TW201947428A - Method and system for automated learning of a virtual assistant - Google Patents

Abstract

A method for automated learning of a virtual assistant is disclosed herein. The method includes the following operations: receiving an audio and identifying the audio to generate a corpus data; utilizing a natural language processing model to analyze the corpus data, and to generate a language characteristic data corresponding to the corpus data; performing a scenario analysis for the language characteristic data according to a situation information confirm an operation corresponding to one of the intentions; if the scenario analysis could not determine the operation corresponding to one of the intentions, performing a segmentation processing for the corpus data; determining whether there is new vocabulary or new corpus data according to results of the segmentation processing; if there is new vocabulary, updating the natural language processing model according to meaning of new vocabulary; if there is new corpus data, updating the scenario analysis according to intention of new corpus data.

Description

   Automatic learning method and system of virtual assistant   

This case relates to an automatic learning method and system, and in particular to an automatic learning method and system of a virtual assistant.

Enterprise resource planning system (Enterprise Resource Planning, ERP), referred to as ERP system, refers to a management platform that provides decision-making for the enterprise decision-making layer based on information technology. Its main purpose is to uniformly manage the flow of people, logistics, information, and funds in order to maximize the use of corporate resources. The ERP system contains three major functions, such as production control, logistics management, and financial management. Therefore, the scale of the ERP system is very large.

Applying the virtual assistant to the ERP system can also help users quickly communicate with the huge ERP system, which can save users the time spent in using the ERP system, but because each user has different habits in using the ERP system, Therefore, there may be situations where the virtual assistant cannot understand the user's problem, which will cause the user's difficulty in using the ERP system.

The main object of the present invention is to provide an automatic learning method and system for a virtual assistant, which is mainly to allow the virtual assistant to have an automatic learning function, so that the virtual assistant can automatically learn the user's speech during the process of communicating with the user. Habits, or special terminology in the industry, enable users to use the ERP system more quickly and conveniently.

In order to achieve the above purpose, the first aspect of the present case is to provide an automatic learning method for a virtual assistant. This method includes the following steps: receiving audio input and identifying the audio to form a corpus data; using a natural language processing model to analyze the corpus data, In order to generate linguistic feature information corresponding to the corpus data, the linguistic feature information includes a plurality of intents, the probability corresponding to the intents, and a plurality of vocabularies; a functional context analysis is performed on the linguistic feature information based on the functional context information to determine the intent of the intents. One of them corresponds to the operation; if the functional context analysis cannot determine the operation corresponding to one of these intentions, the word segmentation processing is performed on the corpus data; according to the result of the word segmentation processing, it is determined whether there is a new vocabulary or new corpus data; If there is a new vocabulary, the natural language processing model is updated according to the meaning of the new vocabulary. If there is a new corpus data, the functional context analysis is updated according to the intention of the new corpus data. Among them, the operation includes one of querying data and performing instruction operations.

The second aspect of the case is to provide an automatic learning system for virtual assistants, which is connected to the enterprise database and the enterprise resource system, respectively, and includes: a processor, a storage device, and an input / output device. The storage device is electrically connected to the processor, and is used for storing a general database, an application knowledge database, a domain knowledge database, and a historical database. The input / output device is electrically connected to the processor to provide an interface for inputting audio. The processor includes a speech recognition module, a corpus analysis module, a situation recognition module, an unknown corpus determination module, and an update information module. The speech recognition module is used to recognize audio to form corpus data. The corpus analysis module and the speech recognition module are electrically connected to analyze the corpus data using a natural language processing model to generate linguistic feature information corresponding to the corpus data, where the linguistic feature information includes a plurality of intents, the intents Corresponding probability and plural words. The context recognition module and the corpus analysis module are electrically connected to perform a functional context analysis on the linguistic feature information according to the functional context information to determine an operation corresponding to one of the intentions. The unknown corpus judgment module is electrically connected to the situation recognition module, and is used to perform word segmentation processing on the corpus data when the situation recognition module cannot recognize the operation corresponding to one of these intentions, and according to the word segmentation processing, As a result, it is determined whether there is a new vocabulary or a new corpus. The update information module and the unknown corpus judgment module are electrically connected to update the natural language processing model according to the meaning of the new vocabulary when a new vocabulary is generated, and according to the new corpus when the new corpus data is generated The intent of the data updates the functional context analysis; wherein the operation includes one of a data query operation and an execution instruction operation.

The automatic learning method of the virtual assistant and the automatic learning system of the virtual assistant of the present invention mainly provide the virtual assistant with an automatic learning function, so that the virtual assistant can automatically learn the user's speaking habits during the process of communicating with the user, or It is a special terminology in the industry to achieve the effect of enabling users to use the ERP system more quickly and conveniently.

