TWI809527B - Method of Constructing PDCA Expert System Using Natural Language Data and Artificial Intelligence and Its Application Method of PDCA Expert System - Google Patents

Abstract

The present invention provides a method for constructing a PDCA expert system using natural language data together with artificial intelligence, and an application method for the PDCA expert system constructed by the method. The construction method uses a smart wearable device or a mobile device with a network communication connection state to transmit an expert's spoken voice stream through the aforementioned device to an artificial intelligence training system through the network and target the spoken voice stream And correspondingly generate natural language information, continue to use the expert's natural language information as machine learning materials, extract and define project feature quantities, execution feature quantities, check feature quantities, and improve feature quantities to group them correspondingly, and execute the aforementioned The process is maintained for a training period, and the PDCA expert system is constructed accordingly. In addition, using the PDCA expert system can also allow users to learn or follow instructions to achieve plan execution.

Description

Method of Constructing PDCA Expert System Using Natural Language Data and Artificial Intelligence and Its Application Method of PDCA Expert System

The present invention is related to the field of "construction and application of expert systems", especially a method for constructing a PDCA expert system using natural language data and artificial intelligence and an application method for the PDCA expert system.

An expert system is a program system that has the ability to solve problems with an expert level in a specific field or purpose. The system can effectively use the effective experience and expertise accumulated by experts over the years, and solve various problems that need to be solved by experts by simulating the thinking process of experts. In view of the fact that experts usually need to have expertise in this field when solving problems, Therefore, the expert system is also called the knowledge base system. In other words, an expert system is to store expert knowledge in a computer-readable form and add it to the control strategy, so that the computer can use this knowledge to solve problems.

There is a positive correlation between the sufficiency of knowledge base construction and the accuracy of problem-solving methods. However, it is not easy to build a complete knowledge base and there are often deficiencies. Therefore, the situation where expert systems perform better is usually just suitable for a specific specialty. In this way, the content of the knowledge base can be as sufficient as possible in this narrow field, so as to obtain better judgment and ability to solve complex problems. A mature expert system is also an ideal tool for education and training. Users can gain corresponding abilities after using the expert system through successive teachings.

On the other hand, the fields contained in the real society are endless, and there are many people who have invested a lot of relevant research and education and training in hot areas, but there is usually little interest or experience in solving problems in special fields or purposes Under the stereotype that there is no need to build an expert system, the precious knowledge inheritance may be lost with the death of experts in the future. Especially for experts in certain special fields, the meaning of knowledge includes the experience accumulation of these experts, that is, these experts often add their own rules of thumb to solve problems when facing problems. The so-called rule of thumb can also be understood as the experience rate for experts to choose problem-solving means after considering various variables when facing a problem. Therefore, the expert system can also formulate the rules of thumb of experts to solve problems. Accordingly, if the knowledge of experts in special fields in human society and their implicit experience accumulation can be stored, this type of expert system will be used as a permanent system resource for effectively inheriting and solving professional problems.

Although the application value of expert systems in various fields is extremely high, how to build a perfect expert system is still difficult, especially how to store enough expert knowledge and turn it into usable resources is the primary issue. In particular, expert knowledge in special and rare fields often has the characteristics of non-data, non-documentation, and lack of experimental analysis data. Therefore, it is very difficult to have sufficient resources to build an expert system under this premise. Not to mention, due to the ever-changing complexity of human problems, even with the aforementioned various documents recording knowledge, the cumulative content of the empirical knowledge base is still often difficult to be sufficient to solve the problem. In other words, in order to impart knowledge to the computer, the expert system must first obtain knowledge from experts. Once the amount of knowledge increases and the rules become more numerous, it will become difficult to properly maintain and manage knowledge. Moreover, for the frame problem (frame problem), that is, when using artificial intelligence to perform a certain task, how to extract the knowledge related to the task and apply it is still difficult; Symbols and the symbol grounding problem linked together with the meaning represented by the symbol are unavoidable difficulties in the construction of expert systems. Generally speaking, after the machine has acquired a large amount of data, how to further find out the feature quantity with classification significance and "harden" it is a topic that machine learning experts want to improve. Especially in the face of various types of data in different fields and different attributes, how to combine the experts in this field to jointly explore the characteristic quantities with characteristic expression meanings as self-exploration, and strengthen them to learn and build the model is still a difficult task.

