Artificial intelligence interface system for updating database by using dialog log, and method therefor

Abstract

An artificial intelligence interface system for updating a database by using a conversation log, and a method therefor are disclosed. According to the disclosed method, the artificial intelligence interface system can classify, by dialog topic, one or more question-answer sets included in a dialog log into one or more dialog records, generate a plurality of topic summaries corresponding to the dialog topics of the dialog record, determine, on the basis of the plurality of topic summaries, the topic summary to be stored in a database, and store the determined topic summary in the database.

Description

Artificial intelligence interface system and method for updating a database using conversation logs

[0001] The present disclosure relates to a technology for accumulating knowledge based on conversation logs and improving the quality of a natural language question-and-answer service using an artificial intelligence model.

[0002] Large Language Models (LLMs) can understand a user's natural language input, determine their intent, and output appropriate natural language responses. This can significantly simplify human-system interactions and enable the design of more efficient communication structures.

[0003] If we define the structure in which users and systems interact using LLM as an artificial intelligence (AI) interface system, the value of AI interface systems can be maximized when they provide personalized services through continuous interaction with users. In particular, AI interface systems can be a key solution in applications such as stock strategy analysis or system design, where the results provided to users must be progressively improved through interaction and continuous feedback.

[0004] However, properly training AI interface systems, including LLMs, and tuning LLM parameters is costly and time-consuming. Therefore, training AI interface systems tailored to individual users is practically challenging, and the likelihood of various problems, such as overfitting, increases as user interactions accumulate. Furthermore, LLM hallucinations can cause serious errors when handling important facts and information, posing problems with the reliability and accuracy of AI interface systems. Previously, fine-tuning only a portion of LLM parameters was used, but this method still required significant time and cost during training and was limited by issues such as hallucinations and overfitting.

[0005] The present disclosure aims to realize a quality improvement effect due to the gradual accumulation of knowledge in a natural language question-and-answer service without additional training of a large-scale language model by updating a database based on the results of interactions between a user and an artificial intelligence interface system.

[0006] The technical problems to be achieved in the present disclosure are not limited to the technical problems described above, and other technical problems can be inferred from the following examples.

[0007] In order to solve the above problem, a method for updating a database using a dialogue log in an artificial intelligence interface system may include a step of classifying one or more question-answer sets included in a dialogue log into one or more dialogue records according to a dialogue topic; a step of generating a plurality of topic summaries corresponding to the dialogue topics of the dialogue records; a step of determining a topic summary to be stored in the database based on the plurality of topic summaries; and a step of storing the determined topic summary in the database.

[0008] In one embodiment, the conversation log includes a question-answer set accumulated from the time of classification into a recent conversation log to the present, and the question-answer set may include a user's question input into the artificial intelligence interface system and a response from the artificial intelligence interface system to the question.

[0009] In one embodiment, the classifying step may include: requesting a language model constituting the artificial intelligence interface system to identify a conversation topic of each of the question-answer sets included in the conversation log at each time a preset condition is satisfied; and classifying each of the question-answer sets into one or more conversation logs based on the conversation topics identified by the language model.

[0010] In one embodiment, the method of updating a database using the conversation log may further include a step of deleting, among the question-answer sets included in the conversation log, a question-answer set whose frequency of use is below a certain level or whose period of time has passed since its inclusion in the conversation log.

[0011] In one embodiment, the generating step may include: searching the database for a topic summary corresponding to the conversation topic of the dialogue transcript; and generating a plurality of topic summaries corresponding to the conversation topic of the dialogue transcript based on a search result of the database. In this case, the plurality of topic summaries may have different word sizes. As an example related to this, the generating the plurality of topic summaries based on the search result may include, if a topic summary corresponding to the conversation topic of the dialogue transcript is searched in the database, inputting the dialogue transcript and the searched topic summary into a language model to request generation of the plurality of topic summaries. As another example, the generating the plurality of topic summaries based on the search result may include, if a topic summary corresponding to the conversation topic of the dialogue transcript is not searched in the database, inputting the dialogue transcript into a language model to request generation of the plurality of topic summaries.

[0012] In one embodiment, the plurality of subject summaries may include a first subject summary that satisfies a preset word size limit and a second subject summary that exceeds the preset word size limit.

[0013] In this case, the determining step may include: evaluating the similarity between the first subject summary and the second subject summary; determining the first subject summary as the subject summary to be stored in the database if the similarity is greater than or equal to a threshold; and determining the subject summary to be stored in the database if the similarity is less than the threshold based on the transcript and the second subject summary. As an example related to this, if the similarity is less than the threshold, the determining the subject summary to be stored in the database may include: inputting the transcript and the second subject summary into a language model to request generation of a third subject summary that satisfies the word size restriction; and determining the third subject summary as the subject summary to be stored in the database. Meanwhile, as another example, if the similarity is less than the reference value, the step of determining the subject summary to be stored in the database may include the step of inputting the transcript and the second subject summary into a language model to request generation of a third subject summary that satisfies the word size limit and a fourth subject summary that exceeds the word size limit; and the step of determining the subject summary to be stored in the database based on the third subject summary and the fourth subject summary.

[0014] Also, in this case, the plurality of subject summaries may further include a fifth subject summary that satisfies the word size restriction and has a word size greater than the word size of the first subject summary, and in this case, the determining step may include: evaluating a first similarity between the first subject summary and the second subject summary; when the first similarity is equal to or greater than a threshold, determining the first subject summary as a subject summary to be stored in the database; when the first similarity is less than the threshold, evaluating a second similarity between the second subject summary and the fifth subject summary; when the second similarity is equal to or greater than the threshold, determining the fifth subject summary as a subject summary to be stored in the database; and when the second similarity is less than the threshold, determining the subject summary to be stored in the database based on the transcript and the second subject summary.

[0015] In one embodiment, the storing step may include a step of updating a topic summary corresponding to a conversation topic of the determined topic summary among the topic summaries stored in the database with the determined topic summary, or a step of adding the determined topic summary as a new topic summary to the database.

