you can refer to the DM / calendar / proposal'"

[0035] Then, in step S27, the application interface server 3 generates a question-and-answer text containing the question text, the approximate clause text, and the prompt word corresponding to the approximate clause text, and sends it to the artificial intelligence server 4.

[0036] Finally, in step S28, the AI ​​server 4 inputs the question-and-answer text into the first or second model indicated by the approximate terms text to generate an answer message related to the question-and-answer text, and transmits the answer message to the question-and-answer interface of the communication server 2 via the application interface server 3. Continuing the previous example, the answer message corresponding to the first question text is, for example, "Main payment items ● Longevity insurance benefit ● Death benefit or funeral expense insurance benefit ● Total disability insurance benefit"; the answer message corresponding to the second question text is, for example, "Hello, according to the current policy terms and summary document, we cannot answer the question about 'guaranteed principal'"; the answer message corresponding to the third question text is, for example, "Hello, you can refer to the DM / calculation sheet / proposal".

[0037] In summary, the AI ​​server 4 inputs the question text into the text vectorization model to obtain the digital text; the application interface server 3 searches the clause text knowledge base based on the vectorized keywords contained in the digital text to obtain the approximate clause text; the AI ​​server 4 inputs the question-and-answer text containing the question text, the approximate clause text, and the prompt word corresponding to the approximate clause text into the first or second model indicated by the approximate clause text to generate the answer message related to the question-and-answer text, and transmits the answer message to the question-and-answer interface of the communication server 2 via the application interface server 3, so that the answer message that fits the question text can be generated even without a preset knowledge point correspondence. In this way, the salesperson can obtain the knowledge reserve to reply to the customer. Therefore, the purpose and effect of the present invention are indeed achieved.

[0038] However, the above description is merely an embodiment of the present invention and should not be construed as limiting the scope of the present invention. Any simple equivalent changes and modifications made in accordance with the scope of the patent application and the contents of the patent specification of the present invention shall still fall within the scope of the patent of the present invention.

[0039] 1: Terminal device 2: Communication Server 3: Application Interface Server 4: Artificial Intelligence Server S21~S28: Steps

Claims

1. A smart question-and-answer system, applicable to a terminal device, and comprising: a communication server, capable of establishing a connection with the terminal device and providing a question-and-answer interface for the terminal device to input a question text; An application interface server, electrically connected to a communication server, stores a clause text knowledge base containing multiple clause texts and multiple preset prompts corresponding to each clause text; and an artificial intelligence server, electrically connected to the application interface server, stores a text vectorization model, a first model, and a second model, either the first model or the second model being a generative model, each clause text indicating one of the first model or the second model, and each clause text relating to one of a medical insurance, an accident insurance, and a general insurance; wherein the application interface server transmits the question text from the communication server to the artificial intelligence server, the artificial intelligence server inputs the question text into the text vectorization model to obtain a digital text related to the question text, and sends the digital text back to the application interface server; the application interface server searches the clause text knowledge base based on a vectorized keyword contained in the digital text to obtain an approximate clause text; The application interface server generates a question-and-answer text containing the question text, the similar terms text, and the prompt word corresponding to the similar terms text, and sends it to the artificial intelligence server; the artificial intelligence server inputs the question-and-answer text into the first model indicated by the medical insurance or the accident insurance related to the similar terms text, or the second model indicated by the general insurance related to the similar terms text, to generate an answer message related to the question-and-answer text, and sends the answer message to the question-and-answer interface of the communication server via the application interface server.

2. The intelligent question-answering system as described in claim 1, wherein, When the communication server determines that the content of the question text contains sensitive personal information, it will not send the question text to the application interface server, and will generate and display a prompt message on the question and answer interface to re-enter the question. Only when the communication server determines that the content of the question text entered by the terminal device does not contain sensitive personal information will the question text be sent to the application interface server.

3. The intelligent question-answering system as described in claim 1, wherein, The definition of the approximate clause text is the clause text among those clause texts that has the shortest Euclidean distance to the vectorized keyword.

4. The intelligent question-answering system as described in claim 1, wherein, Either the first model or the second model is a large language model.

5. A smart question-answering method, applicable to a terminal device, comprising the following steps: A communication server provides a question-answering interface for the terminal device to input a question text; An application interface server transmits the question text from the communication server to an artificial intelligence server; The artificial intelligence server inputs the question text into a text vectorization model stored therein to obtain a digital text related to the question text, and sends the digital text back to the application interface server; The application interface server searches a stored clause text knowledge base based on a vectorized keyword contained in the digital text to obtain an approximate clause text; wherein, The term knowledge base contains multiple term texts and multiple preset prompts corresponding to each term text. Each term text is related to one of a medical insurance, an accident insurance, and a general insurance. The application interface server generates a question-and-answer text containing the question text, the similar term text, and the prompts corresponding to the similar term text, and sends it to the artificial intelligence server. The artificial intelligence server inputs the question-and-answer text into the storage, whereby the similar term text is related to a first model indicated by the medical insurance or the accident insurance, or the similar term text is related to a second model indicated by the general insurance, to generate an answer message related to the question-and-answer text. The answer message is then sent to the question-and-answer interface of the communication server via the application interface server. Either the first model or the second model is a generative model, and each term text indicates one of the first model or the second model.

6. The intelligent question-answering method as described in claim 5, wherein, When the communication server determines that the content of the question text contains sensitive personal information, it will not send the question text to the application interface server, and will generate and display a prompt message on the question and answer interface to re-enter the question. Only when the communication server determines that the content of the question text entered by the terminal device does not contain sensitive personal information will the question text be sent to the application interface server.

7. The intelligent question-answering method as described in claim 5, wherein, The definition of the approximate clause text is the clause text among those clause texts that has the shortest Euclidean distance to the vectorized keyword.

8. The intelligent question-answering method as described in claim 5, wherein, Either the first model or the second model is a large language model.

Images