In view of this, the inventor of the present invention considers the shortcomings and deficiencies of the above-mentioned various prior technologies, and then conceives a method through the present invention, which can accumulate expert knowledge and experience, and continuously transform it into knowledge base content in life, and build a Set up the big data database under the field or purpose, and provide it as an expert system construction material. On the other hand, the content of these big data is disorganized, such as which of these data is an effective means, which is a problem, which is an additional factor and other uncertainties. Therefore, considering the optimal flow of the operation model of the expert system, the present invention uniquely uses PDCA (Plan-Do-Check-Act) cyclical quality management as the feature quantity classification standard. This covers applications in various fields, so there is a follow-up relationship between the classified data. The so-called PDCA is to carry out activities according to planning, execution, inspection and action for quality work, so as to ensure the achievement of reliability goals, and then promote continuous improvement of quality, and construct an expert system for logical problem solving.

The purpose of the present invention is to improve the difficulty in collecting effective expert data or the insufficient amount of data when the conventional expert system is constructed. Especially for the inheritance of expertise in special fields, expert systems should be relied on to assist the continuation of expertise. Therefore, how to build a perfect expert system and allow users to use the expert system to be taught and learned sequentially is the focus of the present invention.

In order to achieve the above purpose, the present invention proposes a method for constructing a PDCA expert system using natural language data together with artificial intelligence. The voice stream is transmitted to an artificial intelligence training system through the smart wearable device or mobile device through the network, and corresponding natural language information is generated for the spoken voice stream, followed by the natural language information input by the expert. Semantics, extract and define planning features, execution features, inspection features, and improvement features and group them accordingly, and execute the aforementioned process and maintain a training period to construct a PDCA expert system. In this way, the construction data of the expert system can be effectively collected, and a better feature quantity classification basis for artificial intelligence in the construction process can be proposed.

Furthermore, in order to effectively convert highly complex natural language into machine learning materials, in one embodiment, the artificial intelligence training system uses preset project feature words as the starting point for extracting natural language information Points, and define the grouping of continuously input natural language information into the project feature quantity; according to the preset execution feature words, as the starting point of natural language information extraction, and define the continuous input of natural language information grouping into the execution feature quantity; According to the pre-set check feature words, as the starting point of natural language information extraction, and define the continuous input of natural language information into groups for inspection features; according to the pre-set improvement feature words, as the starting point of natural language information extraction points, and define continuous input of natural language information grouping to improve feature quantity. In this way, the training materials can be logically grouped according to the characteristic quantity for the natural language information with variable nature.

Furthermore, in order to obtain appropriate machine learning materials, among them, when the artificial intelligence training system collects the feature words of the trigger plan, executes the feature words, checks the feature words and improves the feature words, in the improved feature words After the word is triggered, the execution of the characteristic word is not triggered after a set time, then the training period ends. Or, when the artificial intelligence training system collects triggering project feature words, execution feature words, check feature words and improvement feature words, when the similarity between the improved feature quantity and the project feature quantity exceeds a threshold , the training period ends. Or, when the artificial intelligence training system does not trigger the planned feature words, execute the feature words, check the feature words and improve the feature words in sequence, the training period will be terminated at any time, and the natural language information collected and stored Neither are used as training material.

After collecting a large amount of natural language information, on the other hand, deep learning can also be used as a grouping method. In one embodiment, the feature quantity of the project, the execution feature quantity, the inspection feature quantity and the improvement feature quantity are the expert's Natural language information codes are used as machine learning materials, and then automatically grouped by the artificial intelligence training system through deep learning of more than three layers of neural networks.

Inheritance from previous embodiments, wherein, when constructing and storing more than two PDCA expert systems, if the project feature quantity judgment of the two or more PDCA expert systems has an attribute correlation, then define that these PDCA expert systems are all The subsystems will be connected in series and merged with each other to generate a parent system with a higher purpose; the parent system includes all the integrated subsystems.

Based on the above-mentioned same purpose, the present invention continues to propose a kind of application method of the PDCA expert system that utilizes the above-mentioned method to construct, and is operated accordingly for a user, it is characterized in that: this PDCA expert system corresponds to the user to generate a According to the target plan, the user converts the voice stream of the user's spoken language into natural language through the smart wearable device or mobile device with network transmission function according to the target plan and inputs it into the PDCA expert system to judge the target The implementation level of the plan, and then the PDCA expert system conducts a check procedure and generates a corresponding improvement plan after analysis, and then feeds back to the user for execution. In this way, users can follow the PDCA expert system to effectively guide, educate, train, execute or supervise until the expected target plan is achieved.