[0016] In one embodiment, a method for updating a database using the conversation log may further include: receiving a user's query input into the artificial intelligence interface system; performing natural language processing on the query through a language model constituting the artificial intelligence interface system to identify a subject of the query; identifying a subject summary corresponding to the subject of the query in the database; inputting the identified subject summary into the language model to generate a response to the query; and providing the response to the user.

[0017] In this case, the step of identifying the subject summary may include a step of identifying the most recently added subject summary and the most frequently updated subject summary among the subject summaries corresponding to the subject of the query in the database.

[0018] In addition, in this case, the method for updating the database using the conversation log may further include a step of calling an API (Application Program Interface) for an external system connected to the artificial intelligence interface system when it is determined that interaction with the external system connected to the artificial intelligence interface system is necessary for processing the query as a result of natural language processing for the query; and a step of receiving a system result processed through the API from the external system, and the step of generating a response to the query may include a step of inputting the identified subject summary and the system result together into the language model to generate a response to the query.

[0019] Additionally, in this case, the method of updating a database using the conversation log may further include a step of recording a question-response set composed of the question and the response in the conversation log.

[0020] Meanwhile, in order to solve the above-described problem, an artificial intelligence interface system for updating a database using a dialogue log includes a memory and a processor for storing commands, and the processor is connected to the memory and can classify one or more question-answer sets included in a dialogue log into one or more dialogue records by dialogue topic, generate a plurality of topic summaries corresponding to the dialogue topics of the dialogue records, determine a topic summary to be stored in the database based on the plurality of topic summaries, and store the determined topic summaries in the database.

[0021] Specific details of other embodiments are included in the detailed description and drawings.

[0022] According to the present disclosure, by updating a database using conversation logs resulting from interactions between a user and an artificial intelligence interface system, an artificial intelligence interface system can be trained in a user-specific manner without directly training a language model.

[0023] In addition, according to the present disclosure, by generating a response to a user's query using a subject summary searched from an updated database and providing the response to the user, the user can be provided with a progressively customized response simply by exchanging questions and answers with the artificial intelligence interface system.

[0024] The effects of the invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description of the claims.

[0025] FIG. 1 is a schematic diagram of a system for providing a natural language question-and-answer service, including an artificial intelligence interface system according to one embodiment.

[0026] FIG. 2 is a flowchart illustrating a method for updating a database using a conversation log in an artificial intelligence interface system according to one embodiment.

[0027] Figure 3 is a conceptual diagram illustrating a method of classifying one or more question-answer sets included in a conversation log into one or more conversation logs by conversation topic.

[0028] Figure 4 is a conceptual diagram illustrating a method for generating multiple subject summaries.

[0029] Figure 5 is a conceptual diagram illustrating a method for determining a subject summary to be stored in a database.

[0030] FIG. 6 is a flowchart illustrating a method for providing a response to a user's query using an updated database according to one embodiment.

[0031] Figure 7 is a conceptual diagram exemplifying the method of Figure 6.

[0032] FIG. 8 is a block diagram illustrating an artificial intelligence interface system that updates a database using a conversation log according to one embodiment.

[0033] Hereinafter, specific embodiments will be described with reference to the drawings. The following detailed description is provided to facilitate a comprehensive understanding of the methods, devices, and / or systems described herein. However, these are merely examples, and the disclosed embodiments are not limited thereto.

[0034] In describing the embodiments, if it is determined that a detailed description of related known technology may unnecessarily obscure the gist of the disclosed embodiments, the detailed description will be omitted. In addition, the terms described below are terms defined in consideration of functions in the disclosed embodiments, and this may vary depending on the intention or custom of the user or operator. Therefore, the definitions should be made based on the contents throughout this specification. The terminology used in the detailed description is only for the purpose of describing the embodiments and should never be limited. Unless clearly used otherwise, the singular form includes the plural form. In this description, expressions such as "comprises" or "having" are intended to indicate certain features, numbers, steps, operations, elements, parts or combinations thereof, and should not be construed to exclude the presence or possibility of one or more other features, numbers, steps, operations, elements, parts or combinations thereof other than those described.

[0035] The terms used in the embodiments have been selected from widely used and common terms, taking into account the functions of the present disclosure. However, these terms may vary depending on the intentions of those skilled in the art, precedents, the emergence of new technologies, etc. Furthermore, in certain cases, terms may be arbitrarily selected by the applicant, in which case their meanings will be described in detail in the relevant description. Therefore, the terms used in this disclosure should not be defined simply as names of terms, but rather based on the meanings of the terms and the overall content of the present disclosure.

[0036] When a part of the specification is said to "include" a component, this does not exclude other components, but rather implies the inclusion of other components, unless otherwise specifically stated. Furthermore, terms such as "part," "module," etc., used in the specification refer to a unit that processes at least one function or operation, which may be implemented in hardware or software, or a combination of hardware and software, and, unlike the illustrated examples, may not be clearly distinguished in their specific operations.

[0037] The expression "at least one of a, b, and c" described throughout the specification may encompass 'a alone', 'b alone', 'c alone', 'a and b', 'a and c', 'b and c', or 'all of a, b, and c'.

[0038] The "terminal" mentioned below may be implemented as a computer or portable terminal capable of connecting to a server or other terminal via a network. Here, the computer includes, for example, a notebook, desktop, or laptop equipped with a web browser, and the portable terminal may include, for example, a wireless communication device that guarantees portability and mobility, such as a communication-based terminal such as IMT (International Mobile Telecommunication), CDMA (Code Division Multiple Access), W-CDMA (W-Code Division Multiple Access), LTE (Long Term Evolution), and all types of handheld-based wireless communication devices such as smartphones and tablet PCs.

[0039] In the following description, the terms “transmission,” “communication,” “sending,” “receiving,” or other similar terms of a signal or information include not only the direct transmission of a signal or information from one component to another, but also the transmission via another component.

[0040] In particular, "transmitting" or "sending" a signal or information as a single component indicates the final destination of the signal or information, not the direct destination. The same applies to "receiving" a signal or information. Furthermore, in this specification, "relating" two or more pieces of data or information means that, upon acquiring one piece of data (or information), at least a portion of the other piece of data (or information) can be acquired based on that piece of data (or information).