100‧‧‧ Automatic learning system for virtual assistants

110‧‧‧ processor

130‧‧‧Storage device

150‧‧‧ input / output device

131‧‧‧Overall Database

132‧‧‧Application Knowledge Database

133‧‧‧ Domain Knowledge Database

134‧‧‧History database

111‧‧‧Speech recognition module

112‧‧‧ Corpus Analysis Module

113‧‧‧Situation recognition module

114‧‧‧Unknown corpus judgment module

115‧‧‧Update Information Module

121‧‧‧ Training Module

122‧‧‧Template Creation Module

123‧‧‧ Semantic model building module

124‧‧‧vocabulary model building module

125‧‧‧Scenario Training Module

126‧‧‧ Vocabulary Training Module

300‧‧‧ Automatic learning method of virtual assistant

S310 ~ S360, S410 ~ S480, S321 ~ S323, S331 ~ S332, S341 ~ S342, S361 ~ S363‧‧‧Steps

In order to make the above and other objects, features, advantages, and embodiments of the present invention more comprehensible, the description of the drawings is as follows: FIG. 1 is an automatic learning of a virtual assistant according to some embodiments of the present invention A schematic diagram of the system; FIG. 2 is a schematic diagram of a processor according to some embodiments of the present case; FIG. 3 is a flowchart of an automatic learning method of a virtual assistant according to some embodiments of the present case; FIG. 5 is a flowchart of a training data model according to some embodiments of the present case; FIG. 5 is a flowchart of step S320 according to some embodiments of the present case; and FIG. 6 is a flowchart according to some embodiments of the present case. FIG. 7 is a flowchart of step S330; FIG. 7 is a flowchart of step S340 according to some embodiments of the present case; and FIG. 8 is a flowchart of step S360 according to some embodiments of the present case.

The following disclosure provides many different embodiments or illustrations to implement different features of the invention. The elements and configurations in the particular example are used in the following discussion to simplify the present disclosure. Any illustrations discussed are for illustrative purposes only and do not in any way limit the scope and meaning of the invention or its illustrations. In addition, the present disclosure may repeatedly refer to numerical symbols and / or letters in different examples, and these repetitions are for simplification and explanation, and do not themselves specify the relationship between different embodiments and / or configurations in the following discussion.

The terms used throughout the specification and the scope of patent applications, unless otherwise specified, usually have the ordinary meaning of each term used in this field, in the content disclosed here, and in special content. Certain terms used to describe this disclosure are discussed below or elsewhere in this specification to provide additional guidance to those skilled in the art on the description of this disclosure.

As used herein, "coupling" or "connection" can mean that two or more components make direct physical or electrical contact with each other, or indirectly make physical or electrical contact with each other, and "coupling" or " "Connected" may also mean that two or more elements operate or act on each other.

In this article, the terms first, second, third, etc. are used to describe various elements, components, regions, layers, and / or blocks that are understandable. However, these elements, components, regions, layers and / or blocks should not be limited by these terms. These terms are limited to identifying single elements, components, regions, layers, and / or blocks. Therefore, a first element, component, region, layer, and / or block in the following may also be referred to as a second element, component, region, layer, and / or block without departing from the intention of the present invention. As used herein, the term "and / or" includes any combination of one or more of the associated listed items. The "and / or" mentioned in this document refers to any, all or any combination of at least one of the listed elements.

See Figure 1. FIG. 1 is a schematic diagram of an automatic learning system 100 for a virtual assistant according to some embodiments of the present invention. As shown in FIG. 1, the automatic learning system 100 of the virtual assistant includes a processor 110, a storage device 130, and an input / output device 150. The storage device 130 is configured to store a general database 131, an application knowledge database 132, a domain knowledge database 133, and a historical database 134, and a general database 131, an application knowledge database 132, a domain knowledge database 133, and a historical database 134. Electrically connected to the processor 110. The input / output device 150 is electrically connected to the processor 110 for providing an interface for inputting audio. In one embodiment, the input / output device 150 may be a keyboard, a touch screen, a microphone, a speaker, or other suitable input / output devices. The user can input audio through the interface provided by the input / output device.

In various embodiments of the present invention, the processor 110 may be implemented as an integrated circuit such as a microcontroller, a microprocessor, a digital signal processor, and an application specific integrated circuit. integrated circuit (ASIC), logic circuit, or other similar components or a combination of the above. The storage device 150 may be implemented as a memory, a hard disk, a flash drive, a memory card, and the like.

