TWI846528B - Customizing setting and updating download system with proactive chat response mode and method thereof - Google Patents

Abstract

A customizing setting and updating download system with proactive chat response mode and method thereof is disclosed. By generating a rough inquiry message with natural language structure based on a user behavior state and a plurality of customized parameters, and inputting the rough inquiry message into a logic circuit to generate a precise inquiry message and sending the precise inquiry message to an artificial intelligence (AI) platform to obtain a corresponding response message as a first training data, and using the customized parameters as a second training data, and inputting the first training data and the second training data into an AI model for training. After training is completed, extracting a plurality of weight values and translating them to correspond to different customized parameters, and providing the translated results to a client for download, so as to execute customization setting and updating. The mechanism is help to improve the convenience and specificity of model customization.

Description

Customized settings and update download system with active chat response and method thereof

The present invention relates to a system and method for customized settings and update downloads, and in particular to a system and method for customized settings and update downloads with active chat response.

In recent years, with the popularization and rapid development of artificial intelligence, various applications of artificial intelligence have emerged like mushrooms after rain. Among them, chatbots have attracted the most attention.

Generally speaking, traditional AI-based chatbots use AI models to output dialogues. However, general AI models only give standardized answers to users' questions, so there are problems such as dull dialogues and lack of authenticity, which greatly reduces users' willingness to talk to chatbots. On the other hand, since the answer method is the same for all users, it lacks exclusivity and may cause problems with different users.

In view of this, some manufacturers have proposed technical means to train artificial intelligence models to make their answers emotional. For example, when generating answers, they can further embed emotional words and sentences based on the answers. However, although this method can make the answers no longer rigid, it also cannot produce exclusive dialogues, that is, there will not be different dialogues for different users, so it still cannot effectively solve the problem of insufficient convenience and unclear exclusivity of model customization. On the other hand, if the artificial intelligence model is to be customized, the server host needs a lot of computing power, which is also a problem to be solved.

In summary, it can be seen that the previous technology has long had problems such as insufficient convenience in model customization and unclear exclusivity, as well as excessive computing power load on the server host. Therefore, it is necessary to propose improved technical means to solve this problem.

The present invention discloses a customized setting and update downloading system with active chat response and its method.

First, the present invention discloses a customized setting and update downloading system with active chat response, which includes: an artificial intelligence platform, a client host and a server host. Among them, the artificial intelligence platform is used to receive precise question messages through an application programming interface (API), and input the precise question messages into a large language model to generate answer messages, and then send the answer messages to the server host through the application programming interface. Then, in the client host part, it includes: a sensor, a first non-transient computer-readable storage medium and a first hardware processor. The sensor is used to continuously sense at least one of physiological state, facial expression and body movement to generate user behavior state; the first non-transitory computer-readable storage medium is used to store multiple first computer-readable instructions; the first hardware processor is electrically connected to the first non-transitory computer-readable storage medium and the sensor to execute the multiple first computer-readable instructions, so that the client host downloads the update package and continuously transmits the user behavior state and customized parameters, wherein the update package provides a state template to allow the customized parameters to be set and updated. In addition, the server host is connected to the client host to receive user behavior status and customized parameters. The server host includes: a logic circuit, a second non-transitory computer-readable storage medium and a second hardware processor. The logic circuit includes a first finite state machine and a second finite state machine connected in series, wherein the first finite state machine receives a rough question message and an answer message, and the output of the first finite state machine is used as the input of the second finite state machine, so that the second finite state machine generates a precise question message and outputs it to the artificial intelligence platform through the application program interface; the second non-transient computer-readable storage medium is used to store a plurality of second computer-readable instructions; the second hardware processor is electrically connected to the second non-transient computer-readable storage medium and the logic circuit, and is used to execute a plurality of second computer-readable instructions, so that the server host executes: generating a self-generated instruction according to the received user behavior state; A rough question message of a language structure is input into a logic circuit; when the logic circuit generates a corresponding precise question message according to the rough question message and outputs it to an artificial intelligence platform, an answer message corresponding to the precise question message is received from the artificial intelligence platform as a first training data; a customized parameter is used as a second training data, and the second training data and the first training data are input into an artificial intelligence model for training. After the training is completed, a plurality of weight values of the artificial intelligence model are captured; and the weight values are translated to correspond to different customized parameters respectively, and the translated results are embedded into a status template of an update package to provide a client host with download.