In order to facilitate the implementation of the PDCA expert system, in one embodiment, the user continuously converts the spoken voice stream into natural language information and inputs it into the PDCA expert system, repeatedly executes the inspection procedure and proposes the improvement plan until This goal is planned to be completed. In this way, the user also collects data in the same way as when building the PDCA expert system to facilitate the execution level check, so that the user can be effectively supervised, and then play the terminal guidance function of the PDCA expert system.

Based on the previous embodiment, in order to enhance the accuracy of semantic recognition, the semantic recognition of the natural language information is supplemented with information content stored in a cloud database in the corresponding field for comparison and analysis, and the cloud database further stores the used The natural language information input by the user, classify the features and create corresponding labels for the natural language information, and then record the personalization factors belonging to the user. In this way, the PDCA expert system can guide, educate, train, execute or supervise multiple people at the same time.

In summary, the present invention is a method of constructing PDCA expert system using natural language data and artificial intelligence engine. This method can accumulate expert knowledge and experience, and continuously transform it into knowledge base content in daily life. This is to build a big data database under the field or purpose, and provide it as an expert system construction material. In addition, considering the optimal process of the operation model of the expert system, the present invention uniquely uses PDCA (Plan-Do-Check-Act) cycle quality management as the characteristic quantity classification standard of artificial intelligence, and then plans, executes, Check and act to carry out activities to ensure the achievement of reliability goals and promote continuous quality improvement. Therefore, the expert system construction data can be effectively collected by using the present invention, and the PDCA expert system established accordingly can be used as the best training and learning tool for the inheritance of expertise in various fields.

In order to enable those skilled in the art to clearly understand the content of the present invention, the following descriptions are provided together with the drawings for your reference.

The present invention is essentially a part of machine learning, but the basic technology of machine learning is not the focus of this case. To be precise, the present invention proposes a solution for how to effectively obtain sufficient knowledge for constructing an expert system, and how to effectively classify a large amount of data (that is, knowledge) after obtaining a technical means. Please also refer to Figures 1-3, which are respectively a schematic diagram of the construction process of the expert system, a schematic diagram of the grouping of natural language information features, and a schematic diagram of the construction method of the expert system in a preferred embodiment of the present invention. As can be seen from the figure, the present invention utilizes the actual experts 1 and the artificial intelligence training system 3 , complementing the management concept of PDCA as the characteristic quantity classification standard, and constructs the corresponding expert system accordingly. Specifically, a method of using natural language information 5 and artificial intelligence to construct a PDCA expert system 4 according to the present invention is characterized in that, using a smart wearable device 2 or a mobile device 9 with a network communication connection status, an expert 1 The spoken voice stream is transmitted to an artificial intelligence training system 3 through the smart wearable device or mobile device 9 through the network, and the corresponding natural language information 5 is generated for the spoken voice stream, according to the expert 1 The semantics of the input natural language information 5, extract and define the project feature, execute the feature, check the feature and improve the feature and group them correspondingly, and execute the aforementioned process and maintain a training period, and construct accordingly PDCA expert system4. The data acquisition method of the present invention is done through the most natural and easiest way for the expert 1 to implement, that is, the expert 1 only needs to wear the smart wearable device 2 or the mobile device 9 held by the expert 1. When the system is built, the oral content is used to further convert it into usable training materials. In this way, the system can overcome the problem of insufficient amount of traditional training data. On the other hand, because in the past, machine learning was often unable to effectively extract the correct data during the classification stage, resulting in the computer being unable to correctly classify as the corresponding correct execution output. Therefore, after obtaining sufficient amount of data, further use the unique classification rules proposed by the present invention in the field of machine learning, that is, divide all the data into four categories by planning feature quantities, executing feature quantities, checking feature quantities, and improving feature quantities. Large types, and then according to the logical relationship between the execution order of the four major characteristics, can be applied to any expert system construction model, which is the PDCA expert system defined in the present invention.