Please refer to FIG. 2. FIG. 2 is a schematic diagram of a processor 110 according to some embodiments of the present invention. The processor 110 includes a speech recognition module 111, a corpus analysis module 112, a situation recognition module 113, an unknown corpus judgment module 114, an update information module 115, a training module 121, a template creation module 122, and a semantic model A building module 123, a vocabulary model building module 124, a situation training module 125, and a vocabulary training module 126. Corpus analysis module 112 is electrically connected with speech recognition module 111, context recognition module 113 is electrically connected with corpus analysis module 112, unknown corpus judgment module 114 is electrically connected with context judgment module 113, and updated The information module 115 is electrically connected to the unknown corpus judgment module 114. The training module 121 is electrically connected to the corpus analysis module 112. The template creation module 122 is electrically connected to the training module 121. The semantic model creation module 123 and the vocabulary model creation module 124 and the template creation module 122 are electrically connected. Connected, the situation training module 125 is electrically connected with the situation recognition module 113, and the unknown corpus judgment module 114 is electrically connected with the vocabulary training module 126.

Please refer to Figures 1 to 3 together. FIG. 3 is a flowchart of an automatic learning method 300 for a virtual assistant according to some embodiments of the present invention. As shown in FIG. 3, the automatic learning method 300 of the virtual assistant includes the following steps: Step S310: Receive audio input and identify the audio to form corpus data; Step S320: Use natural language processing model to analyze the corpus data to generate and language Step S330: Perform a functional situation analysis on the linguistic feature information according to the functional situation information to determine the operation corresponding to one of the intentions; Step S340: If the functional situation analysis cannot determine which of the intentions One of the corresponding operations, word segmentation processing is performed on the corpus data; step S350: judging whether a new vocabulary or new corpus data exists according to the result of the word segmentation processing; and step S360: if a new vocabulary exists, update according to the meaning of the new vocabulary Natural language processing model, if there is new corpus data, the functional situation analysis is updated according to the intent of the new corpus data.

In step S310, audio input is received and audio is recognized to form corpus data. In one embodiment, the audio received through the input / output device 150 can be recognized by the speech recognition module 111 of the processor 110 to convert the user's natural language into corpus data. In another embodiment, speech recognition can also send audio to the cloud speech recognition system through the Internet. After the audio is recognized by the cloud speech recognition system, the recognition result is used as corpus data. For example, cloud speech recognition The system can be implemented as Google's speech recognition system.

Before executing step S320, a common vocabulary model and a common semantic model need to be established. Therefore, please refer to FIG. 4, which is a flowchart of a training data model according to some embodiments of the present invention. As shown in Figure 4, the training data model stage includes the following steps: Step S410: Generate a system domain vocabulary based on the application knowledge database and domain knowledge database; Step S420: The system domain vocabulary and a plurality of service application parameters are formed as the key Entity set; step S430: classify a plurality of training corpora into one of query data operation and execute instruction operation; step S440: distinguish the training corpus corresponding to query data operation according to the category in the enterprise database The plurality of query data operation intentions and the intention to distinguish the training corpus corresponding to the execution of the instruction operation according to the service behavior provided by the enterprise resource system form a plurality of execution instruction operation intentions; step S450: establishing a template of the query data operation intention and the execution The template of the instruction operation intention; Step S460: Establish an overall database according to the key entity set, the query data operation intention template, and the template for executing the instruction operation intention; Step S470: identify the system domain words in the key entity set appear in the training corpus Plural first Rate, and analyze the plurality of sentence structure structures of the training corpus through the identified system domain vocabulary, and the plurality of correlations between the system domain vocabularies, and establish a common vocabulary model based on the first probability and correlation; and steps S480: Analyze a plurality of second chances of a system domain vocabulary appearing in the query data operation intention and the execution instruction operation intention, and establish a common semantic model according to the sentence structure and the second probability.

In steps S410 and S420, a system domain vocabulary set is generated according to the application knowledge database 132 and the domain knowledge database 133, and then the system domain vocabulary set and a plurality of service application parameters are used to form a key entity set. The key entity set includes a plurality of System domain vocabulary. For example, the key entity set contains information such as enterprise domain vocabulary and service application parameters of the enterprise system. The corporate vocabulary refers to the vocabulary that companies in different fields may need. For example, the vocabulary used in the medical industry and the vocabulary used in the transportation industry must be different. Enterprises vary. The service application parameters of the enterprise system are the parameters corresponding to various services provided by the enterprise system. For example, the leave function in the enterprise system may require information such as leave time and leave, and the system domain vocabulary in the key entity collection needs Contains information on leave, annual leave, sick leave, and business leave.

In detail, the key entity collection also includes the data field names that will be available when accessing the data, the service name provided by the enterprise system to the user, the parameter values of the restriction conditions set by the user during the query, and the parameter values of the service application As well as operation functions of the enterprise system, the operation functions of the enterprise system may be operation functions such as leave of absence, overtime application, business application, and reimbursement. The above-mentioned information may also have corresponding aliases, which also need to be entered in the training database. For example, the shipping order may have different names such as shipping details or sales orders for manufacturers in specific fields.