[0041] Additionally, while terms such as "first" and "second" may be used to describe various components, these components should not be limited by these terms. These terms may be used to distinguish one component from another. For example, without departing from the scope of the present invention, a first component may be referred to as a "second component," and similarly, a second component may also be referred to as a "first component."

[0042] In describing the embodiments, descriptions of technical details that are well known in the technical field to which the present invention pertains and are not directly related to the present invention will be omitted. This is to avoid obscuring the gist of the present invention by omitting unnecessary explanations and to convey the gist more clearly.

[0043] For the same reason, some components in the attached drawings are exaggerated, omitted, or schematically depicted. Furthermore, the dimensions of each component do not entirely reflect its actual size. Identical or corresponding components in each drawing are assigned the same reference numbers.

[0044] The advantages and features of the present invention, and the methods for achieving them, will become clearer with reference to the embodiments described in detail below together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various different forms. These embodiments are provided only to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Like reference numerals designate like elements throughout the specification.

[0045] It will be appreciated that each block of the flowchart drawings and combinations of the flowchart drawings can be performed by computer program instructions. These computer program instructions can be installed in a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing equipment, such that the instructions, when executed by the processor of the computer or other programmable data processing equipment, create a means for performing the functions described in the flowchart block(s). These computer program instructions can also be stored in a computer-available or computer-readable memory that can direct a computer or other programmable data processing equipment to perform the functions in a specific manner, such that the instructions stored in the computer-available or computer-readable memory can produce an article of manufacture that includes instruction means for performing the functions described in the flowchart block(s). Since the computer program instructions may be installed on a computer or other programmable data processing device, a series of operational steps may be performed on the computer or other programmable data processing device to create a computer-executable process, and the instructions that cause the computer or other programmable data processing device to perform the steps for performing the functions described in the flowchart block(s) may also provide steps for performing the functions described in the flowchart block(s).

[0046] Additionally, each block may represent a module, segment, or portion of code that contains one or more executable instructions for performing a specific logical function(s). It should also be noted that in some alternative implementation examples, the functions described in the blocks may occur out of order. For example, two blocks depicted in succession may actually be executed substantially concurrently, or the blocks may sometimes be executed in reverse order, depending on their respective functions.

[0047] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the attached drawings so that those skilled in the art can easily practice the present disclosure. However, the present disclosure may be implemented in various different forms and is not limited to the embodiments described herein. The embodiments illustrated in FIGS. 2 to 8 may be performed, for example, by an artificial intelligence interface system (100) described with reference to FIG. 8. The artificial intelligence interface system (100) is an exemplary device that updates a database using a conversation log in the present disclosure, and may be a single device, but depending on the embodiment, may be a system composed of multiple detailed devices that communicate with each other via a wired or wireless network.

[0048] FIG. 1 is a schematic diagram of a system for providing a natural language question-and-answer service, including an artificial intelligence interface system according to one embodiment. More specifically, FIG. 1 illustrates a system for verifying a stock strategy established by a user and performing a simulation or actual execution thereof. The SYSTEM illustrated in FIG. 1 represents a server-side system (10) that provides a natural language question-and-answer service, USER represents a user-side system (20) that is not directly controlled by the server-side system (10), and EXTERNAL represents an external real-time data source (30) accessed by the server-side system (10).

[0049] The server-side system (10) may include an artificial intelligence interface system (100), a user database (DB), a past data DB, a strategy verification engine, and a strategy execution engine. The user-side system (20) may include a user terminal (referred to as a "user" in this disclosure for convenience of explanation) and a user securities firm account.

[0050] The AI ​​interface system (100) receives user queries and provides responses. During the response generation process, it may access a user database containing past query-response exchanges with users and reference necessary information. For example, the AI ​​interface system (100) may reference information stored in the user database to identify information necessary for strategy verification or strategy execution, and, if additional information is required, request it from the user.

[0051] When the AI ​​interface system (100) receives a strategy verification request from a user, if the information or conditions required for strategy verification are met, it may request the strategy verification engine to verify the user's strategy. In this process, the AI ​​interface system (100) may write a request in a format that the strategy verification engine can process. For example, assuming that the strategy verification engine can process requests in the JSON format of a REST API, the AI ​​interface system (100) may generate a request in the JSON format using a language model that constitutes the AI ​​interface system (100).

[0052] Thereafter, the strategy verification engine can access the past data DB to verify the user's strategy, refer to past data related to the user's strategy, and return the result of verifying the user's strategy to the artificial intelligence interface system (100). Based on the result received from the strategy verification engine, the artificial intelligence interface system (100) can directly execute a strategy that is worth executing through the strategy execution engine, or provide the result received from the strategy verification engine to the user-side system (20) and then request the strategy execution engine to execute the strategy according to the user's request.

[0053] The strategy execution engine can execute the strategy in a simulation form (virtual execution) or actually execute it. The execution form may be determined according to the user's request, but depending on the embodiment, if the strategy verification engine verifies that the strategy does not meet the predefined execution conditions, it may be designed to be executed in a simulation form, and if the execution conditions are met, it may be designed to be actually executed. The strategy execution engine can access the user's securities account of the user-side system (20) to execute the strategy. In the present disclosure, the "user securities account" refers to the account information of the user account managed by the user's securities company. When the strategy is actually executed, the strategy execution engine can invest the funds deposited in the user's account according to the strategy and reflect the investment results in the user's securities account. On the other hand, when the strategy is executed in a simulation form, the strategy execution engine only refers to the amount of funds deposited in the user's account, virtually invests funds equivalent to that amount, and provides the user-side system (20) so that the virtual investment results can be confirmed in the user's account, but may not reflect information about the virtual investment results in the user's securities account. Meanwhile, the strategy execution engine can transmit the results of executing the user's strategy (or executing it in a simulation form) to an external real-time data source (30), and the transmitted execution results can be processed or partially extracted and stored in the past data DB of the server-side system (10).

