TWI805008B - Customized intent evaluation system, method and computer-readable medium - Google Patents

Abstract

The present invention provides a customized intent evaluation system, method and computer-readable medium, including a natural language processing module and an intention evaluation module, wherein, the natural language processing module processes a text messages to generate corresponding candidate intention commands, and then the intention evaluation module uses a custom logic decision tree or an intention prediction results obtained through machine learning to analyze the candidate intention commands to quickly and accurately obtain a intention command that meets the user's true intentions.

Description

System, method and computer-readable medium for selection of customized intent

The present invention relates to an intention selection technology, in particular to a customized intention selection system, method and computer readable medium.

Today's users are accustomed to giving instructions to smart interactive devices (such as smart phones, smart speakers, etc.) by voice, so as to complete the services required by users through smart interactive devices, but the instructions issued by users may contain multiple meanings at the same time , causing the smart interactive device to easily generate an error response.

For example, when the user's instruction is "I want to listen to a story", there may be a song titled "Story" and a book titled "Story" in the smart interactive device. In this situation, although conventional technologies such as Google Home or Amazon Alexa can still directly trigger device actions for multilingual commands, this action may not necessarily satisfy the user's true intention. Therefore, how to quickly and accurately respond to the user's satisfactory answer will be an important task of the intelligent interactive device.

Therefore, how to provide an innovative intention selection technology to overcome the lack of the above-mentioned conventional technology or provide related functions has become an urgent problem to be solved.

In order to solve the above problems or provide related functions, the present invention provides a customized intent selection system, which includes: a natural language processing module, which performs segmentation processing on a text message to generate plural segmentation words, and the plural segmentation Words are tagged with part-of-speech and named entities so that the plural segmented words have part-of-speech tagging and identification tags, and then the natural language processing module combines the plural segmented words with part-of-speech tagging and identification tags with the plural intentions in a semantic list Templates are matched to obtain the relevant intent templates of the plurality of segmented words with part-of-speech tags and identification tags, so that the natural language processing module can obtain the corresponding intent templates through the plurality of segmented words with part-of-speech tags and identification tags. A plurality of candidate intent commands in the semantic list; and an intent selection module that receives the plurality of candidate intent commands in the semantic list from the natural language processing module, so that the intent selection module is based on a custom logic The decision tree judges the execution order of the plurality of candidate intended instructions, and generates an execution intention instruction after calculation according to the execution order.

The present invention provides a method for selecting customized intentions, which includes: after a natural language processing module performs segmentation processing on a text message to generate plural segmentations, the plural segmentations are tagged with part of speech and named entities, so that the Plural segmented words have part-of-speech tagging and identification tags; the natural language processing module matches the plural segmented words with part-of-speech tagging and identification tags with the plural intent template in a semantic list, and obtains the plural with part-of-speech tagging and identification The association intention template of the tagged word segmentation; the natural language processing module obtains the corresponding plural candidate meanings in the semantic list through the association intent template of the plural segmentation words with part-of-speech tags and identification tags Graph instruction: an intent selection module receives the plurality of candidate intent instructions from the semantic list of the natural language processing module, so that the intent selection module judges the plurality of candidate intent instructions according to a custom logic decision tree an execution order; and the intention selection module generates an execution intention instruction after calculation according to the execution order.

In one embodiment, the natural language processing module includes a part-of-speech tagging unit and a named entity unit, wherein the part-of-speech tagging unit performs part-of-speech tagging on the plural segmented words, so that the plural segmented words all have the part-of-speech tagging, The named entity unit then performs named entity tagging on the plural segmented words with part-of-speech tagging, so that the plural segmented words further have the identification tag.

In one embodiment, the natural language processing module includes a logic and semantic editing unit, which is to provide a service provider to edit the plural intent templates and corresponding plural intent instructions and plural service behaviors, so as to base on the plural intent templates After the semantic list is formed by the multiple intent commands and the multiple service behaviors, the logic and semantic editing unit further provides the service provider to set the execution order of the multiple intent commands, so as to form the custom system according to the execution order of the multiple intent commands Logical decision tree.

In one embodiment, the intention selection module includes a subscription information unit, which is used to screen candidate intention instructions among the plurality of candidate intention instructions that are not related to the subscription information according to a subscription information.