In step S430, the plurality of training corpora are classified into one of a query data operation and a command execution operation. The training corpus can be natural language data such as the user's instructions or questions that may be asked. After the key entity set is established, the training corpus will be classified according to the intention. In one embodiment, the user's intention It is divided into query data operation and execution instruction operation, but the user's intention can be classified more finely, and the present invention is not limited to this. For example, if the user says to the virtual assistant: "Please help me find the shipping order of XX company", it will be classified as query data operation in the intent classification of the present invention, and the virtual assistant will go to the enterprise database to help Check the shipping order of company XX. If the user says to the virtual assistant: "Please ask me for a business trip on January 30." In the intent classification of the present invention, it will be classified as performing an instruction operation, and the virtual assistant will enter the enterprise resource system to help the user ask for leave.

In step S440, the intention of distinguishing the training corpus corresponding to the query data operation according to the category in the enterprise database is to form a plurality of query data operation intentions, and to distinguish the corresponding execution instruction operations according to the service behavior provided by the enterprise resource system. The intent of the training corpus forms a plurality of intents to execute the instruction operation. In one embodiment, the operation of querying data is distinguished according to the enterprise database in each different domain. For example, the database stored in the enterprise database of the medical industry must be different from the enterprise database of the transportation industry, so the user needs of the two are not necessarily the same. For example, users in the medical industry may have different intents for querying medical records, checking room vacancies, etc., and users in the transportation industry may have querying shipping records, querying parcel delivery status, etc. Different intentions of data manipulation. Of course, according to the service behavior provided by the enterprise resource system in each different field, the intention is to distinguish the execution of the instruction operation. As mentioned above, the services provided by the enterprise resource system in the medical industry will of course be different from the transportation industry. Each different field The query data operation or service behavior operation provided by the company may not be universal, so it is also necessary to distinguish the intent of the service provided by the enterprise in each different field. For example, users in the medical industry may have registered services. The provision of services such as hospitalization and ordering healthy meals are different intentions of service behavior operations. For users in the transportation industry, there may be different intentions of service behavior operations to provide services of automatically classifying goods and arranging the order of shipment of goods. .

In steps S450 and S460, a template for querying data operation intention and a template for executing instruction operation is established, and an overall database 131 is established according to a set of key entities, a template for querying data operation intention and a template for executing instruction operation. For example, after distinguishing the query data operation intention and execution instruction operation intention that a user's virtual assistant operating in a certain field company has, the corresponding template can be generated for each intent. According to the above example, medical treatment There will be 4 templates corresponding to querying medical records, querying vacancies in the ward, providing registered services, and providing services for ordering healthy meals in hospitals. The transportation industry will have corresponding querying shipping records, querying parcel delivery status, and providing automatic sorting of goods. Services, four services for arranging the order in which goods are shipped, and then an overall database 131 will be created based on these templates and a collection of key entities.

In step S470, identify the first multiple occurrences of the system domain vocabulary in the key entity set in the training corpus, and analyze the plural sentence structure of the training corpus through the identified system domain vocabulary, and the system domain There are multiple correlations between words, and a common vocabulary model is established according to the first probability and the correlation. In one embodiment, two algorithms, n-GRAM and Context-free grammar (CFG), are used to calculate the probability of each system domain vocabulary appearing in the training corpus. The domain vocabulary analysis training corpus's sentence structure and system domain vocabulary are related to each other to establish a common vocabulary model. For example, if the training corpus contains "I want to query the quotation of XX company" and "I want to query the XX company's shipping order", and "XX company", "quotation" and "shipment order" are all Is a vocabulary in the system field, but in the above example, because "XX company" may appear on average for each query data operation intent, the probability of "XX company" is almost the same in each query data operation intention. The "quotations" and "shipments" only appear in large numbers in the training corpus intended to query some specific data, and will not appear in the training corpus intended to query other data, so the "quotation" And the probability of "shipment order" will be particularly high in the corresponding intent, and will be lower in other intents.

In step S480, a plurality of second probabilities of vocabulary of the system domain appear in the query data operation intention and the execution instruction operation intention, and a common semantic model is established according to the sentence structure and the second probability. In an embodiment, a Hidden Markov Model (HMM) algorithm is used to calculate the probability of a system vocabulary appearing in query data operation intentions and executing instruction operation intentions to establish a common semantic model. For example, During the training data model phase, many training corpora are input. The hidden Markov model algorithm must calculate the probability of the system domain vocabulary appearing at different intents. In combination with the above example, if the training corpus contains "I want to query the shipping order of XX company", according to the n-gram syntax and context-free grammar, it can be found that "XX company" and "shipping order" are both system vocabulary. The hidden Markov model algorithm can further determine the "shipment order" and the query shipment according to the relationship between all the identified system domain vocabularies in the query data operation intention, the probability of executing the instruction operation intention, and the system domain vocabulary. The intent of the data is related, and combined with the system field vocabulary of "XX company", users can automatically query the company's shipping related data in the enterprise database.