[0054] FIG. 2 is a flowchart illustrating a method for updating a database using a conversation log in an artificial intelligence interface system according to an embodiment. The method illustrated in FIG. 2 may be performed by the artificial intelligence interface system, and the AI ​​interface (100) illustrated in FIG. 1 may serve as an example of the artificial intelligence interface system. Hereinafter, the term "database" refers to any database that the artificial intelligence interface system directly or indirectly references to provide a natural language question-and-answer service, and the user DB illustrated in FIG. 1 may be an example thereof.

[0055] In step S210, the artificial intelligence interface system can classify one or more question-answer sets included in a dialog log into one or more dialog records by conversation topic.

[0056] In the present disclosure, a "conversation log" refers to data recording conversation content in the process of providing a natural language question-and-answer service, and a "conversation record" refers to a collection of question-and-answer sets for the same conversation topic. For example, a conversation log may include a set of accumulated question-and-answers from the time of classification as a conversation log to the present. In addition, the set of question-and-answers included in the conversation log may include a user's question input into an artificial intelligence interface system and the artificial intelligence interface system's response to the question. Meanwhile, the conversation log may be stored in a temporary storage space (e.g., cache) or storage medium within the artificial intelligence interface system, or in a physical storage medium or cloud storage external to the artificial intelligence interface system. As conversations accumulate while the natural language question-and-answer service is provided, the capacity occupied by the stored conversation log also increases. To prevent problems caused by such capacity increase, the artificial intelligence interface system may adjust the capacity of the conversation log to an appropriate size. As an example of such a method, the AI ​​interface system may delete question-and-answer sets included in the conversation log that have a usage frequency below a certain level or that have been used for a certain period of time since they were included in the conversation log. In the present disclosure, "usage frequency" refers to the number of times the AI ​​interface system re-references a question-and-answer set recorded in the conversation log during a specific period of time, and may refer to, for example, the number of times the question-and-answer set was input into a language model for generating a topic summary during a specific period of time.

[0057] In one embodiment, the AI ​​interface system may request a language model constituting the AI ​​interface system to identify a conversation topic of each question-and-answer set included in a conversation log at each point in time when a preset condition is met, in order to classify the question-and-answer set into a dialogue log, and may classify each question-and-answer set into one or more dialogue transcripts based on the conversation topics identified by the language model. In the present disclosure, the 'language model' refers to a deep learning model trained based on big data, and may be implemented as a set of neural networks having an encoder and a decoder according to a transformer structure, for example. A technique of obtaining a desired result by providing appropriate instructions and dictionary information to a language model is called prompt engineering, and in the present disclosure, a request transmitted to the language model is made in the form of inputting information to be input by the AI ​​interface system according to a predefined prompt and then transmitting it to the language model. Meanwhile, in the present disclosure, the term "language model constituting an artificial intelligence interface system" means that the artificial intelligence interface system transmits a request to the language model and receives a natural language processed result from the language model in the process of providing a natural language question-and-answer service, and does not only mean that the language model is installed and operates within a device on which the artificial intelligence interface system is implemented. In other words, it should be noted that the language model in the present disclosure may include not only a language model installed on a device on which the artificial intelligence interface system is implemented, but also a language model operating on a cloud server.

[0058] In step S220, the AI ​​interface system can generate multiple topic summaries corresponding to the conversation topics in the conversation transcript. In this disclosure, a "topic summary" refers to a document summarizing the content of the conversation transcript regarding a specific conversation topic. Multiple topic summaries may exist for the same conversation topic. For example, the most recently generated topic summary and the most updated topic summary may exist together for the same conversation topic.

[0059] In one embodiment, the AI ​​interface system may search for a topic summary corresponding to a conversation topic of a dialogue transcript from a database storing multiple topic summaries, and, based on the search results, generate multiple topic summaries corresponding to the conversation topic of the dialogue transcript but having different word sizes. In the present disclosure, the "word size" refers to the number of words included in the text constituting the topic summary. For example, if the number of words included in topic summary a is 500, the word size of topic summary a may be set to 500. Accordingly, for example, when the AI ​​interface system generates two topic summaries (a first topic summary and a second topic summary), the first topic summary may be generated to have a relatively smaller word size than the second topic summary. This means that both the first topic summary and the second topic summary are documents corresponding to the conversation topic of the dialogue transcript, but the first topic summary is a relatively more concise document.

[0060] Since subject summaries are also data that occupy storage space when stored temporarily or long-term, a word size limit may be set for them to increase efficiency during storage and use. In such a case, some of the multiple subject summaries generated by the AI ​​interface system may be created to meet the word size limit, while others may be created to exceed the word size limit. In other words, when a word size limit is set for subject summaries, the AI ​​interface system may set different word sizes for some subject summaries and the remaining subject summaries based on the set word size limit. This is to improve the efficiency of the logic associated with step S230, which will be described in more detail below.

[0061] Meanwhile, the process of generating multiple topic summaries based on search results can be subdivided depending on whether a topic summary corresponding to the conversation topic of the transcript is found in the database. For example, if a topic summary corresponding to the conversation topic of the transcript is found in the database, the AI ​​interface system can input the transcript and the retrieved topic summary into the language model and request the generation of multiple topic summaries. Alternatively, if a topic summary corresponding to the conversation topic of the transcript is not found in the database, the AI ​​interface system can input the transcript into the language model and request the generation of multiple topic summaries.

[0062] In step S230, the artificial intelligence interface system can determine a subject summary to be stored in the database based on the plurality of generated subject summaries.

[0063] When a word size limit is set for a subject summary, the AI ​​interface system can set a different word size for some subject summaries and the remaining subject summaries based on the set word size limit. When the AI ​​interface system creates two subject summaries (a first subject summary that satisfies the set word size limit and a second subject summary that exceeds the word size limit), the subject summary to be stored in the database can be determined according to the following embodiments. However, the following embodiments are exemplary and the method for determining the subject summary to be stored in the database is not limited thereto.