In one embodiment, the intent selection module includes a support vector machine classifier, which converts the text message into a vector, and performs intent prediction through a machine learning semantic understanding model to generate a predicted intent instruction, wherein, when the intent When the selection module cannot determine which of the plurality of candidate intent instructions has the highest execution order according to the custom logic decision tree, the support vector machine classifier compares the predicted intent instruction with the plurality of candidate intent instructions, and then selects from the plurality of candidate intent instructions plural candidates The intention instruction consistent with the predicted intention instruction is compared among the intention instructions as the execution intention instruction.

In one embodiment, the intent selection module includes a feature engineering unit, which extracts the features of the user's service history and personal information to obtain the required multiple features, and the feature engineering unit builds the multiple features into The feature vector is provided to the support vector machine classifier to train the machine learning semantic understanding model.

In addition, the computer-readable medium of the present invention is applied in a computing device or a computer, and stores instructions to execute the above-mentioned customizing intent selection method.

As can be seen from the above, the present invention provides a system, method, and computer-readable medium for customizing intention selection, which analyzes the user's candidate intention instructions by using a preset custom logical decision tree or intention prediction results obtained through machine learning. , and then provide the correct intent command to the user device to execute the service required by the user. Therefore, compared with the conventional technology that cannot correctly judge the instruction with multiple semantics issued by the user, the present invention can not only provide the service provider to set a customized logical decision tree according to the target customer group for service guidance, but also can use Quickly and accurately analyze the user's true intentions from the user's past usage history and personal data.

1: Customized intent selection system

10: Natural language processing module

11: Segmentation unit

12: Part-of-speech tagging unit

13: Named Entity Units

14: Semantic Comprehension Unit

15: Logic and Semantic Editing Unit

16:Semantic understanding database

20: Intent Selection Module

21:Subscription information unit

22: Logic execution unit

23: Support Vector Machine Classifier

24: Feature engineering unit

25: User Journey Database

26: Personal database

30: Use end device

40: Service Provider

C1: Feature vector of service history

C2: Feature vector of personal data

S31 to S36: Steps

FIG. 1 is a schematic diagram of the structure of the customized intention evaluation system of the present invention;

Figure 2 is a schematic diagram of the user's service history and personal data feature vectors of the present invention; and

FIG. 3 is a schematic flow chart of the method for selecting customized intentions according to the present invention.

The implementation of the present invention is described below through specific specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification.

It should be noted that the structures, proportions, sizes, etc. shown in the drawings attached to this specification are only used to match the content disclosed in the specification, for the understanding and reading of those familiar with this technology, and are not used to limit the implementation of the present invention Therefore, it has no technical substantive meaning. Any modification of structure, change of proportional relationship or adjustment of size shall still fall within the scope of this invention without affecting the effect and purpose of the present invention. The technical content disclosed by the invention must be within the scope covered. At the same time, terms such as "a", "first", "second", "upper" and "lower" quoted in this specification are only for the convenience of description and are not used to limit the scope of the present invention. The scope of implementation, the change or adjustment of its relative relationship, without substantial change in technical content, shall be regarded as the scope of implementation of the present invention.

FIG. 1 is a schematic diagram of the architecture of a customized intention evaluation system 1 of the present invention. As shown in FIG. 1 , the customized intention selection system 1 includes a natural language processing module 10 and an intention selection module 20, and may further include at least one (such as plural) user-end devices 30 and at least one (such as Plural) service provider 40, wherein, "plural" represents more than two (such as two, three, four or more than five). The natural language processing module 10 and the intention selection module 20 are respectively connected to the user device 30 and the service provider 40 through communication, and the natural language processing module 10 receives a text message generated by the voice conversion performed by the user device 30 . In one embodiment, the client device 30 converts the voice into a text message by utilizing the speech recognition technology of the prior art, and then sends the text message to the natural language processing module 10 , which speech recognition technology is not limited here.

The natural language processing module 10 includes a word segmentation unit 11, a part-of-speech tagging unit 12, a named entity unit 13, a semantic understanding unit 14, a logic and semantic editing unit 15 and a semantic understanding database 16, wherein the The natural language processing module 10 is used to receive text messages, and perform segmentation processing on the text messages, so as to perform part-of-speech tagging and named entity tagging on the results of word segmentation processing, and finally perform semantic understanding to obtain plural candidates intent instruction.