After the common vocabulary model and the common semantic model are established, step S320 is performed, and the corpus data is analyzed by using a natural language processing model to generate linguistic feature information corresponding to the corpus data. The linguistic feature information includes a plurality of intents and intent correspondence Probability and plural words. For the detailed flow of step S320, please refer to FIG. 5. FIG. 5 is a flowchart of step S320 according to some embodiments of the present invention. As shown in FIG. 5, step S320 includes the following steps: Step S321: use a common vocabulary model to identify whether the corpus data has a system domain vocabulary that matches the key entity set, set the recognition result to the vocabulary in the language feature information, and Analyze the probability of occurrence of vocabulary in the linguistic feature information; step S322: analyze the sentence structure of the corpus data based on the vocabulary in the feature information; and step S323: use a common semantic model based on the probability of vocabulary occurrence in the feature information and the corpus data The structure of the sentence recognizes the intent of the corpus and the probability of the intent.

In step S321 and step S322, the common vocabulary model is used to identify whether the corpus data has vocabulary in the system domain in the key entity set, set the recognition result to the vocabulary in the language feature information, and analyze the occurrence of the vocabulary in the language feature information. Probability, and then analyze the sentence structure of the corpus data based on the vocabulary in the feature information. For example, the corpus data input by the user is used to identify the vocabulary in the corpus data that contains the vocabulary of the system domain, and then the sentence structure of the corpus data is further determined. For example, if the user says to the virtual assistant: "I want to check the shipping order of XX company last month", the common vocabulary model can identify "XX company", "last month" and "shipping order" "And other words that fit the vocabulary of the system domain.

In step S323, the common semantic model is used to identify the intention of the corpus data and the probability corresponding to the intent according to the probability of vocabulary appearance in the feature information and the sentence structure of the corpus data. According to the example above, after identifying the words "XX company", "last month", and "shipment order", we will further judge the probability of these words in all intents. All intents here include all query data operation intentions and the probability of performing instruction operation intentions.

In step S330, functional context analysis is performed on the linguistic feature information according to the functional context information to determine an operation corresponding to one of the intentions. Before performing the functional situation analysis, a functional situation model and a functional vocabulary model need to be established. When performing the functional situation analysis, the functional situation model first converts the feature vectors from the data in the historical database 134, and then uses machine learning algorithms. The data in the historical database 134 is classified according to various scenarios, and the strong and weak relationship between the feature vector and each scenario is calculated, and then a functional scenario model is generated. Machine learning algorithms suitable for establishing the above functional scenarios include: Support Vector Machines (SVMs) commonly used in traditional machine learning, and Convolutional Neural Networks (CNN) related to deep learning. , Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) algorithms.

Continuing the above, functional vocabulary models are based on a large amount of input training corpus using hidden Markov model algorithm analysis and then word segmentation processing. Then the frequency of word segmentation is counted to generate a word segmentation frequency table, and then a functional vocabulary model is established. For the detailed flow of step S330, please refer to FIG. 6, which is a flowchart of step S330 according to some embodiments of the present invention. As shown in FIG. 6, step S330 includes the following steps: step S331: using corpus data and functional context information to compare with the functional context model and generating a functional context identification result; and step S332: judging the result based on the functional context identification result One of these intentions corresponds to one of querying data operations and executing instruction operations.

In step S331, the corpus data and the functional context information are compared with the functional context model, and a functional context identification result is generated. Functional context information includes the user's identity, the user's position, the user's department, time, and location. Part of the information of the functional context information can be sensed by the input / output device 150, for example, it can detect the current status of the user (for example, whether or not he is back on a business trip). According to the probability and vocabulary of all intents obtained after identifying the user's corpus data, combined with the functional context information, the similarity between the user's corpus data and the data in the training data model can be further estimated as the corresponding intention. Chance.

In step S332, it is determined that one of the intentions corresponds to one of a query data operation and an execution instruction operation according to the functional context recognition result. Because there will be multiple query data operation intentions and multiple instruction operation operation intentions in the training data model, and after the calculation of the common semantic model described above, the probability corresponding to each intent will be generated. Intents with lower probability values can be Use threshold filtering to get the most likely intent and confirm the corresponding operation. As can be seen from the foregoing example, when the words "XX company", "last month", and "shipment order" are identified, these words will be judged with the functional context information to find the most suitable query data operation intention or execute the command operation The intention is that after the above operation, the user will be judged that the user said to the virtual assistant: "I want to check the shipping order of XX company last month". Most likely, you will want to check the shipping order of XX company, so you can Corresponding to what the user wants to perform is a query data operation. The judgement of the functional situation is needed because the user's position, department, operation time, operation place and other information have different needs. For example, purchasers and financial staff will look at [vendor monthly statistics table] However, it is possible that the statistical goals of the [vendor monthly statistics table] of the two are not the same: one is to count the purchase status of the manufacturer, and the other is to count the payment status of the company to the manufacturer. However, when talking with the virtual assistant, the user may not necessarily specify exactly what the [vendor monthly statistics table] is needed. It may only say, "I need the monthly monthly statistics table of the manufacturer last month." It is even more necessary to make a more accurate judgment with the user's functional situation information.