[0064] The AI ​​interface system may evaluate the similarity between the first and second subject summaries. If the similarity is above a threshold, the first subject summary may be determined as the subject summary to be stored in the database. If the similarity is below the threshold, the AI ​​interface system may determine the subject summary to be stored in the database based on the transcript and the second subject summary. More specifically, if the similarity is below the threshold, the AI ​​interface system may input the transcript and the second subject summary into a language model and request the generation of a third subject summary that satisfies a word size restriction. The third subject summary thus generated may then be determined as the subject summary to be stored in the database.

[0065] Alternatively, if the similarity is below a threshold, the AI ​​interface system may input the transcript and the second topic summary into the language model, requesting the generation of a third topic summary that satisfies the word size limit and a fourth topic summary that exceeds the word size limit. Based on the third and fourth topic summaries generated accordingly, the AI ​​interface system may determine which topic summaries to store in the database. As one example, the AI ​​interface system may recursively apply the method of determining which topic summaries to store in the database using the first and second topic summaries described above to the third and fourth topic summaries.

[0066] Meanwhile, if the artificial intelligence interface system generates three subject summaries (a first subject summary that satisfies a preset word size limit, a second subject summary that exceeds the word size limit, and a fifth subject summary that satisfies the word size limit and has a word size larger than that of the first subject summary), the subject summaries to be stored in the database may be determined according to the following embodiments. However, the embodiments below are also exemplary, and the method of determining the subject summaries to be stored in the database is not limited thereto.

[0067] The artificial intelligence interface system can evaluate the first similarity between the first subject summary and the second subject summary, and if the first similarity is greater than a threshold, determine the first subject summary as the subject summary to be stored in the database.

[0068] Meanwhile, if the first similarity is below the threshold, the AI ​​interface system may evaluate the second similarity between the second subject summary and the fifth subject summary, and if the second similarity is above the threshold, determine the fifth subject summary as the subject summary to be stored in the database, and if the second similarity is below the threshold, determine the subject summary to be stored in the database based on the transcript and the second subject summary. Specifically, if the second similarity is below the threshold, the AI ​​interface system may input the transcript and the second subject summary into the language model and request the generation of a sixth subject summary that satisfies the word size limit, and determine the sixth subject summary thus generated as the subject summary to be stored in the database.

[0069] On the other hand, if the first similarity is below the threshold, the AI ​​interface system may evaluate the second similarity between the second subject summary and the fifth subject summary, and if the second similarity is above the threshold, determine the fifth subject summary as the subject summary to be stored in the database, and if the second similarity is below the threshold, input the transcript and the second subject summary into the language model to request the generation of a sixth subject summary that satisfies the word size limit and a seventh subject summary that exceeds the word size limit, and determine the subject summaries to be stored in the database based on the sixth and seventh subject summaries generated accordingly. As one example, the AI ​​interface system may recursively apply the method of determining the subject summaries to be stored in the database using the first and second subject summaries described above to the sixth and seventh subject summaries.

[0070] In step S240, the AI ​​interface system may store the determined topic summary in a database. In the present disclosure, "storing" the topic summary in the database means reflecting data referred to as the topic summary in the database, and the reflection method may be determined in various ways. For example, the AI ​​interface system may update the topic summary corresponding to the conversation topic of the topic summary determined in step S230 among the topic summaries stored in the database with the determined topic summary. Accordingly, the existing topic summary may be replaced with the determined topic summary. Alternatively, the AI ​​interface system may add the topic summary determined in step S230 to the database as a new topic summary. In this case, only a new topic summary is added to the database, and no topic summary is deleted or replaced accordingly.

[0071] Figure 3 is a conceptual diagram illustrating a method for classifying one or more question-and-answer sets contained in a conversation log into one or more conversation logs based on conversation topics. Therefore, the process illustrated in Figure 3 may be an example of concretizing step S210 of Figure 2.

[0072] A conversation log may record a question-answer set consisting of a user's query and an AI interface system's response. The question-answer sets (a, b, c) recorded in the conversation log may be categorized according to a sequential conversation history. If a new question-answer set (n) is generated, it may be added to the existing question-answer set (a, b, c). "Added to" here means that it is added chronologically according to the conversation history, not the actual recorded location.

[0073] A language model can analyze conversation logs via prompts to identify the core conversational topic of each question-and-answer set and reconstruct question-and-answer sets related to the same conversational topic by classifying them into the same transcript. For example, if the question-and-answer sets 'a', 'b', and 'c' relate to topic A, and the question-and-answer sets 'c', 'd', and 'n' relate to topic B, then each question-and-answer set is classified as a transcript corresponding to topic A and topic B, respectively.

[0074] Figure 4 is a conceptual diagram illustrating a method for generating multiple subject summaries. Therefore, the process illustrated in Figure 4 may be an example of concretizing step S220 of Figure 2.

[0075] The AI ​​interface system can search the database for topic summaries related to the subject of the user's query. Figure 4 illustrates the process of executing a query for topic A on the database to retrieve topic summaries for topic A. By using existing topic summaries, the newly generated transcripts generated by the language model can be integrated with existing knowledge in the database. If multiple topic summaries for topic A exist in the database, one of them can be selected. For example, the most recently generated topic summary for topic A can be selected. Conversely, if no topic summary related to the user's query exists in the database, the language model can generate a new topic summary based on transcript A.

[0076] An artificial intelligence interface system can input a dialogue A corresponding to topic A and an existing topic summary A in the form of a prompt into a language model to generate a topic summary that integrates the dialogue A and the existing topic summary A. The topic summary may have a word size limit, which may cause newly added information, newly generated insight information, and contextual information to be lost during the topic summary generation process. To prevent this, as illustrated in Fig. 4, multiple topic summaries A' and A'' with different word sizes may be temporarily generated for the same topic A. The word size of each of the multiple temporarily generated topic summaries may be determined arbitrarily, but some may be set to meet the word size limit, and others may be set to exceed the word size limit. Furthermore, some may be set to meet the word size limit, while others may exceed the word size limit and be set to be a multiple of the word size of the portion mentioned above. For example, if the word size limit is 800, when temporarily creating two subject summaries A' and A'', the word size of subject summary A' can be set to 500 and the word size of subject summary A'' can be set to 1000.