The intent selection module 20 includes a subscription information unit 21, a logic execution unit 22, a support vector machine (Support Vector Machine, SVM) classifier 23, a feature engineering unit 24, a user history database 25 and a personal database 26. Wherein, the intention selection module 20 is connected to the natural language processing module 10 by communication, and is used to evaluate the plurality of candidate intention instructions understood by the natural language processing module 10, so as to select from the plurality of candidate intention instructions Find the most suitable execution intent instruction for the user, and then execute the service behavior corresponding to the execution intent instruction.

Specifically, the natural language processing module 10 and the intent selection module 20 are built on servers (such as general-purpose servers, file-type servers, storage unit-type servers, etc.) and electronic devices with appropriate calculation mechanisms such as computers. Among them, the natural language processing module 10 and the intention selection module 20 can be software, hardware or firmware; if it is hardware, it can be a processing unit, processor, computer or servo with data processing and computing capabilities If it is software or firmware, it may include instructions executable by a processing unit, processor, computer or server, and may be installed on the same hardware device or distributed across multiple different hardware devices. Furthermore, the user device 30 can be a smart speaker, a smart phone, a tablet computer, a personal computer or other electronic devices, but is not limited to the above, and the service provider 40 refers to providing various services (such as video streaming services) etc.) service providers or service providing equipment.

The word segmentation unit 11 performs word segmentation processing after receiving the text message, so as to divide the text message into plural segment words. In one embodiment, the purpose of word segmentation is to divide the input text information into units with the smallest meaning, and the word segmentation methods include statistical, dictionary, or a combination of the two, and the word segmentation methods are not limited here. . For example, word segmentation unit 11 receives the instruction of the user's text message (for example [I want to listen to Jay Chou's Daoxiang]), and performs word segmentation on the instruction of text message to obtain plural word segmentation (such as [I, want, listen , Jay Chou, of, Daoxiang]), where symbols (such as commas) are used to separate each hyphen.

The part-of-speech tagging unit 12 performs part-of-speech tagging on the plural segmented words respectively, so that the plural segmented words all have part-of-speech tagging, wherein, the methods of part-of-speech tagging include statistical, dictionary or a mixture of the two. Furthermore, statistical formulas are mainly used at present, but this embodiment adopts statistical formulas based on the definition of Chinese Treebank (Chinese Treebank) 3.0. For example, the part-of-speech tagging unit 12 performs part-of-speech tagging on plural hyphenated words (such as [I, want, listen, Jay Chou, de, Daoxiang]) to generate plural hyphenated words with part-of-speech tagging (such as [我_PN, 要_VV , listen to _VV, Jay Chou_NR, _DEG, Daoxiang_NR]).

The named entity unit 13 uses named entity recognition (Named Entity Recognizer, NER) to perform named entity tagging on plural segmented words with part-of-speech tags, so that plural segmented words have part-of-speech tagging and identification tags, wherein named entity recognition Statistical, dictionary or a combination of both. Furthermore, statistical formulas are mainly used at present, and the identification labels in this embodiment are determined by the labels marked in the training data, which means they can be customized. For example, the named entity unit 13 uses named entity recognition to mark [I_PN, Want_VV, Listen_VV, Jay Chou_NR, De_DEG, Daoxiang_NR] to generate [I_PN_o, To _VV_o, listen to _VV_o, Jay Chou_NR_SINGER, _DEG_o, Daoxiang_NR_SONG], where o represents a non-special named entity, SINGER represents a singer and SONG represents a song, and SINGER and SONG are custom labels .

The semantic comprehension unit 14 performs semantic comprehension by using a plurality of intent templates in the word segmentation matching semantic list with identification tags to match a plurality of candidate intent instructions. In detail, the semantic understanding unit 14 uses a pattern-based semantic understanding method to perform semantic understanding through intent templates. Therefore, the semantic understanding unit 14 matches the plurality of segmented words with part-of-speech tags and identification tags with the plural intent templates in the semantic list, and then obtains the plural intent templates associated with the plurality of segmented words with part-of-speech tags and identification tags, and The semantic understanding unit 14 obtains the plurality of candidate intention instructions in the corresponding semantic list through the plurality of intent templates associated with the segmented words with part-of-speech tags and identification tags. In addition, if the semantic understanding unit 14 fails to match the corresponding intent template, the semantic understanding unit 14 responds through the client device 30 that the service cannot be executed; or, if the semantic understanding unit 14 only matches a corresponding intent template, Then, a candidate intent instruction corresponding to the intent template is provided to the intent selection module 20, so that the intent selection module 20 can directly use the candidate intent instruction as an execution intent instruction.