In step S340, if the functional context analysis cannot determine the operation corresponding to one of the intentions, the word segmentation processing is performed on the corpus data. For the detailed flow of step S340, please refer to FIG. 7, which is a flowchart of step S340 according to some embodiments of the present invention. As shown in FIG. 7, step S340 includes the following steps: step S341: segment the corpus data according to the functional vocabulary model to generate a plurality of segmentations; and step S342: calculate the frequency of the segmentations.

In step S341 and step S342, the corpus is segmented according to the functional vocabulary model to generate a plurality of segmentations; then the frequency of the segmentations is calculated. If the functional context analysis cannot determine the operation corresponding to the input corpus data in step S330, the corpus data needs to be segmented. First, the corpus is segmented based on the vocabulary stored in the pre-established functional vocabulary model, and then the frequency of multiple segmentations generated after segmentation is calculated.

In step S350 and step S360, it is determined whether there is a new vocabulary or new corpus data according to the results of the word segmentation processing; if there is a new vocabulary, the natural language processing model is updated according to the meaning of the new vocabulary; if there is a new corpus data, the new corpus is Data intent update functional situation analysis. For the detailed flow of step S360, please refer to FIG. 8. FIG. 8 is a flowchart of step S360 according to some embodiments of the present invention. As shown in FIG. 8, step S360 includes the following steps: step S361: judging whether the frequency of the participles calculated by the word segmentation process is lower than a threshold value; step S362: if one of the participles is lower than the threshold value, the One of the participles is a new vocabulary and receives a definition of the new vocabulary to update the common vocabulary model and the common semantic model; and step S363: if the participles are higher than the threshold value, the corpus data is the new corpus data , And receive the intent of the new corpus material to update the functional context model.

In step S361 and step S362, it is determined whether the frequency of the participles calculated by the segmentation process is lower than the threshold value. If one of the participles is lower than the threshold value, one of the participles is a new word, and Receive new vocabulary definitions to update common vocabulary models and common semantic models. In one embodiment, after calculating the frequency of the segmentation after segmentation processing, set the segmentation below the threshold as a new vocabulary, the virtual assistant will ask the user for the definition of the new vocabulary, and define the new vocabulary and the definition of the new vocabulary. Stored in the common vocabulary model and common semantic model together. For example, the corpus entered by the user is "I want to find the contact person of XX company", and if the virtual assistant cannot determine the meaning of "I want to find the contact person of XX company", it will sort out "I "," Want to find "," XX company "," of "," contact "and other words. If" XX company "is lower than the threshold, the virtual assistant will ask the user what" XX company "means, and then use The answer of the author and the "XX company" are stored in the common vocabulary model and the common semantic model; the new vocabulary also needs to be stored in the system domain vocabulary collection and shared with everyone.

In step S363, if the participles are higher than the threshold, the corpus data is the new corpus data, and the intent of the new corpus data is received to update the functional context model. Continuing the example of "I want to find the contact person of XX company" above, after the word segmentation processing, the words "I", "want to find", "XX company", "of", and "contact person" are separated. If there is no vocabulary Below the threshold value, it means that the virtual assistant does not understand the intention of the corpus. It is possible that the training corpus when training the smart assistant is all about “help me check the contacts of XX company”, so the virtual assistant will not be able to Understand the intent of "I want to find the contact person of XX company", and the virtual assistant needs to ask the user what "I want to find the contact person of XX company", and then the user's response and "I want to find the contact person of XX company" The new corpus of "contacts" is stored in the functional situation model. Before storing the functional model, you need to determine whether the new corpus is a common corpus. If it is, it means that other people will also use the new corpus when using the virtual assistant. Therefore, the new corpus needs to be stored in the system domain vocabulary collection, so that all People share it; but if it is not, it means that the new corpus is just a different term for the user ’s own speaking habits, so only the functional context model needs to be updated, and there is no need to update the system domain vocabulary set.

It can be known from the implementation of the above-mentioned case that the virtual assistant is mainly provided with an automatic learning function, so that the virtual assistant can communicate with the user. If there is a vocabulary that the smart assistant does not understand, you can update the virtual assistant after asking the user The database allows the virtual assistant to automatically learn the user's speaking habits or the special terms in the industry, so that users can use the ERP system more quickly and conveniently.