[0077] Figure 5 is a conceptual diagram illustrating a method for determining a topic summary to be stored in a database. Therefore, the process illustrated in Figure 5 may be an example of concretizing step S230 of Figure 2.

[0078] An AI interface system can use a similarity comparison model to evaluate the similarity between two subject summaries A' and A'' with different word sizes. The similarity comparison model may refer to the same language model, but depending on the embodiment, it may also refer to a separate, pre-trained AI model. Methods for evaluating the similarity between subject summaries include taking the inner product of the embedding vectors of each subject summary, inputting a similarity evaluation prompt into the language model, or utilizing any other known similarity evaluation algorithm.

[0079] If the similarity between the two subject summaries exceeds a threshold, the newly created subject summary A' can be considered a document that sufficiently summarizes the required information without losing content while meeting the word size limit. Therefore, subject summary A' with a smaller word size can be stored in the database.

[0080] Meanwhile, if the similarity between two topic summaries is below the threshold, the newly generated topic summary A' can be judged to be a document with content loss due to the word size limit. Therefore, topic summary A'' with a larger word size can be input as a prompt to the language model along with the transcript A to generate a new topic summary A''' that satisfies the word size limit and store it in the database. This process can have the effect of weighting the new content included in the transcript and reflecting it in the topic summary. Although not shown in Figure 5, the AI ​​interface system can also recursively repeat the similarity evaluation process by generating additional topic summaries A'''' that exceed the word size limit in addition to topic summary A''''. If this similarity evaluation process is repeated recursively, it can be assumed that the topic summary ultimately stored in the database will be a higher-quality topic summary that well summarizes the content of the conversation topic.

[0081] Although not shown in Fig. 5, the process of Fig. 5 can be extended to a case where three subject summaries are temporarily created. For example, if the word size limit is 800, subject summaries A' with a word size of 500, A'' with a word size of 1000, and A* with a word size of 700 can be assumed. In this case, the AI ​​interface system first evaluates the first similarity between subject summaries A' and A'', and if the first similarity is greater than or equal to a threshold, subject summary A' can be stored in the database. This is because subject summary A' is determined to be a document that sufficiently summarizes the necessary information in the smallest word size without losing content. Meanwhile, the AI ​​interface system evaluates the second similarity between subject summaries A'' and A* if the first similarity is less than the threshold, and if the second similarity is greater than or equal to the threshold, subject summary A* can be stored in the database. This is because the subject summary A* was judged to be a document that sufficiently summarizes the necessary information in a word size smaller than the subject summary A' without any content loss. Meanwhile, if the second similarity is below the standard, the AI ​​interface system can input the dialogue A and the subject summary A'' into the language model in the form of a prompt to generate a new subject summary that satisfies the word size limit and store it in the database. This is because both the subject summaries A' and A*, which satisfy the word size limit, were judged to have suffered from content loss, and therefore a new subject summary that satisfies the word size limit was judged necessary.

[0082] Furthermore, the above logic can be applied in the case where four or more subject summaries are generated. The subject summary (A_n) with the largest word size will be generated to exceed the word size limit, and the remaining subject summaries will be generated to satisfy the word size limit but have different word sizes. The AI ​​interface system can implement the above logic by sequentially evaluating the similarity with subject summary A_n starting from the subject summary with the smallest word size, and if the similarity is above a threshold, storing the subject summary used for the similarity evaluation with subject summary A_n in the database, and if the similarity is below the threshold, evaluating the similarity between subject summary A_n and the subject summary with the next smallest word size.

[0083] Meanwhile, when storing topic summaries in a database, the AI ​​interface system can replace (update) existing topic summaries with newly stored ones if the similarity exceeds a threshold. If the similarity falls below the threshold, the existing topic summaries can be preserved while the newly stored ones are added. This allows for the integration of new knowledge and insights into the topic, while preserving a history of past updates and keeping knowledge related to the topic up to date. In other words, it can simultaneously achieve both incremental knowledge accumulation and improvement.

[0084] FIG. 6 is a flowchart illustrating a method for providing a response to a user's query using an updated database according to one embodiment. The method illustrated in FIG. 6 may be performed by an AI interface system, and the AI ​​interface (100) illustrated in FIG. 1 may serve as an example of such an AI interface system.

[0085] In step S610, the artificial intelligence interface system can receive a user's query input into the artificial intelligence interface system.

[0086] In step S620, the artificial intelligence interface system can identify the subject of the user's query by performing natural language processing on the user's query through a language model constituting the artificial intelligence interface system.

[0087] In step S630, the AI ​​interface system may identify a subject summary corresponding to the subject of the query in the database. However, step S630 does not necessarily mean that the AI ​​interface system identifies "all" subject summaries corresponding to the subject of the query in the database. For example, the AI ​​interface system may identify the most recently added subject summary corresponding to the subject of the query in the database. As another example, the AI ​​interface system may identify the most frequently updated subject summary corresponding to the subject of the query in the database.

[0088] In step S640, the artificial intelligence interface system can input the identified topic summary into a language model constituting the artificial intelligence interface system to generate a response to the query.

[0089] However, there may be cases where the AI ​​interface system, after processing the user's query in natural language at step S620, requires interaction with an external system connected to the AI ​​interface system to process the query. The strategy verification engine illustrated in FIG. 1 is an example of such an external system. In this case, the AI ​​interface system can call an Application Program Interface (API) for the external system and receive the processed system result from the external system through the API. Thereafter, at step S640, the AI ​​interface system can input the system result together with the identified topic summary into a language model to generate a response to the user's query. In this embodiment, the process of the AI ​​interface system calling the API for the external system and receiving the system result accordingly may be performed before or after step S630 described above, or may be performed simultaneously in parallel.

[0090] In step S650, the artificial intelligence interface system can provide a response to the query to the user.

[0091] Meanwhile, the AI ​​interface system can also record a question-and-answer set consisting of the query and its responses in a conversation log. The AI ​​interface system can then categorize these recorded conversation logs into conversation logs by conversation topic, generate multiple topic summaries corresponding to the conversation topics in the conversation logs, and determine a specific topic summary based on the generated topic summaries and store it in a database. In other words, the AI ​​interface system can not only provide answers to user queries, but also use newly accumulated question-and-answer sets to update the database.