For example, the semantic understanding unit 14 compares all intent templates sequentially with [I_PN_o, wants_VV_o, listen_VV_o, Jay Chou_NR_SINGER, _DEG_o, Daoxiang_NR_SONG], and can compare [I_PN_o , want_VV_o, listen_VV_o, Jay Chou_NR_SINGER's _DEG_o, Daoxiang_NR_SONG] are consistent or identical to the intent template of [I, want, listen, SINGER, de, SONG], and then obtain candidate intent instructions.

The logic and semantic editing unit 15 provides that the service provider 40 can pre-edit the plural intention templates and corresponding plural intention instructions and plural service behaviors according to its target customer groups, so as to base on the plural intention templates, plural intention instructions and plural The service behavior forms a semantic list (as shown in Table 1). For example, the logic and semantic editing unit 15 provides an interface (such as a webpage or a smart phone application program, not limited here) to the service provider 40 to build the intention template of [I, want, listen, SINGER, of, SONG], and Set the intent instruction as [I1] and the service behavior as [start music service S1].

Table 1: Semantic Checklist

In one embodiment, the logic and semantic editing unit 15 can also provide the service provider 40 to set the execution sequence when multiple segmented words with part-of-speech tags and identification tags trigger services with different intent commands, so as to form a customized logical decision tree, For example, the service provider 40 (such as the first service provider) can set according to its target customer group when the first service behavior and the second service behavior are triggered at the same time, the first service behavior will always be executed prior to the second service behavior, Another service provider 40 (such as the second service provider) can also set the second service behavior to be executed always prior to the third service behavior according to its target customer group.

The semantic understanding database 16 stores the semantic list customized by the service provider 40, and provides the intent templates in the semantic list to the semantic understanding unit 14 for similarity comparison.

The subscription information unit 21 is based on the subscription information of the user to screen the candidate intention instructions irrelevant to the subscription information among the plurality of candidate intention instructions, so as to exclude the candidate intention instructions irrelevant to the subscription information. In one embodiment, the subscription information refers to the service list that the user can use in the user-end device 30. For example, the user-end device 30 can provide the user with a service list with multiple service behaviors, and the user chooses to subscribe ( Access) service behaviors in the service list, thereby forming a service list (that is, subscription information). For example, when a candidate intent is triggered, if subscribe If there is no music subscription service in the information, the user device 30 will respond to the user with no music subscription service.

The logic execution unit 22 receives multiple candidate intent commands related to the subscription information from the subscription information unit 21, and judges the execution order of the multiple candidate intent commands according to the custom logic decision tree of the logic and semantic editing unit 15, and then The one with the highest execution order among the plurality of candidate intended instructions is regarded as an execution intended instruction. In another embodiment, the execution order can also be calculated to generate an execution intent instruction.

The support vector machine (SVM) classifier 23 is customized through the support vector machine (SVM) in machine learning and based on the feature vector built by the user's past service history and personal information The machine learning semantic understanding model is trained, and after the text information is converted into a vector, the intention prediction is performed through the machine learning semantic understanding model to generate a predicted intention instruction (ie, the intention prediction result). In one embodiment, when the logic execution unit 22 can use the customized logic decision tree to select the highest execution order among the plurality of candidate intention instructions as an execution intention instruction, and transmit the execution intention instruction to the support vector machine classifier 23, Although the support vector machine classifier 23 also generates the prediction intent instruction, it still directly triggers the execution intent instruction transmitted from the logic execution unit 22 .

The feature engineering unit 24 extracts the features of the user's past service history and personal information to obtain the required complex features, and constructs the complex features into feature vectors to provide the support vector machine classifier 23 for machine learning. Semantic understanding model for training.

The user history database 25 records the user's service history. For example, the user will listen to Jay Chou's music by using the command "I want to listen to Jay Chou's Daoxiang" at 7 o'clock every Monday night, and the user history database 25 will record this service history.

The personal database 26 records personal information. For example, personal information such as the user's gender, occupation, education level, location and age, but not limited to the above. In addition, this personal information will be registered when the client device 30 is first used.

The following embodiment is the first embodiment of the customized intent evaluation system 1 of the present invention, and the main content of the first embodiment is as follows, and the rest of the content is the same as the description of FIG. 1 above.