Furthermore, since the feature quantity is the input variable of machine learning, and its value can quantitatively present the characteristics of the target, that is to say, the accuracy of machine learning will vary with the selected feature quantity. Therefore, the present invention utilizes the time series evolution characteristic of PDCA, and can prompt the expert 1 to use certain verbal rules for the artificial intelligence training system 3 to collect training data more accurately and conveniently. Please refer to Figure 2 again, where the artificial intelligence training system 3 uses the preset project feature words Pc as the starting point for extracting natural language information, and defines the continuously input natural language information to be grouped into project features The amount of continuously input data during this period of time is represented by Lt1; then, according to the preset execution feature word Dc, it is used as the starting point for natural language information extraction, and the grouping of the continuously input natural language information is defined as the execution feature amount , the data continuously input during this period is represented by Lt2; then, according to the pre-set check feature word Cc, as the starting point of natural language information extraction, and define the continuous input of natural language information into groups of check features , the data continuously input during this period is represented by Lt3; finally, according to the preset improved feature word Ac, it is used as the starting point of natural language information extraction, and the grouping of the continuously input natural language information is defined to improve the feature value, The data continuously input during this period is represented by Lt4. In terms of the characteristics of spoken language, there is a logical sequence for planning, execution, inspection, and improvement. Therefore, using the preset characteristic words can effectively trigger the machine as an appropriate classification. In this way, even for the extremely complex human language of the machine, the classification is relatively simple and more correct. At the same time, this method also solves the problem of frame and symbol grounding. Afterwards, the machine only needs to use the existing semantic recognition tools to regenerate appropriate output.

Continuing, in order to effectively control whether the learning data of the machine is sufficient to construct the PDCA expert system, the present invention further proposes the following several solutions. After executing the characteristic words, checking the characteristic words and improving the characteristic words, if the improved characteristic words are triggered and the execution of the characteristic words is not triggered after a set time, then the training period ends. In terms of its intention, it is mainly considered that after the improvement of the characteristic words is triggered, it is no longer necessary for execution. At this time, it is determined that the current situation has reached the expected target state. In other words, there should be no need to execute at this time, that is, to trigger the execution of the characteristic word. Therefore, when the set time has elapsed without triggering the execution of the characteristic word again, it means that the goal should have been achieved, and the amount of data learned by the machine is sufficient to end the training period.

In another embodiment, when the artificial intelligence training system 3 collects triggering project feature words, execution feature words, inspection feature words and improvement feature words, the improvement feature quantity is similar to the project feature quantity When the degree exceeds a threshold value, the training period ends. Empirically, when the meaning of the improved characteristic quantity and the spoken language corresponding information of the project characteristic quantity tend to be consistent, it is essentially extremely high probability that the degree of improvement has reached the target level. In other words, at this point the goal should have been achieved and the amount of data for machine learning is sufficient to end the training period. In addition, the threshold value can be judged by using various correlation existing formulas in statistics, so it will not be repeated here.

Continuing, in yet another embodiment, in order to avoid collecting wrong semantic correspondence data, and increase the burden of machine learning, wherein, when the artificial intelligence training system 3 does not sequentially trigger the project feature word, execute the feature word, check Feature words and improved feature words means that the training period will be terminated at any time, and the collected and stored natural language information will not be used as training materials. Although in experience, a better system can only be trained with more learning data, but miscognized data will instead cause obstacles and burdens to machine learning, that is, miscognized feature quantities will lead to incorrect classification, which will greatly affect the machine after calculation. The correct answer is output. Therefore, continuing the concept of logical sequence of PDCA, if there is an unreasonable data collection sequence, it should be a better practice to discard the data in principle.

In fact, the method of defining feature quantities in advance in the present invention has greatly reduced the burden of machine learning, so at this time, the machine can also use these feature quantities to classify these data by itself. Please refer to Fig. 3 again. In one embodiment, after a large amount of learning data is collected, the feature quantity of planning, execution feature quantity, inspection feature quantity and improvement feature quantity are layered through the artificial intelligence training system 3 Grouping after neural network-like deep learning; or grouping through decision tree by the artificial intelligence training system. Among them, the hierarchical neural network is a neural network with more than three layers, and an automatic encoder is used to set the collected data as the same input and output, and to further give meaning to the nodes of the multi-layer neural network. Extract the data grouping of the above-mentioned planning characteristic quantity, execution characteristic quantity, inspection characteristic quantity and improvement characteristic quantity.