In addition, the present invention also discloses a customized setting and update downloading method with active chat response, the steps of which include: connecting the server host to the artificial intelligence platform and the client host respectively; the client host continuously senses at least one of the physiological state, facial expression and body movement through the sensor to generate the user behavior state; the client host downloads the update package, wherein the update package provides a state template to allow the setting and update of the customized The client host continuously transmits the user behavior state and customized parameters to the server host; the server host generates a rough question message with a natural language structure according to the received user behavior state and customized parameters, and inputs the rough question message into the first finite state machine and the second finite state machine connected in series for parsing and state conversion to generate a precise question message, wherein the first finite state machine receives the rough question message and the information from the artificial intelligence. The output of the first finite state machine is used as the input of the second finite state machine, and the second finite state machine outputs the precise question message to the artificial intelligence platform through the application programming interface of the artificial intelligence platform; the artificial intelligence platform inputs the precise question message into the large language model to generate the answer message, and then transmits the answer message to the server host through the application programming interface; the server host receives the precise question message from the artificial intelligence platform. The answer message is used as the first training data, and the received customized parameters are used as the second training data, and the second training data and the first training data are input into the artificial intelligence model for training. When the training is completed, multiple weight values of the artificial intelligence model are extracted; the server host translates the weight values to correspond to different customized parameters respectively, and embeds the translated results into the status template of the update package to provide the client host with download.

The system and method disclosed in the present invention are as described above. The difference from the prior art is that the present invention generates a rough question message with a natural language structure through user behavior status and customized parameters, and then inputs the rough question message into a logic circuit to generate a precise question message and transmits it to an artificial intelligence platform to obtain a corresponding answer message as the first training data, and uses the customized parameters as the second training data, and then inputs the two types of training data into an artificial intelligence model for training. After the training is completed, the weight value is extracted and translated to correspond to different customized parameters respectively, and the translated result is provided to the client host for downloading so as to perform customized settings and updates.

Through the above-mentioned technical means, the present invention can achieve the technical effect of improving the convenience and exclusivity of model customization.

The following will be used in conjunction with drawings and embodiments to explain the implementation of the present invention in detail, so that the implementation process of how the present invention applies technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

First, please refer to "Figure 1", which is a system block diagram of the customized setting and update download system with active chat response of the present invention, which includes: an artificial intelligence platform 110, a client host 120 and a server host 130. Among them, the artificial intelligence platform 110 is used to receive a precise question message through an application program interface, and input the precise question message into a large language model to generate an answer message, and then send the answer message to the server host 130 through the application program interface. In actual implementation, the artificial intelligence platform 110 is a chat robot using a large language model, such as a generative pre-trained transformer model (GPT), PaLM, Galactica, LLaMA, LaMDA or the like.

The client host 120 includes: a sensor 121, a first non-transitory computer-readable storage medium 122 and a first hardware processor 123. The sensor 121 is used to continuously sense at least one of physiological state, facial expression and body movement to generate user behavior state. In actual implementation, the sensor 121 can sense physiological characteristics such as blood pressure, heartbeat, pulse, blood sugar, etc. to determine physiological state, such as: happiness, excitement, frustration, etc.; or by sensing the face, iris, etc. to determine facial expression and mood; or by wearing sensors on the limbs of the human body, such as: three-axis acceleration sensor, gyroscope, etc. to sense the user's body movement. For example: walking, running, dancing, etc.