In this embodiment, the user subscribes to music service S1, music service S2, and music service S3 through the client device 30, and when the user speaks to the client device 30 [I want to listen to the song of Andy Lau's Bingyu] After the voice message, the user device 30 converts the voice message into a text message and sends it to the customized intention selection system 1 .

Specifically, the natural language processing module 10 of the customized intent selection system 1 receives the text message from the user device 30, and the word segmentation unit 11 converts the text message (such as [I want to listen to Andy Lau's Bing Yu song]) Segmentation into plural breaks (such as [I, want, listen, Andy Lau, of, Bingyu, this, first, song]), and the part-of-speech tagging unit 12 carries out part-of-speech tagging of plural breaks to generate plural breaks with part-of-speech tags Words (such as [I_PN, Want_VV, Listen_VV, Andy Lau_NR, _DER, Bingyu_NR, This_PN, Song_M, Song_NN]), wherein, PN is a pronoun, VV is a verb, NR is a proper noun, DER is the unique part of speech of the word "的", M is a quantifier, and NN is a noun. Furthermore, the named entity unit 13 performs named entity tagging on the plural part-of-speech tagged words to generate plural part-of-speech tagged and identification tags (such as [I_PN_o, want_VV_o, listen_VV_o, Andy Lau_ NR_SINGER, _DER_o, Bingyu_NR_SONG, _PN_, Song_M_o, Song_NN_o]), where o means that the vocabulary is not a named entity, SINGER means singer, and SONG means song.

Next, the semantic understanding unit 14 divides the word segmentations with part-of-speech tagging and identification tags (such as [我_PN_o, want_VV_o, listen_VV_o, Andy Lau_NR_SINGER, _DER_o, Bingyu_NR_SONG, this_PN_, first _M_o, song_NN_o]) matches the intent template in the semantic list (as shown in Table 1 above), to obtain two candidate intent instructions I1 and I2 through the matched intent template, as shown in Table 2.

Table 2: Candidate Intent Instructions

The subscription information unit 21 of the intent selection module 20 screens the candidate intent commands I1 and I2 according to the user's subscription information to confirm that the user has subscribed to related services, and then the logic execution unit 22 uses the service provider 40 in the logic and semantic editing unit The custom logic decision tree (as shown in Table 3) set in 15 determines the execution sequence of the candidate intention instructions, wherein the logic execution unit 22 takes the candidate intention instruction I1 as the execution intention instruction according to the rule R1.

Table 3: Customized Logical Decision Tree

Afterwards, the support vector machine classifier 23 converts the original text message of "I want to listen to Andy Lau's Ice Rain" into a vector, and performs intent prediction through the machine learning semantic understanding model to generate a prediction intent instruction, and the prediction intent The instruction is compared with the execution intention instruction I1, and no matter whether the predicted intention instruction is the same as the execution intention instruction I1 after comparison, the execution intention instruction I1 is triggered to make the user device 30 start the music service S1.

In addition, the feature engineering unit 24 will perform feature extraction on the user's service history and personal information from the user history database 25 and personal database 26 to generate the feature vector C1 of the service history and the feature vector C2 of the personal data, so as to provide Intent to support vector machine classifier 23 predict. For example, as shown in FIG. 2 , they are the feature vectors C1 and C2 generated by the user using the intention command to trigger the music service S1 at 7:00 p.m. on Monday in the past.

What's more, as shown in Figure 2 and Table 4, the Bag of Words model in the text vector of the feature vector C1 of the service history will change due to the number of intention instructions accumulated by the user, and the user history data The library 25 stores the correspondence table of each feature in the feature vector C1 of the service history, for example, the time feature correspondence table in Table 5, the week feature correspondence table in Table 6, and the intent command feature correspondence table in Table 7, wherein the intent command feature corresponds to The list will be pre-numbered by the service provider 40 in the semantic understanding database 16. For example, the semantic list in Table 1 can correspond to the intent instruction feature correspondence table in Table 7. Also, the personal database 26 stores the correspondence table of each feature in the feature vector C2 of the personal data, for example, the sex feature correspondence table of Table 8, the industry characteristic correspondence table of Table 9, the regional characteristic correspondence table of Table 10 and Table 11 The corresponding table of educational level characteristics, in which, the age is the characteristic value based on the original age excluding months. Therefore, the aforementioned corresponding tables of feature vectors C1 and C2 (such as Table 4 to Table 11) can be feature vector tables for machine learning, and are used to provide the support vector machine classifier 23 for machine learning on text messages.