In addition, the above-mentioned illustration includes sequential exemplary steps, but the steps need not be performed in the order shown. It is within the scope of this disclosure to perform these steps in different orders. Within the spirit and scope of the embodiments of the present disclosure, these steps may be added, replaced, changed, and / or omitted as appropriate.

Although this case has been disclosed as above in the form of implementation, it is not intended to limit the case. Any person skilled in this art can make various modifications and retouches without departing from the spirit and scope of the case. Therefore, the scope of protection of this case should be considered after The attached application patent shall prevail.

Claims (16)

    An automatic learning method for a virtual assistant includes: receiving an audio input and identifying the audio to form a corpus data; analyzing the corpus data using a natural language processing model to generate a language feature information corresponding to the corpus data , Where the linguistic feature information includes a plurality of intentions, the probability corresponding to the intents, and a plurality of words; according to a functional context information, a functional context analysis is performed on the linguistic feature information to determine an operation corresponding to one of the intentions ; If the functional context analysis cannot determine the operation corresponding to one of the intentions, a word segmentation process is performed on the corpus data; based on the result of the word segmentation process, determine whether a new vocabulary or a new corpus data exists ; If the new vocabulary exists, update the natural language processing model according to the meaning of the new vocabulary; if the new corpus data exists, update the functional context analysis according to the intention of the new corpus data; wherein the operation includes a query data operation And one of performing an instruction operation.         The automatic learning method for a virtual assistant according to claim 1, further comprising: generating a system domain vocabulary set according to an application knowledge database and a domain knowledge database; the system domain vocabulary set and a plurality of service application parameters are formed into one A set of key entities, which includes a plurality of system domain vocabularies; classify a plurality of training corpora into one of the query data operation and the execution instruction operation; distinguish the corresponding query data according to the category in the enterprise database The intentions of the training corpora that are operated form a plurality of query data operation intentions, and the intention of the training corpora that should perform the instruction operation is formed according to the service behavior provided by the enterprise resource system to form a plurality of execution instruction operation intentions; establish The query data operation intention template and the execution instruction operation intention template; the overall database is established according to the set of key entities, the query data operation intention template and the execution instruction operation intention template; identifying the System domain words in the set of key entities A plurality of first chances appearing in the training corpora, and analyzing the sentence structure of the training corpora through the identified system domain vocabularies, and the complex numbers of the system domain vocabularies among each other Correlations, and build a common vocabulary model based on the first probabilities and the correlations; and analyze the query data operation intentions and the plurality of second chances of the system domain words appearing in the execution instruction operation intentions , And build a common semantic model based on the sentence structure and the second chances.         The automatic learning method for the virtual assistant according to claim 2, further comprising: using a classifier to classify the data in a historical database to generate a functional context model; and segmenting the training corpus And analyze, and generate a functional vocabulary model based on the data in the historical database.         The automatic learning method for a virtual assistant according to claim 3, wherein the functional context analysis further comprises: using the corpus data and the functional context information to compare with the functional context model and generating a functional context identification result; and According to the result of the functional context identification, it is determined that one of the intentions corresponds to one of the data query operation and the execution instruction operation.         The automatic learning method for a virtual assistant according to claim 4, wherein the word segmentation processing further comprises: segmenting the corpus data according to the functional vocabulary model to generate a plurality of segmentation words; and calculating the frequency of the segmentation words.         The automatic learning method for a virtual assistant according to claim 5, further comprising: determining whether the frequency of the participles calculated by the segmentation processing is lower than a threshold value; if one of the participles is lower than the threshold value, One of the participles is the new vocabulary, and the definition of the new vocabulary is received to update the common vocabulary model and the common semantic model; if the participles are above the threshold, the corpus data is The new corpus material, and receive the intent of the new corpus material to update the functional context model.         The automatic learning method for a virtual assistant according to claim 6, further comprising: judging whether the new corpus data is a common corpus, and if so, updating the vocabulary set of the system domain according to the new corpus data; and updating the vocabulary according to the new vocabulary; System domain vocabulary collection.         The automatic learning method for a virtual assistant according to claim 2, wherein the analysis of the corpus data by the natural language processing model further comprises: using the common vocabulary model to identify whether the corpus data has those that match the key entity set. Vocabulary in the system domain, set the recognition result to the vocabulary, and analyze the probability of the vocabulary appearing; analyze the sentence structure of the corpus data according to the vocabulary; and use the common semantic model according to the probability of the vocabulary appearing and The sentence structure of the corpus data identifies the intentions of the corpus data and the probability corresponding to the intents.         