The first non-transitory computer-readable storage medium 122 is used to store a plurality of first computer-readable instructions. In practical implementation, the first non-transitory computer-readable storage medium 122 may include a hard disk, an optical disk, a flash memory or the like. In addition, the first computer-readable instructions refer to instructions that can be interpreted and executed by the client host 120, such as a client computer (or computer).

The first hardware processor 123 is electrically connected to the first non-transitory computer-readable storage medium 122 and the sensor 121 to execute the plurality of first computer-readable instructions, so that the client host 120 downloads the update package and continuously transmits the user behavior status and customized parameters, wherein the update package provides a status template to allow the customized parameters to be set and updated. In actual implementation, the customized parameters may include time information and filtering parameters. Taking time messages as an example, they may include year, month, day, hour, minute, second, or even time range, etc., which are used as the basis for determining whether the answer message is related to time, for example, to determine whether the answer message has expired, set timed feedback (such as sending an on-demand dialogue message to remind you to eat at 12:00 a.m. every day), or other situations related to time, such as: morning is related to breakfast, 00:00 to 4:00 a.m. is related to sleep, national holidays are related to fixed dates, birthdays are related to specified dates, etc. Therefore, when the time falls within the morning range, answer messages related to breakfast can be filtered out, when the time falls between 00:00 and 4:00 a.m., answer messages related to sleep can be filtered out, and when the time is a national holiday, answer messages containing this national holiday can be filtered out. In addition, taking filtering parameters as an example, the filtering parameters are used to set the reply messages that are agreed to be received and the reply messages that are refused to be received. For example, the user can set it according to his or her own preferences, such as: setting to filter out chat content containing politics and religion, setting to receive only entertainment chat content, setting to filter out content with a depressing tone, etc.

Next, in the part of the server host 130, it is connected to the client host 120 to receive the user behavior state and customized parameters. The server host 130 includes: a logic circuit 131, a second non-transitory computer-readable storage medium 132 and a second hardware processor 133. Among them, the logic circuit 131 includes a first finite state machine and a second finite state machine connected in series, wherein the first finite state machine receives a rough question message and an answer message, and the output of the first finite state machine is used as the input of the second finite state machine, so that the second finite state machine generates a precise question message and outputs it to the artificial intelligence platform 110 through the application program interface. In actual implementation, the first finite state machine can be a Mealy Machine, and its output is affected by the current state, the rough question message and the answer message. In addition, the second finite state machine can be a Moore type finite state machine (Moore Machine) state machine, and its output is only affected by the current state. In practice, each customized parameter can be converted into a state table first, and then the transition table of the flip-flop can be set according to the excitation table of the flip-flop, and then the input equation of each flip-flop can be obtained by using the Karnaugh Map to generate the final circuit diagram of the finite state machine, thereby realizing the logic circuit 131.

The second non-transitory computer-readable storage medium 132 is used to store a plurality of second computer-readable instructions. In actual implementation, the second non-transitory computer-readable storage medium 132 is different from the first non-transitory computer-readable storage medium 122 in that the former is a non-transitory computer-readable storage medium of the server host 130, storing computer-readable instructions for execution by the server host 130 (i.e., second computer-readable instructions), while the latter is a non-transitory computer-readable storage medium of the client host 120, storing computer-readable instructions for execution by the client host 120 (i.e., first computer-readable instructions).