Table 4: Bag of words model

Table 5: Time Feature Correspondence Table

Table 6: Correspondence table of week features

Table 7: Correspondence table of intent instruction features

Table 8: Correspondence table of gender characteristics

Table 9: Correspondence table of industry characteristics (according to the classification of information from the Accounting Office of the Executive Yuan)

Table 10: Correspondence table of regional characteristics

Table 11: Correspondence table of educational level characteristics

On the other hand, the user history database 25 records user context characteristics (as shown in Table 12) to provide training data for the support vector machine classifier 23 .

Table 12: User Situation Characteristics Table

Moreover, the personal data base 26 also records the user's personal data characteristics (as shown in Table 13) to provide training data for the support vector machine classifier 23 .

Table 13: Profile Characteristics Table

The following embodiment is the second embodiment of the customized intention evaluation system 1 of the present invention, and the main content of this second embodiment is as follows, and the rest of the content is the same as the description of the above-mentioned FIG. 1 and the first embodiment.

In this embodiment, the user subscribes to the music service S2, the story service S4 and the broadcast service S5 through the user device 30, and when the user tells the voice message of "I want to listen to the story" to the user device 30, the The voice message is converted into a text message to be sent to the customized intention evaluation system 1 .

Specifically, the natural language processing module 10 of the customized intent selection system 1 receives text messages from the user-end device 30, and the word segmentation unit 11 divides the text messages (such as [I want to listen to a story]) into plural segment words (such as [I, want, listen, story]), and the part-of-speech tagging unit 12 carries out part-of-speech tagging with plural part-of-speech tags to generate plural part-of-speech tagged part-of-speech tags (such as [我_PN, want_VV, listen_VV, Story_NN]). Furthermore, the named entity unit 13 performs named entity tagging on the plural part-of-speech tagged words to generate plural part-of-speech tagged and identification tags (such as [我_PN_o, want_VV_o, listen_VV_o, story_ NN_SONG/STORY]), where "Story" may represent a song titled Story, or simply means to listen to a story.

The semantic understanding unit 14 matches the plural segmented words (such as [我_PN_o, want_VV_o, listen_VV_o, story_NN_SONG/STORY]) with part-of-speech tags and identification tags to the intent templates in the semantic list (as shown in Table 1 above) shown), to obtain two candidate intent instructions 12, I4 through the matched intent template, as shown in Table 14.

Table 14: Candidate Intent Instructions

The subscription information unit 21 of the intent selection module 20 screens the candidate intent commands I2, I4 according to the user's subscription information to confirm that the user has subscribed to related services, and the logic execution unit 22 uses the service provider 40 in the logic and semantic editing unit 15 to determine the execution order of the candidate intent instructions. However, when the customized logic decision tree does not set the execution sequence of the candidate intent instructions I2, I4, the logic execution unit 22 will directly send the candidate intent instructions I2, I4 to the support vector machine classifier 23.

At this time, the support vector machine classifier 23 converts the text message of "I want to listen to a story" into a vector and then performs intent prediction to generate a predicted intent instruction, and then compares the predicted intent instruction with the candidate intent instructions I2, I4, so that Obtain the intersection between the predicted intention instruction and the candidate intention instructions I2, I4; in other words, the support vector machine classifier 23 can obtain the candidate intention instruction consistent with or the same as the predicted intention instruction to obtain the candidate intention instruction I2, and finally the candidate intention instruction I2 As the execution intent instruction, the user device 30 starts the music service S2.

Fig. 3 is a schematic diagram of the method flow chart of the customization intention selection method of the present invention, and the main content of the method flow chart of the customization intention selection method is as follows, and the rest of the content is the same as that described in Fig. 1 above, and will not be repeated here. , the method flow includes the following steps S31 to S36.

In step S31 , the natural language processing module 10 performs segmentation processing on the user's text message to generate plural segmented words.

In step S32 , the natural language processing module 10 performs part-of-speech tagging and named-entity tagging on the segmented words, so that plural segmented words have part-of-speech tags and identification tags.

In step S33 , the natural language processing module 10 uses the plurality of segmentation words with identification tags to match the plurality of intent templates in the semantic list for semantic understanding, and then matches a plurality of candidate intent instructions.