Please also refer to FIG. 4, which is a schematic diagram of an expert system application in a preferred embodiment of the present invention. Furthermore, in view of the high correlation between the abundance of knowledge and the accuracy of the expert system, it is better to build the scale of the expert system from small to large in theory. At this time, when constructing and storing more than two PDCA expert systems 4, if the project feature quantities of the two or more PDCA expert systems 4 are judged to have an attribute correlation, then these PDCA expert systems are defined as The subsystems 6 will be connected in series with each other to form a parent system 7 with a higher purpose; the parent system 7 will include all the integrated subsystems 6 . For example, the PDCA expert system 4 is a Naturopathy Doctor expert system, a dietary advice expert system, an exercise training expert system, a sleep management expert system, or a decompression assistance expert system. Essentially, these PDCA expert systems 4 are constructed for different issues by using different expert knowledge and experiences. However, these PDCA expert systems 4 all have the attribute association of "health", so these natural doctor expert systems, dietary advice expert systems, exercise training expert systems, sleep management expert systems, or decompression assistance expert systems can be further combined The systems are all defined as subsystems 6, and then integrated by the parent system 7 of the upper "health" purpose. When it is subsequently used by others, if the parent system 7 is executed for the upper "health" purpose, these subsystems 6 will also be connected in series at the same time for separate execution. Also, considering the continuity of data collection of the PDCA expert system 4 of the present invention, if the content of its knowledge base is expanded or new solutions are added, and then the original PDCA expert system 4 is amended, different experts can be defined according to the time stamp of the training period system version. In this way, the PDCA expert system 4 built in different periods can be distinguished.

Please refer to Figure 4 again. Taking the health field as an example, the computer-aided system in the past can only provide corresponding improvement measures for the user 8 to implement accordingly, but there is no follow-up control method for whether the user 8 has implemented the plan content as scheduled . In other words, no matter how good the plan is, if it cannot be implemented accordingly, its terminal expected result will still be difficult to achieve. Based on this, the present invention is completely different from the prior art, and the focus of execution is more focused on effectively supervising the user 8 to complete the project execution, that is, after these PDCA expert systems 4 are set up, they can be used by the user 8, it is characterized in that: the PDCA expert system 4 generates a target plan corresponding to the user 8, and the user 8 uses the smart wearable device 2 or mobile device 9 with network transmission function according to the target plan Convert the voice stream of the spoken language of the user 8 into natural language and input it to the PDCA expert system 4 to judge the execution degree of the target plan, and then the PDCA expert system 4 performs a check procedure and generates a corresponding result after analysis An improvement plan is fed back to the user 8 for execution. Furthermore, the user 8 can continuously convert the spoken voice stream into natural language and input it to the PDCA expert system 4, and then repeatedly execute the checking procedure and propose the improvement plan until the target plan is completed. In this way, the PDCA expert system 4 built by the present invention is completely different from the conventional expert system which only operates to the extent of providing the initial plan, and can effectively supervise the user when the user 8 is executed. 8 Execute the target plan, and even if the target plan cannot be implemented as scheduled, an improvement plan can be generated to adjust the measures that the final result can still reach the initial target plan.

In addition, in constructing the environmental conditions, in order to enhance the accuracy of semantic recognition, the semantic recognition of the natural language information is supplemented with the information content stored in a cloud database 10 in the corresponding field for comparison and analysis. The cloud database 10 can be a non-associated database, and can record various data materials with different attributes. And the cloud database 10 further stores the natural language input by the user, and classifies the features and establishes corresponding labels for these natural languages, and then records the personalization factors belonging to the user. In this way, the PDCA expert system 4 can guide, educate, train, execute or supervise multiple people at the same time.

In summary, the present invention is a method of constructing PDCA expert system using natural language data and artificial intelligence engine. This method can accumulate expert knowledge and experience, and continuously transform it into knowledge base content in daily life. This is to build a big data database under the field or purpose, and provide it as an expert system construction material. In addition, considering the optimal process of the operation model of the expert system, the present invention uniquely uses PDCA (Plan-Do-Check-Act) cycle quality management as the characteristic quantity classification standard of artificial intelligence, and then plans, executes, Check and act to carry out activities to ensure the achievement of reliability goals and promote continuous quality improvement. Therefore, the expert system construction data can be effectively collected by using the present invention, and the PDCA expert system established accordingly can be used as the best training and learning tool for the inheritance of expertise in various fields.

The above is only a description of the preferred embodiment in the patent scope of the present invention, but not to limit the scope of rights of the present invention based on the content of this embodiment; therefore, the textual changes made without departing from the equivalent scope of the present invention or modifications, should still be covered within the scope of the patent application of the present invention.