The second hardware processor 133 is electrically connected to the second non-transitory computer-readable storage medium 132 and the logic circuit 131 to execute a plurality of second computer-readable instructions to enable the server host 130 to execute: generating a rough question message with a natural language structure according to the received user behavior state to input into the logic circuit 131; when the logic circuit 131 generates a corresponding precise question message according to the rough question message and outputs it to the artificial intelligence platform 110, the artificial intelligence platform 110 generates a precise question message according to the rough question message. The platform 110 receives the answer message corresponding to the precise question message as the first training data; uses the customized parameters as the second training data, and inputs the second training data and the first training data into the artificial intelligence model for training. After the training is completed, multiple weight values of the artificial intelligence model are extracted; and the weight values are translated to correspond to different customized parameters respectively, and the translated results are embedded in the status template of the update package to provide the client host 120 with download. For example, assuming that the user behavior state received by the server host 130 is "excited", the customized parameters include the time message "AM 08:00" and the filter parameter "excluding frustrated sentences", then the second hardware processor 133 will generate a corresponding rough question message according to "excited", "AM 08:00" and "excluding frustrated sentences", such as: "excited, AM 08:00, excluding frustrated sentences", and then input this rough question message into the finite state machine of the logic circuit 131 for parsing and state conversion, for example: define the question type according to "excited", and define the question type according to "AM 08:00” to define the specific time status of the question, and define the scope of the question according to “excluding frustrated statements”, and then use predefined templates or grammatical rules to generate precise question messages, such as “Please list five reasons why you are excited and not frustrated at 8 o’clock in the morning”, “It is 8 o’clock in the morning and I am very excited. Do you have any suggestions?”. Next, the precise question message is sent to the artificial intelligence platform 110 to obtain a corresponding answer message. At this time, the obtained answer message will be fed back to the logic circuit 131 to automatically identify the relevance between the question and the answer, and then dynamically adjust the state setting of the finite state machine to further improve the accuracy of the question. For example, the question with better relevance is automatically set to replace the question with poor relevance to generate an accurate question message. In addition, the answer message will also be used as the first training data, and the customized parameters will be used as the second training data, and the first training data and the second training data will be input into the artificial intelligence model for training. When the training is completed, multiple weight values of the artificial intelligence model are captured. Finally, the second hardware processor 133 translates the weight values to correspond to different customized parameters, and embeds the translated results into the status template of the update package to provide the client host 120 with download. For example, interpretive techniques such as Local Interpretable Model-agnostic Explanations (LIME), Shapley Additive Explanations (SHapley Additive ExPlanationsl, SHAP) and so on to obtain the influence of weight values on specific instances, and translate these influences into human-understandable explanations or parameters. For example, the output of this data by the artificial intelligence model is affected by certain keywords of the customized parameters, so the weight values are mapped to these keywords and embedded into the state template. In this way, users can easily understand from the state template which customized parameters have a greater impact on the output results of the artificial intelligence model and which customized parameters have a smaller impact, and then control the customized parameters according to the impact, for example, avoid using customized parameters with a smaller impact.

It should be particularly noted that, in actual implementation, the present invention may be partially or completely implemented based on hardware. For example, one or more components in the system may be implemented through hardware processors such as integrated circuit chips, system on chip (SoC), complex programmable logic devices (CPLD), field programmable gate arrays (FPGA), etc. The non-transitory computer-readable storage medium described in the present invention is loaded with computer-readable instructions (or computer program instructions) that are useful for the processor to implement various aspects of the present invention. The non-transitory computer-readable storage medium may be a tangible device that can retain and store instructions used by an instruction execution device. The non-transitory computer-readable storage medium may be, but is not limited to, an electric storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: hard drive, random access memory, read-only memory, flash memory, optical disk, floppy disk, and any suitable combination of the above. As used herein, non-transitory computer-readable storage media is not to be construed as a transient signal per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., optical signals through optical fiber cables), or electrical signals transmitted through wires. In addition, the computer-readable instructions described herein may be downloaded from a non-transitory computer-readable storage medium to various computing/processing devices, or downloaded to an external computer device or external storage device through a network, such as the Internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, hubs, and/or gateways. The network card or network interface in each computing/processing device receives the computer readable instructions from the network and forwards the computer readable instructions for storage in the non-transitory computer readable storage medium in each computing/processing device. The computer-readable instructions for executing the operations of the present invention may be assembly language instructions, instruction set architecture instructions, machine instructions, machine-related instructions, microinstructions, firmware instructions, or source code or object code written in any combination of one or more programming languages, wherein the programming languages include object-oriented programming languages such as Common Lisp, Python, C++, Objective-C, Smalltalk, Delphi, Java, Swift, C#, Perl, Ruby and PHP, as well as conventional procedural programming languages such as C or similar programming languages.