In step S34 , after receiving the plurality of candidate intent commands from the natural language processing module 10 , the intent selection module 20 screens the plurality of candidate intent commands according to the subscription information of the user, so as to exclude candidate intent commands irrelevant to the subscription information.

In step S35 , the intention selection module 20 judges the execution order of the plurality of candidate intention instructions according to the customized logic decision tree, and then takes the highest execution order among the plurality of candidate intention instructions as an execution intention instruction.

In step S36, the intent selection module 20 converts the original text message into a vector and then performs intent prediction to generate a predicted intent command, wherein, if it is impossible to judge the execution order of the plurality of candidate intent commands according to the customized logic decision tree When it is the highest, the intention selection module 20 compares the predicted intention instruction with the plurality of candidate intention instructions, and then compares the intention instruction consistent with or the same as the predicted intention instruction from the plurality of candidate intention instructions as an execution intention instruction.

In addition, the present invention also discloses a computer-readable medium, which is applied to a computing device or computer having a processor (for example, CPU, GPU, etc.) and/or memory, and stores instructions, and can be used by this computing device or The computer executes the computer-readable medium through the processor and/or memory, so as to execute the above-mentioned method and each step when executing the computer-readable medium.

In summary, the present invention provides a system, method, and computer-readable medium for customizing intention selection, which analyzes the user's candidate intentions by using a preset custom logical decision tree or intention prediction results obtained through machine learning. Instructions, and then provide the correct intent instructions to the user device to perform the services required by the user, so compared with the inability to correctly judge the multi-semantics issued by the user According to the conventional technology of command, the present invention can not only provide the service provider to set a customized logical decision tree according to the target customer group for service guidance, but also quickly and accurately analyze it by using past usage history and personal data the user's true intent.

Therefore, compared with the prior art, the present invention has the following technical features and effects thereof:

1. The present invention can provide the service provider to adjust the service priority to meet the target customer group, thereby forming a customized logical decision tree, and allowing the user to conduct service guidance according to the customized logical decision tree set by the service provider.

2. The present invention uses the user's service history and personal information to train the machine learning semantic understanding model to provide a customized machine learning semantic understanding model for the user's personal habits, so that the machine learning semantic understanding model can be based on the user's The habit of generating intention prediction results, and the correct service response, in order to better meet the user's true intentions, and thus avoid providing wrong services to users.

3. The present invention performs part-of-speech tagging and named entity tagging after segmenting the user's text message, so as to quickly find the corresponding candidate intent commands by matching plural segmented words and intent templates with part-of-speech tagging and identification tags .

The above-mentioned embodiments are only illustrative to illustrate the principles and effects of the present invention, and are not intended to limit the present invention. Anyone skilled in the art can modify and change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the rights of the present invention should be listed in the scope of the patent application.

1: Customized intent selection system

10: Natural language processing module

11: Segmentation unit

12: Part-of-speech tagging unit

13: Named Entity Units

14: Semantic Comprehension Unit

15: Logic and Semantic Editing Unit

16:Semantic understanding database

20: Intent Selection Module

21:Subscription information unit

22: Logic execution unit

23: Support Vector Machine Classifier

24: Feature engineering unit

25: User Journey Database

26: Personal database

30: Use end device

40: Service Provider

Claims (11)