1: Expert 2: Smart wearable device 3: Artificial intelligence training system 4: PDCA expert system 5: Natural language information 6: Subsystem 7: Mother system 8: User 9:Mobile device 10:Cloud database

Figure 1 is a schematic diagram of the construction flow of the expert system of the preferred embodiment of the present invention. Fig. 2 is a schematic diagram of natural language information feature quantity grouping in a preferred embodiment of the present invention. Fig. 3 is a schematic diagram of an expert system construction method in a preferred embodiment of the present invention. Figure 4 is a schematic diagram of an expert system application in a preferred embodiment of the present invention.

1: Expert

2: Smart wearable device

3: Artificial intelligence training system

4: PDCA expert system

5: Natural language information

9:Mobile device

Claims (10)

A method of using natural language information and artificial intelligence to construct a PDCA expert system, characterized in that: using a smart wearable device or a mobile device with a network communication connection state, an expert's oral voice stream is passed through the smart wearable device device or mobile device, the network transmits to an artificial intelligence training system and generates corresponding natural language information for the spoken voice stream, and then extracts and defines the plan according to the semantic meaning of the natural language information input by the expert Feature quantity, execute feature quantity, check feature quantity and improve feature quantity to group them correspondingly, execute the aforementioned steps and maintain a training period, and construct the PDCA expert system accordingly. The method for constructing a PDCA expert system as described in Claim 1, wherein, the artificial intelligence training system uses the preset project feature words as the starting point for extracting project feature quantities of natural language information, and defines The continuously input natural language information is grouped into the project feature quantity; according to the pre-set execution feature words, it is used as the starting point for extracting the execution feature quantity of the natural language information, and the continuously input natural language information is grouped into the execution feature quantity ;According to the pre-set check feature words, as the starting point of the check feature quantity extraction of natural language information, and define the continuous input of natural language information grouping to check feature quantities; according to the preset improvement feature words, as the object The improvement feature quantity of natural language information extracts the starting point, and defines the continuous input natural language information grouping to improve the feature quantity. The method for constructing a PDCA expert system as described in claim 2, wherein, when the artificial intelligence training system collects the triggering project characteristic words, executes the characteristic words, checks the characteristic words and improves the characteristic words, after the improvement If the characteristic words are not triggered and executed within a set period of time after the characteristic words are triggered, the training period ends. The method for constructing a PDCA expert system as described in claim item 2, wherein, after the artificial intelligence training system collects trigger project feature words, executes feature words, checks feature words and improves feature words, the improvement feature When the similarity between the quantity and the feature quantity of the plan exceeds a threshold value, the training period ends. The method for constructing a PDCA expert system as described in claim 2, wherein, when the artificial intelligence training system does not trigger the planned feature words, execute the feature words, check the feature words and improve the feature words in sequence, it will be terminated at any time The training period and the collected and stored natural language information are not used as training materials. The method for constructing a PDCA expert system as described in claim item 1, wherein, the feature quantity of planning, execution feature quantity, inspection feature quantity and improvement feature quantity are the natural language information encoding of the expert as machine learning material, and then by The artificial intelligence training system is automatically grouped after deep learning of more than three layers of neural networks. The method for constructing a PDCA expert system as described in any one of claims 1 to 6, wherein, when constructing and storing more than two PDCA expert systems, if the project feature quantity judgment of the two or more PDCA expert systems If there is an attribute correlation, these PDCA expert systems are all defined as subsystems and will be connected and integrated with each other to generate a parent system with a higher purpose; the parent system includes all the integrated subsystems. An application method of the PDCA expert system constructed by the method described in claim 7, for a user to operate, it is characterized in that: the PDCA expert system generates a target plan corresponding to the user, and the user generates a target plan according to the target plan Through the smart wearable device or mobile device with network transmission function, the voice stream of the user's spoken language is converted into natural language and input to the PDCA expert system to judge the implementation degree of the target plan, and then the PDCA expert system Perform a check procedure and generate a corresponding improvement plan after analysis, and then give feedback to the user for execution. The application method of the PDCA expert system as described in Claim 8, wherein the user continuously converts the spoken voice stream into natural language information and inputs it into the PDCA expert system, repeatedly executes the inspection procedure and proposes the improvement plan until This goal is planned to be completed. The application method of the PDCA expert system as described in claim item 9, wherein the semantic recognition of the natural language information is supplemented with information content stored in a cloud database in the corresponding field for comparison and analysis, and the cloud database is further stored The natural language information input by the user, classify the features and create corresponding labels for the natural language information, and then record the personalization factors belonging to the user.