Please refer to "FIG. 2A" and "FIG. 2B", which are the method flow charts of the method for customized setting and update downloading of active chat response of the present invention, and the steps include: connecting the server host 130 to the artificial intelligence platform 110 and the client host 120 respectively (step 210); the client host 120 continuously senses at least one of the physiological state, facial expression and body movement through the sensor 121 to generate the user behavior state (step 220); the client host 120 downloads the update package , wherein the update kit provides a state template to allow setting and updating of customized parameters (step 230); the client host 120 continuously transmits the user behavior state and customized parameters to the server host 130 (step 240); the server host 130 generates a rough question message with a natural language structure according to the received user behavior state and customized parameters, and inputs the rough question message into the first finite state machine and the second finite state machine connected in series for parsing and state conversion to generate a precise question message, wherein the first finite state machine The rough question message and the answer message from the artificial intelligence platform 110 are received, the output of the first finite state machine is used as the input of the second finite state machine, and the second finite state machine outputs the precise question message to the artificial intelligence platform 110 through the application programming interface of the artificial intelligence platform 110 (step 250); the artificial intelligence platform 110 inputs the precise question message into the large language model to generate the answer message, and then transmits the answer message to the server host 130 through the application programming interface (step 260); the server host 130 receives the answer message from the artificial intelligence platform The station 110 receives the answer message corresponding to the precise question message as the first training data, and the received customized parameters as the second training data, and inputs the second training data and the first training data into the artificial intelligence model for training. After the training is completed, multiple weight values of the artificial intelligence model are extracted (step 270); the server host 130 translates the weight values to correspond to different customized parameters respectively, and embeds the translated results into the status template of the update package to provide the client host 120 with download (step 280). Through the above steps, a rough question message with a natural language structure can be generated through the user behavior status and customized parameters, and then the rough question message is input into the logic circuit to generate a precise question message and transmitted to the artificial intelligence platform 110 to obtain the corresponding answer message as the first training data, and the customized parameters are used as the second training data, and then the two types of training data are input into the artificial intelligence model for training. After the training is completed, the weight value is extracted and translated to correspond to different customized parameters respectively, and the translated result is provided to the client host 120 for downloading so as to perform customized settings and updates.

The following is explained in the form of an example with reference to "Figure 3" and "Figure 4". As shown in "Figure 3", "Figure 3" is a schematic diagram of the client host, server host and artificial intelligence platform to which the present invention is applied. First, the server host 130 is connected to the artificial intelligence platform 110 and the client host 120 respectively. The client host 120 continuously senses at least one of the user's physiological state, facial expressions and body movements through sensors to generate user behavior status, and downloads an update package, wherein the update package provides a status template to allow the setting and updating of multiple customized parameters. Then, the client host 120 continuously transmits the user behavior status and customized parameters to the server host 130 through the input interface. The state management module of the server host 130 generates a rough question message with a natural language structure according to the received user behavior state and customized parameters, and inputs the generated rough question message into the first finite state machine (e.g., Mealy type finite state machine) and the second finite state machine (e.g., Moore type finite state machine) connected in series for parsing and state conversion to generate a precise question message. Then, the server host 130 transmits the precise question message to the artificial intelligence platform 110 through the API of the artificial intelligence platform 110. At this time, the artificial intelligence platform 110 inputs the precise question message into the large language model to generate an answer message, and then transmits the answer message to the server host 130 through the API. When the server host 130 receives the answer message corresponding to the precise question message from the artificial intelligence platform 110, the answer message is input into the first finite state machine to confirm the relevance between the question and the answer, so as to further optimize the rough question message into the precise question message. In addition, the server host 130 also uses the received answer message as the first training data and the received customized parameters as the second training data, and inputs these two training data into the artificial intelligence model for training. When the training is completed, multiple weight values of the artificial intelligence model are extracted for translation so as to correspond to different customized parameters respectively, and the translation results are embedded into the state template of the update package to provide the client host 120 with download. At this point, the client host 120 can obtain the weight value of the customized artificial intelligence model. Afterwards, even if the artificial intelligence model is reset or initialized, the user can still upload the customized parameters and weight values to the server host 130 through the state template to update and set the artificial intelligence model. In this way, without retraining the artificial intelligence model, the output result of the artificial intelligence model will still be the same as the previously trained artificial intelligence model. It should be particularly noted that the state management module and the artificial intelligence model are implemented by a hardware processor executing computer-readable instructions. In actual implementation, the artificial intelligence model is trained with neural networks such as convolutional neural networks (CNN), recurrent neural networks (RNN), long short-term memory (LSTM), transformers or their similarities. In addition, back propagation algorithms and gradient descent methods can also be used for optimization during the training process.