A customized intent selection system, comprising: a natural language processing module, which processes a text message to generate plural sentence segments, and then tags the plural sentence words with part of speech and named entities, so that the plural sentence sentences All words have part-of-speech tagging and identification tags, and then the natural language processing module matches the plural segmentation words with part-of-speech tagging and identification tags with the plural intention templates in a semantic list to obtain the plural words with part-of-speech tagging and identification tags The related intent templates of the segmented words, so that the natural language processing module can obtain the corresponding plurality of candidate intent instructions in the semantic list through the related intent templates of the plurality of segmented words with part-of-speech tags and identification tags; and an intent The selection module receives the plurality of candidate intent instructions from the semantic list of the natural language processing module, so that the intent selection module judges the execution order of the plurality of candidate intent instructions according to a custom logic decision tree, and then According to the execution sequence, an execution intent instruction is generated after calculation, wherein the natural language processing module includes a logic and semantic editing unit, which is to provide a service provider to edit the plural intent templates and their corresponding plural intent instructions and plural After the service behavior forms the semantic list based on the plural intent templates, the plural intent instructions and the plural service behaviors, the logic and semantic editing unit provides the service provider to set the execution order of the plural intent instructions according to the plural The order of execution of the intended instructions forms the custom logical decision tree. The customized intent selection system as described in claim 1, wherein the natural language processing module includes a part-of-speech tagging unit and a named entity unit, and the part-of-speech tagging unit performs part-of-speech tagging on the plural segmented words, so that the plural Segmented words all have the part-of-speech tag, and the named entity The unit then performs named entity tagging on the plural segmented words with part-of-speech tagging, so that the plural segmented words further have the identification tag. The customized intent selection system as described in claim item 1, wherein the intent selection module includes a subscription information unit, which is based on a subscription information to filter the candidate intent instructions among the plurality of candidate intent instructions that are not related to the subscription information, so as to Exclude candidate intent commands that are not relevant to the feed. The customized intention selection system as described in claim 1, wherein the intention selection module includes a support vector machine classifier, which converts the text message into a vector, and performs intention prediction through a machine learning semantic understanding model to generate A predicted intent instruction, wherein, when the intent selection module cannot determine which of the plurality of candidate intent instructions has the highest execution order according to the customized logical decision tree, the support vector machine classifier compares the predicted intent instruction For the plurality of candidate intention instructions, an intention instruction consistent with the predicted intention instruction is compared from the plurality of candidate intention instructions as the execution intention instruction. The customized intent selection system as described in claim 4, wherein the intent selection module includes a feature engineering unit, which extracts the features of the user's service history and personal information to obtain the required multiple features, the The feature engineering unit constructs the complex features into feature vectors, which are provided to the support vector machine classifier for training the machine learning semantic understanding model. A method for selecting customized intentions, comprising: after a natural language processing module performs word segmentation processing on a text message to generate plural word breaks, and then carries out part-of-speech and named entity tagging on the plural word breaks, so that the plural word breaks All have part-of-speech tagging and identification tags; The natural language processing module matches the plurality of segmented words with part-of-speech tags and identification tags with the plurality of intent templates in a semantic list to obtain the relevant intent templates of the plurality of segmented words with part-of-speech tags and identification tags; The natural language processing module obtains the plurality of candidate intent instructions corresponding to the semantic list through the relevant intent templates of the plurality of segmented words with part-of-speech tags and identification tags; an intent selection module receives from the natural language processing module The plurality of candidate intent instructions in the semantic list is used to determine the execution sequence of the plurality of candidate intent instructions by the intent selection module according to a custom logical decision tree; and the intent selection module calculates according to the execution order Generate an execution intent command, wherein the natural language processing module provides a service provider to edit the multiple intent templates and corresponding multiple intent commands and multiple service behaviors, based on the multiple intent templates, the multiple intent commands and After the plurality of service actions form the semantic list, the natural language processing module provides the service provider to set the execution order of the plurality of intent instructions, so as to form the customized logical decision tree according to the execution order of the plurality of intent instructions. The customized intent selection method as described in claim item 6, further includes the part-of-speech tagging of the plural sentence by the natural language processing module, so that all the plural sentence words have the part-of-speech tag, and then processed by the natural language The module tags the plural part-of-speech tagged words with named entities, so that the plural sentence words further have the identification tag. The customized intent selection method as described in claim item 6 further includes the intent selection module screening the plurality of candidate intent commands that are not related to the subscription information according to a subscription information, so as to exclude the candidate intent commands that are not related to the subscription information Candidate intent directives for . The customized intent selection method as described in claim 6 further includes converting the text message into a vector by the support vector machine classifier of the intent selection module, and performing intent prediction through a machine learning semantic understanding model to generate a prediction Intention instructions, wherein, when the intention selection module cannot determine which of the plurality of candidate intention instructions has the highest execution order according to the customized logic decision tree, the support vector machine classifier is used to compare the predicted intention instructions The plurality of candidate intention instructions, and then comparing the plurality of candidate intention instructions to find an intention instruction consistent with the predicted intention instruction as the execution intention instruction. The customized intent selection method as described in claim item 9 further includes that the intent selection module extracts features from the user’s service history and personal information to obtain the required multiple features, and then constructs the multiple features After being converted into a feature vector, it is provided to the support vector machine classifier to train the machine learning semantic understanding model. A computer-readable medium, used in a computing device or a computer, is stored with instructions to execute the method for selecting a customized intent as described in any one of Claims 6-10.

Images