[0092] Figure 7 is a conceptual diagram exemplifying the method of Figure 6.

[0093] When a user's query is input into a language model in the form of a prompt, the AI ​​interface system can process the query in natural language through the language model to identify topics A and B contained in the query. The language model can then query the database for topics A and B, respectively, to search for topic summaries A and B. Although not illustrated in Figure 7, the language model can also search for multiple topic summaries for a single topic in the database. In such cases, the criteria for searching for topic summaries can be designed in various ways. For example, assuming that two topic summaries for topic A are being searched, the most recent topic summary and the most frequently updated topic summary among the multiple topic summaries for topic A stored in the database can be searched, respectively. This is because the most recent topic summary contains the most up-to-date information reflecting current trends and developments, while the most frequently updated topic summary is more likely to contain proven, important information about the topic and have a relatively high proportion of data that has been verified over time.

[0094] Meanwhile, if the AI ​​interface system determines that interaction with an external system connected to the AI ​​interface system is necessary for query processing, it can call the external system's Application Program Interface (API). This allows communication with the external system when querying specific data on a specific topic or processing and manipulating such data. For example, if the query includes a request for verification of a specific stock strategy, the AI ​​interface system can generate a request using a language model to utilize the capabilities of the strategy verification engine connected to the AI ​​interface system and make a request to the strategy verification engine.

[0095] Meanwhile, the AI ​​interface system can generate a response to a query by inputting the retrieved topic summary (and system results processed via the API) in the form of a prompt into a language model. The user's query and its response are recorded in a conversation log as a single question-and-answer set, which can then be utilized to generate the aforementioned topic summary. This allows the AI ​​interface system to continuously learn and generate more accurate and relevant answers based on past conversations.

[0096] Applying the above-described process to a stock strategy verification case, referring to Figure 7, the following scenario is assumed. If a user queries, "Evaluate a strategy that sells when the stock price enters a buy signal on the Moving Average Convergence Divergence (MACD) and touches the upper Bollinger Band," the AI ​​interface system would process this query.

[0097] First, the AI ​​interface system processes queries in natural language using a language model, identifying two key topics: "MACD buy signal" and "Bollinger Band upper touch sell strategy." Each identified topic is then categorized into a conversation transcript, along with related question-and-answer sets from the existing conversation log.

[0098] Afterwards, the AI ​​interface system can query the database for each topic and explore existing thematic summaries related to MACD and Bollinger Bands.

[0099] Meanwhile, the AI ​​interface system connects to the strategy verification engine to simulate the user's stock strategy. During this process, the AI ​​interface system references the topic summary to identify the information necessary for strategy execution. If additional information is needed, it can ask the user additional questions. Otherwise, the AI ​​interface system can reference a predefined API to create a request in a format that the strategy verification engine can process. Upon receiving the request, the strategy verification engine can perform a simulation on the given strategy and return the results to the AI ​​interface system.

[0100] Afterwards, the AI ​​interface system can input simulation results (system results) and a topic summary into a language model in the form of prompts to generate a report for the user. For example, this report could include information on the strategy's effectiveness, potential risks, and expected returns.

[0101] Meanwhile, the AI ​​interface system may store a new set of question-response responses in the conversation log, which includes the user's query and the AI ​​interface system's response (e.g., the report itself may be used as the response, or at least part of the text of the report may be used as the response).

[0102] FIG. 8 is a block diagram illustrating an artificial intelligence interface system (100) that updates a database using a conversation log according to one embodiment.

[0103] The artificial intelligence interface system (100) may include, according to one embodiment, a transceiver (101), a processor (103), and a memory (105). In one embodiment, the artificial intelligence interface system (100) may exchange data with the outside world through the transceiver (101).

[0104] The processor (103) can perform at least one method described above with reference to FIGS. 2 and 6. The memory (105) can store information for performing at least one method described above with reference to FIGS. 2 and 6. The memory (105) can be a volatile memory or a non-volatile memory. The processor (103) can control the artificial intelligence interface system (100) to execute a program and provide information. The code of the program executed by the processor (103) can be stored in the memory (105). In one embodiment, the processor (103) is connected to the memory (105) and can classify one or more question-answer sets included in a dialogue log into one or more dialogue records by dialogue topic, generate a plurality of topic summaries corresponding to the dialogue topics of the dialogue log, determine a topic summary to be stored in a database based on the generated plurality of topic summaries, and store the determined topic summaries in the database.

[0105] Additionally, the artificial intelligence interface system (100) according to one embodiment may further include an interface capable of providing information to a user.

[0106] The artificial intelligence interface system (100) illustrated in FIG. 8 only illustrates components related to the present embodiment. Therefore, those skilled in the art will understand that, in addition to the components illustrated in FIG. 8, other general-purpose components may be included.

[0107] The device according to the above-described embodiments may include a processor, a memory for storing and executing program data, permanent storage such as a disk drive, a communication port for communicating with an external device, a user interface device such as a touch panel, a key, a button, etc. The methods implemented as software modules or algorithms may be stored on a computer-readable recording medium as computer-readable codes or program instructions executable on the processor. Here, the computer-readable recording medium includes a magnetic storage medium (e.g., read-only memory (ROM), random-access memory (RAM), floppy disk, hard disk, etc.) and an optical reading medium (e.g., CD-ROM, DVD: Digital Versatile Disc)). The computer-readable recording medium may be distributed to computer systems connected to a network, so that the computer-readable code may be stored and executed in a distributed manner. The medium may be readable by a computer, stored in a memory, and executed by a processor.