As shown in "Figure 4", "Figure 4" is a schematic diagram of applying the present invention to set and update customized parameters through a status template. In actual implementation, after the client host 120 downloads the update package from the server host 130, the client host 120 can view the status template provided by the update package through the setting window 410. As shown in "Figure 4", the setting window 410 can display the customized parameters and their impact on the weight value. For example, using a bar graph, the longer the line, the greater the impact, and vice versa, the shorter the line, the smaller the impact. In this example, the customized parameter "A" has a greater impact on the weight value than the customized parameter "C", and the customized parameter "C" has a greater impact on the weight value than the customized parameter "B". That is, the user can choose to delete the customized parameter "B" that has the smallest impact on the output result of the artificial intelligence model through the filtering component 411. Assuming that the user deletes the customized parameter "B", the user can click the confirmation component 413 to save the setting. It should be particularly noted that in actual implementation, in addition to allowing the setting and updating of customized parameters, the state template can also allow the setting and updating of hyperparameters. The hyperparameters usually determine the architecture of the artificial intelligence model and various details of the training process, such as: learning rate, regularization parameter, tree depth, number of layers of the neural network, etc. The user can make settings through the input block 412, and after the settings are completed, the client host 120 can upload them to the server host 130, so that the server host 130 directly applies the weight values corresponding to the customized parameters, or even the hyperparameters, etc., to the artificial intelligence model according to the settings of the state template. In other words, the updated package can customize the artificial intelligence model into an artificial intelligence model dedicated to the client host 120 without training after setting, and because the generation of these weight values is affected by the user behavior status and customized parameters of the client host 120, for example: the first training data is affected by the user behavior status and customized parameters; the second training data itself is a customized parameter. Therefore, these weight values are exclusive to the client host 120, and the artificial intelligence model using these weight values will be very suitable for the client host 120. Moreover, the updated package setting is that the server host 130 extracts the weight values trained based on the user behavior status sensed by the client host 120 and the customized parameters sent previously, which can effectively reduce the computing load of the server host 130 for training the model.

In summary, the difference between the present invention and the prior art is that a rough question message with a natural language structure is generated through user behavior status and customized parameters, and then the rough question message is input into the logic circuit to generate a precise question message and sent to the artificial intelligence platform to obtain the corresponding answer message as the first training data, and the customized parameters are used as the second training data, and then the two training data are input into the artificial intelligence model for training. After the training is completed, the weight value is extracted for translation to correspond to different customized parameters respectively, and the translation result is provided to the client host. Download for customized settings and updates. In practice, the update package downloaded by the client host will be fed back to the user for customized settings based on the weight values trained on the server host. However, first-time users still need to rely on the learning/training process of the server host. Therefore, another practical significance of the update package is to extract the customized weight values after learning/training to reduce the computing power load of the server host's repeated learning. This technical means can solve the problems existing in previous technologies, thereby achieving the technical effect of improving the convenience and exclusivity of model customization.