An automatic learning system for a virtual assistant, which is respectively connected to an enterprise database and an enterprise resource system and includes: a processor; a storage device electrically connected to the processor for storing an overall database and an application knowledge A database, a domain knowledge database, and a historical database; an input / output device electrically connected to the processor to provide an interface for inputting an audio; wherein the processor includes: a speech recognition module A group for identifying the audio to form a corpus data; a corpus analysis module electrically connected to the speech recognition module for analyzing the corpus data using a natural language processing model to generate the corpus data A linguistic feature information corresponding to the source data, wherein the linguistic feature information includes a plurality of intents, a probability corresponding to the intents, and a plurality of vocabularies; a context recognition module is electrically connected to the corpus analysis module to A functional contextual information performs a functional contextual analysis of the linguistic feature information to determine an operation corresponding to one of the intentions; an unknown corpus judgment A module electrically connected to the context recognition module for performing a word segmentation process on the corpus data when the situation recognition module cannot recognize the operation corresponding to one of the intentions, and according to the word segmentation The processed result determines whether there is a new vocabulary or a new corpus data; and an updated information module, which is electrically connected to the unknown corpus determination module, and is used to generate a new vocabulary based on the new vocabulary. Update the natural language processing model, and when the new corpus data is generated, update the functional context analysis according to the intention of the new corpus data; wherein the operation includes one of a query data operation and an execution instruction operation.         According to the automatic learning system of the virtual assistant according to claim 9, the processor further comprises: a training module, which is electrically connected to the corpus analysis module, according to the application knowledge database and the field knowledge database A system domain vocabulary set is generated, and the system domain vocabulary set and a plurality of service application parameters are formed into a key entity set, the key entity set includes a plurality of system domain vocabularies, and a plurality of training corpora are classified into the query data operation and One of the execution instruction operations, according to the category in the enterprise database, distinguishes the intention of the training corpus corresponding to the query data operation to form a plurality of query data operation intentions, and distinguishes according to the service behavior provided by the enterprise resource system. Forming a plurality of execution instruction operation intentions for the intentions of the training corpus that should execute the instruction operations; a template creation module, which is electrically connected to the training module, establishes the query data operation intention templates, and the execution The template of the instruction operation intention, according to the set of key entities, the operation intention of the query data The template and the templates for executing the instruction operation intention establish the overall database; a vocabulary model building module is electrically connected with the template building module to identify the system domain words in the key entity set in the training languages The first probabilities appearing in the data, and analyze the plural sentence structure of the training corpora through the identified system domain vocabularies, and the plurality of correlations between the system domain vocabularies with each other, and Establish a common vocabulary model according to the first chances and the correlations; and a semantic model establishment module, which is electrically connected to the template establishment module, analyzes the query data operation intentions and the execution instruction operation intentions A plurality of second probabilities of the vocabularies of the system domains appear, and a common semantic model is established according to the sentence structure and the second probabilities.         The automatic learning system for a virtual assistant according to claim 10, wherein the processor further includes: a situation training module, which is electrically connected to the situation analysis module, and is configured to use a classifier to store the historical data in the historical database. Classify the relationship strength of the data to generate a functional context model; and a vocabulary training module that is electrically connected to the unknown corpus judgment module to perform word segmentation and analysis on the training corpus and based on the historical data The data in the database produces a functional vocabulary model.         The automatic learning system for a virtual assistant according to claim 11, wherein the context analysis module is further configured to compare the functional context information and the functional context information with the functional context model and generate a functional context identification result, And it is judged according to the result of the functional context identification that one of the intentions corresponds to one of the data query operation and the execution instruction operation.         The automatic learning system for a virtual assistant according to claim 12, wherein the unknown corpus judgment module is further configured to perform word segmentation on the corpus data according to the functional vocabulary model to generate a plurality of participles to calculate the participles. Frequency of.         The automatic learning system for a virtual assistant according to claim 13, wherein the update information module is further configured to determine whether the frequency of the participles calculated by the segmentation processing is lower than a threshold value; if one of the participles is Below the threshold, one of the participles is the new vocabulary, and the definition of the new vocabulary is received to update the common vocabulary model and the common semantic model; if the participles are above the threshold, then The corpus data is the new corpus data, and receives the intention of the new corpus data to update the functional context model.         The automatic learning system for a virtual assistant according to claim 14, wherein the update information module is further configured to determine whether the new corpus data is a common corpus, and if so, update the system domain vocabulary set according to the new corpus data; and The system domain vocabulary set is updated based on the new vocabulary.         The automatic learning system for a virtual assistant according to claim 10, wherein the corpus analysis module is further configured to use the common vocabulary model to identify whether the corpus data has vocabularies corresponding to the system domain in the key entity set, Set the recognition result to the words, and analyze the occurrence probability of the words, analyze the sentence structure of the corpus data according to the words, and use the common semantic model according to the occurrence probability of the words and the corpus data The structure of the sentence pattern identifies the intentions of the corpus data and the probability of the intentions.