[0108] The present embodiment may be represented by functional block configurations and various processing steps. These functional blocks may be implemented by various hardware and / or software configurations that perform specific functions. For example, the embodiment may employ direct circuit configurations such as memory, processing, logic, look-up tables, etc., which may perform various functions under the control of one or more microprocessors or other control devices. Similarly, the present embodiment may be implemented in a programming or scripting language such as C, C++, Java, assembler, etc., including various algorithms implemented as a combination of data structures, processes, routines, or other programming configurations. Functional aspects may be implemented as algorithms that execute on one or more processors. Furthermore, the present embodiment may employ conventional techniques for electronic configuration, signal processing, message processing, and / or data processing. Terms like "mechanism," "element," "means," and "composition" can be used broadly and are not limited to mechanical or physical components. These terms can also encompass a series of software routines, such as those associated with a processor.

[0109] The above-described embodiments are merely examples, and other embodiments may be implemented within the scope of the claims set forth below.

Claims

1. A method for updating a database using a conversation log in an artificial intelligence interface system, A step of classifying one or more question-answer sets included in a dialog log into one or more dialog records by conversation topic; A step of generating multiple topic summaries corresponding to the conversation topics of the above conversation transcript; A step of determining a subject summary to be stored in the database based on the plurality of subject summaries; and A method comprising the step of storing the above-determined topic summary in the database.

2. In paragraph 1, The above conversation log is, Contains the accumulated set of questions and answers from the time of classification into recent dialogue records to the present. The above question-answer set is, A method including a user's query entered into the artificial intelligence interface system and a response from the artificial intelligence interface system to the query.

3. In paragraph 1, The above classification steps are: At each point when a preset condition is met, a step of requesting the language model constituting the artificial intelligence interface system to identify the conversation topic of each of the question-answer sets included in the conversation log; and A method comprising the step of classifying each of the question-answer sets into one or more dialogue transcripts based on the conversation topic identified by the language model.

4. In paragraph 1, A method further comprising a step of deleting, among the question-answer sets included in the above conversation log, question-answer sets whose frequency of use is below a certain level or whose time period has passed since their inclusion in the above conversation log.

5. In paragraph 1, The above generating steps are: A step of searching for a topic summary corresponding to the conversation topic of the conversation transcript in the database; and A step of generating a plurality of topic summaries corresponding to the conversation topics of the conversation log according to the search results for the database, The above multiple subject summaries are: A method characterized in that each of the plurality of subject summaries has a different word size.

6. In paragraph 5, The step of generating the multiple subject summaries according to the above search results is: When a topic summary corresponding to the conversation topic of the conversation transcript is searched in the database, a step of requesting the generation of the plurality of topic summaries by inputting the conversation transcript and the searched topic summary into a language model; and A method comprising a step of requesting the generation of the plurality of subject summaries by inputting the conversation transcript into a language model when a topic summary corresponding to the conversation topic of the conversation transcript is not found in the database.

7. In paragraph 1, The above multiple subject summaries are: A method comprising a first subject summary that satisfies a preset word size limit and a second subject summary that exceeds the preset word size limit.

8. In paragraph 7, The above decision step is, A step of evaluating the similarity between the first subject summary and the second subject summary; If the similarity is greater than the standard value, a step of determining the first subject summary as the subject summary to be stored in the database; and A method comprising the step of determining a subject summary to be stored in the database based on the conversation log and the second subject summary when the similarity is less than the criterion.

9. In paragraph 8, If the above similarity is less than the criterion, the step of determining the subject summary to be stored in the database is: A step of requesting the generation of a third subject summary that satisfies the word size restriction by inputting the above conversation transcript and the second subject summary into the language model; and A method comprising a step of determining the third subject summary as a subject summary to be stored in the database.

10. In paragraph 8, If the above similarity is less than the criterion, the step of determining the subject summary to be stored in the database is: A step of requesting the generation of a third subject summary that satisfies the word size limit and a fourth subject summary that exceeds the word size limit by inputting the above-mentioned transcript and the second subject summary into a language model; and A method comprising a step of determining a subject summary to be stored in the database based on the third subject summary and the fourth subject summary.

11. In paragraph 7, The above multiple subject summaries are: Further comprising a fifth subject summary that satisfies the above word size limitation and has a word size larger than the word size of the first subject summary; The above decision step is, A step of evaluating a first similarity between the first subject summary and the second subject summary; A step of determining the first subject summary as a subject summary to be stored in the database when the first similarity is greater than or equal to the criterion; If the above first similarity is below the criterion: A step of evaluating the second similarity between the second subject summary and the fifth subject summary; If the second similarity is greater than or equal to the criterion, a step of determining the fifth subject summary as the subject summary to be stored in the database; and A method comprising the step of determining a subject summary to be stored in the database based on the dialogue record and the second subject summary when the second similarity is less than the reference value.

12. In paragraph 1, The above saving step is, A method comprising the step of updating a subject summary corresponding to a conversation topic of the determined subject summary among the subject summaries stored in the database with the determined subject summary, or adding the determined subject summary as a new subject summary to the database.

13. In paragraph 1, A step of receiving a user's query entered into the above artificial intelligence interface system; A step of identifying the subject of the query by processing the query in natural language through a language model constituting the artificial intelligence interface system; A step of identifying the most recently added subject summary and the most frequently updated subject summary among the subject summaries corresponding to the subject of the query in the database; A step of inputting the identified topic summary into the language model to generate a response to the query; providing the above response to the user; and A method further comprising the step of recording a question-response set comprising the question and the response in the conversation log.

14. In paragraph 13, If, as a result of natural language processing for the above query, it is determined that interaction with an external system connected to the artificial intelligence interface system is necessary for processing the query, a step of calling an API (Application Program Interface) for the external system; and Further comprising a step of receiving the system result processed through the API from the external system, The steps for generating a response to the above query are: A method comprising the step of inputting the identified topic summary and the system result together into the language model to generate a response to the query.

15. An artificial intelligence interface system that updates a database using conversation logs. Includes a memory and a processor that stores instructions, The above processor is connected to the above memory, Classify one or more question-answer sets contained in a dialogue log into one or more dialogue records by conversation topic, Generate multiple topic summaries corresponding to the conversation topics of the above conversation transcript, Based on the above multiple subject summaries, determine the subject summaries to be stored in the database, An artificial intelligence interface system that stores the above-determined topic summary in the above-determined database.

Images