Although the present invention is disclosed as above by the aforementioned embodiments, they are not used to limit the present invention. Anyone skilled in similar techniques can make some changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of patent protection of the present invention shall be subject to the scope of the patent application attached to this specification.

110: artificial intelligence platform 120: client host 121: sensor 122: first non-transient computer-readable storage medium 123: first hardware processor 130: server host 131: logic circuit 132: second non-transient computer-readable storage medium 133: second hardware processor 410: setting window 411: filter component 412: input block 413: determine component Step 210: connect a server host to an artificial intelligence platform and a client host respectively Step 220: The client host continuously senses at least one of physiological state, facial expression and body movement through at least one sensor to generate a user behavior state Step 230: The client host downloads an update package, wherein the update package provides a state template to allow setting and updating at least one customized parameter Step 240: The client host continuously transmits the user behavior state and the customized parameter to the server host Step 250: The server host generates a rough question message with a natural language structure according to the received user behavior state and the customized parameters, and inputs the rough question message into a first finite state machine and a second finite state machine connected in series for parsing and state conversion to generate a precise question message, wherein the first finite state machine receives the rough question message and an answer message from the artificial intelligence platform, the output of the first finite state machine is used as the input of the second finite state machine, and the second finite state machine outputs the precise question message to the artificial intelligence platform through an application programming interface (API) of the artificial intelligence platform Step 260: The artificial intelligence platform inputs the precise question message into a large language model (Large Language Model, LLM) to generate the answer message, and then transmit the answer message to the server host through the application program interface Step 270: The server host receives the answer message corresponding to the precise question message from the artificial intelligence platform as a first training data, and uses the received customized parameters as a second training data, and inputs the second training data and the first training data into an artificial intelligence model for training. After the training is completed, multiple weight values of the artificial intelligence model are extracted Step 280: The server host translates the weight values to correspond to different customized parameters, and embeds the translated results into the status template of the update package to provide the client host with download

Figure 1 is a system block diagram of the system for customized setting and updating downloading of active chat response of the present invention. Figure 2A and Figure 2B are method flow charts of the method for customized setting and updating downloading of active chat response of the present invention. Figure 3 is a schematic diagram of the client host, server host and artificial intelligence platform of the present invention. Figure 4 is a schematic diagram of the application of the present invention to set and update customized parameters through status templates.

110: Artificial Intelligence Platform

120: Client host

121:Sensor

122: First non-transitory computer-readable storage medium

123: First hardware processor

130: Server host

131:Logic circuit

132: Second non-transitory computer-readable storage medium

133: Second hardware processor

Claims (10)

A customized setting and update download system with active chat response, the system comprising: An artificial intelligence platform, for receiving a precise question message through an application programming interface (API), and inputting the precise question message into a large language model (LLM) to generate an answer message, and then transmitting the answer message through the application programming interface; A client host, the client host comprising: At least one sensor, for continuously sensing at least one of physiological state, facial expression and body movement to generate a user behavior state; A first non-transient computer-readable storage medium, for storing a plurality of first computer-readable instructions; and A first hardware processor electrically connected to the first non-transitory computer-readable storage medium and the sensor to execute the plurality of first computer-readable instructions, so that the client host downloads an update package and continuously transmits the user behavior state and at least one customized parameter, wherein the update package provides a state template to allow the customized parameter to be set and updated; and A server host connected to the client host to receive the user behavior state and the customized parameter, the server host comprising: A logic circuit, comprising a first finite state machine and a second finite state machine connected in series, wherein the first finite state machine receives a rough question message and the answer message, and the output of the first finite state machine is used as the input of the second finite state machine, so that the second finite state machine generates the precise question message and outputs it to the artificial intelligence platform through the application program interface; A second non-transient computer-readable storage medium, used to store multiple second computer-readable instructions; and A second hardware processor, electrically connected to the second non-transient computer-readable storage medium and the logic circuit, used to execute the multiple second computer-readable instructions, so that the server host executes: Generate a rough question message with a natural language structure according to the received user behavior state to input into the logic circuit; After the logic circuit generates the corresponding precise question message according to the rough question message and outputs it to the artificial intelligence platform, receive the answer message corresponding to the precise question message from the artificial intelligence platform as a first training data; Use the customized parameter as a second training data, and input the second training data and the first training data together into an artificial intelligence model for training. After the training is completed, extract multiple weight values of the artificial intelligence model; and The weight values are translated to correspond to different customized parameters respectively, and the translated results are embedded into the status template of the update package to provide the client host with download. A customized configuration and update download system with active chat response as claimed in claim 1, wherein the server host selects at least one of natural language processing (NLP), generative model and template matching to generate the rough question message with a natural language structure. As in claim 1, the customized setting and update download system with active chat response, wherein the first finite state machine and the second finite state machine parse keywords and grammatical structures according to the rough question message, and switch states according to the parsing results to determine the question type, and then use predefined templates or grammatical rules to generate the precise question message that is more specific and clear than the rough question message. A customized setting and update download system with active chat response as in claim 1, wherein the first finite state machine is a Mealy Machine, whose output is affected by the current state, the rough question message and the answer message, and the second finite state machine is a Moore Machine, whose output is only affected by the current state. A customized setting and update download system with active chat response as in claim 1, wherein the customized parameters include a filtering parameter and a time message, the filtering parameter is used to set the reply message agreed to be received and the reply message refused to be received, and the time message is used as a basis for determining whether the reply message is related to time. A customized setting and update download method with active chat response, the steps of which include: Connecting a server host to an artificial intelligence platform and a client host respectively; The client host continuously senses at least one of physiological state, facial expression and body movement through at least one sensor to generate a user behavior state; The client host downloads an update package, wherein the update package provides a state template to allow setting and updating at least one customized parameter; The client host continuously transmits the user behavior state and the customized parameter to the server host; The server host generates a rough question message with a natural language structure according to the received user behavior state and the customized parameters, and inputs the rough question message into a first finite state machine and a second finite state machine connected in series for parsing and state conversion to generate a precise question message, wherein the first finite state machine receives the rough question message and an answer message from the artificial intelligence platform, the output of the first finite state machine is used as the input of the second finite state machine, and the second finite state machine outputs the precise question message to the artificial intelligence platform through an application programming interface (API) of the artificial intelligence platform; The artificial intelligence platform inputs the precise question message into a large language model (LLM) to generate the answer message, and then transmits the answer message to the server host through the application programming interface; The server host receives the answer message corresponding to the precise question message from the artificial intelligence platform as a first training data, and uses the received customized parameters as a second training data, and inputs the second training data and the first training data into an artificial intelligence model for training. After the training is completed, multiple weight values of the artificial intelligence model are extracted; and the server host translates the weight values to correspond to different customized parameters respectively, and embeds the translated results into the status template of the update package to provide the client host with download. A customized configuration and update download method with active chat response as in claim 6, wherein the server host selects at least one of natural language processing (NLP), generative model and template matching to generate the rough question message having a natural language structure. A customized setting and update download method with active chat response as in claim 6, wherein the first finite state machine and the second finite state machine parse keywords and grammatical structures according to the rough question message, and switch states according to the parsing results to determine the question type, and then use predefined templates or grammatical rules to generate the precise question message that is more specific and clear than the rough question message. A customized setting and update download method with active chat response as in claim 6, wherein the first finite state machine is a Mealy Machine, whose output is affected by the current state, the rough question message and the answer message, and the second finite state machine is a Moore Machine, whose output is only affected by the current state. A customized setting and update download method for active chat response as in claim 6, wherein the customized parameters include a filtering parameter and a time message, the filtering parameter is used to set the reply message agreed to be received and the reply message refused to be received, and the time message is used as a basis for determining whether the reply message is related to time.

Images