CN114550710A - Recognition method, device, storage medium and electronic device for user's dialog intention - Google Patents

Abstract

The invention discloses a method, a device, a storage medium and an electronic device for recognizing a user's dialogue intention. By obtaining a sentence intention recognition result for the current sentence in the current round of dialogue, the sentence intention recognition result includes: a classification name and a classification score; Determine the corresponding intention state coefficient according to the classification name; determine the corresponding confidence level according to the classification score and the intention state coefficient; The dialogue intention corresponding to the current sentence, wherein the parameters involved in the preset intention determination rule include: the historical dialogue intention corresponding to the historical sentence of the current round of dialogue. The present invention can accurately identify the dialogue intention of each sentence in combination with the foregoing dialogue intention, further improve the fluency of the response, and improve the user experience.

Description

Recognition method, device, storage medium and electronic device for user's dialog intention

technical field

The present invention relates to the field of semantic recognition, and in particular, to a method, device, storage medium and electronic device for recognizing a user's dialog intention.

Background technique

The core function of the intelligent response system is to identify the user's intention and respond based on the user's intention. At present, the intelligent response system mainly recognizes the dialogue intention of a single sentence through the natural language recognition module, that is, only the user input of the sentence is classified into the model.

The specific business needs to collect user information and fill in the slot before answering. During the filling process, the system will ask the user, and the sentence input by the user will be recognized by the natural language recognition module again. At this time, the dialogue intention or dialogue intention may not be recognized. Identify the wrong problem, which further leads to problems such as poor response and poor user experience.

SUMMARY OF THE INVENTION

In view of the above-mentioned problems, the present invention provides a method, apparatus, storage medium and electronic device for identifying a user's dialog intention to overcome the above-mentioned problems or at least partially solve the above-mentioned problems.

A first aspect provides a method for identifying a user's dialog intention, comprising:

Obtain the sentence intent recognition result for the current sentence in this round of dialogue, wherein the sentence intent recognition result includes: classification name and classification score;

According to the classification name, determine the corresponding intention state coefficient;

Determine a corresponding confidence level according to the classification score and the intention state coefficient;

Determine the dialogue intention corresponding to the current sentence of the current round of dialogue according to the confidence level and the preset intention determination rule, wherein the parameters involved in the preset intention determination rule include: the history of the current round of dialogue The historical dialogue intent corresponding to the sentence.

With reference to the first aspect, in some optional implementation manners, the determining the corresponding intention state coefficient according to the classification name includes:

According to the classification name, the corresponding intention state coefficient is determined through a preset intention state determination rule, wherein the intention state coefficient matches the classification name.

With reference to the previous implementation, in some optional implementations, according to the classification name, the corresponding intention state coefficient is determined through a preset intention state determination rule, including:

If the category name matches the newly added intent in the intent state determination rule, determine the newly added coefficient in the intent state determination rule as the intent state coefficient, wherein the newly added coefficient is the same as the Corresponding to the newly added intent, the newly added intent indicates that the intent involved in the current sentence has never appeared in the historical dialogue of the current round of calls;

If the category name matches an exclusive intent in the intent state determination rule, determine an exclusive coefficient in the intent state determination rule as the intent state coefficient, where the exclusive coefficient corresponds to the exclusive intent , the confidence of the exclusive intent representing the intent involved in the current sentence is higher than a certain threshold;

If the category name matches the modification intent in the intent state determination rule, determine a modification coefficient in the intent state determination rule as the intent state coefficient, where the modification coefficient corresponds to the modification intent , the modification intent represents that the confidence level of the intent involved in the current sentence is higher than a certain threshold and has no relationship with each intent involved in the historical dialogue of the current round of calls;

If the category name matches the coexistence intent in the intent state determination rule, determine the coexistence coefficient in the intent state determination rule as the intent state coefficient, wherein the coexistence coefficient corresponds to the coexistence intent , the concurrent intent indicates that the intent involved in the current sentence has a strong correlation with at least one intent of the historical dialogue of the current round of conversation;

If the category name matches an invalid intent in the intent state determination rule, determine an invalid coefficient in the intent state determination rule as the intent state coefficient, wherein the invalid coefficient corresponds to the invalid intent , the invalid intent signifies that the intent involved in the current sentence has an explicit intent switch.

With reference to the first aspect, in some optional implementation manners, the determining a corresponding confidence level according to the classification score and the intention state coefficient includes:

According to formula 1: C=log(score _i ×Z)×ratio _i , the confidence level is calculated, wherein the C is the confidence level, the i is the sentence label, and the score _i is the classification score, the Z is a preset scaling factor, and the ratio _i is the intent state coefficient.

With reference to the first aspect, in some optional implementation manners, determining the dialogue intention corresponding to the current sentence of the current round of dialogue according to the confidence level and a preset intention determination rule, including:

comparing the magnitude between the confidence level and a first preset threshold, and comparing the magnitude between the confidence level and a second preset threshold, wherein the first preset threshold is greater than the second preset threshold ;

If the confidence level is not less than the first preset threshold, determining the dialog intent according to a preset first intent determination rule;

If the confidence level is less than the first preset threshold and greater than the second preset threshold, determining the user dialog intent according to a preset second intent determination rule;

If the confidence level is not greater than the second preset threshold, the user's dialog intent is determined according to a preset third intent determination rule.

In conjunction with the first aspect, in some optional embodiments, the method further includes:

According to the dialogue intention corresponding to the current sentence of the current round of dialogue, the historical dialogue intention corresponding to the historical sentence of the current round of dialogue is updated.

In a second aspect, a device for identifying a user's dialog intention, comprising: an intention state management module and a confidence level calculation module;

The intention state management module includes: a receiving unit, a coefficient determination unit, and an intention determination unit;

The receiving unit is configured to execute and obtain a sentence intent recognition result for the current sentence in the current round of dialogue, wherein the sentence intent recognition result includes: a classification name and a classification score;

The coefficient determination unit is configured to determine the corresponding intention state coefficient according to the classification name;

The confidence calculation module is configured to perform determining a corresponding confidence according to the classification score and the intention state coefficient;

The intention determination unit is configured to execute according to the confidence and a preset intention determination rule to determine the dialogue intention corresponding to the current sentence of the current round of dialogue, wherein the preset intention determination rule involves The parameters include: historical dialogue intentions corresponding to the historical sentences of the current round of dialogue.

With reference to the second aspect, in some optional embodiments, the coefficient determination unit includes: a coefficient determination subunit;

The coefficient determination subunit is configured to determine the corresponding intention state coefficient through a preset intention state determination rule according to the classification name, wherein the intention state coefficient matches the classification name.

In a third aspect, a computer-readable storage medium stores a program thereon, and when the program is executed by a processor, implements any one of the above-mentioned methods for recognizing a user's dialog intention.

In a fourth aspect, an electronic device includes at least one processor, and at least one memory and a bus connected to the processor; wherein the processor and the memory communicate with each other through the bus. communication; the processor is configured to invoke the program instructions in the memory to execute the method for recognizing the user's dialog intention described in any one of the above.

With the above technical solutions, the present invention provides a method, device, storage medium and electronic device for identifying user dialogue intentions, which can obtain a sentence intention identification result for the current sentence in the current round of dialogue, wherein the sentence intention identification The results include: classification name and classification score; according to the classification name, determine the corresponding intention state coefficient; determine the corresponding confidence level according to the classification score and the intention state coefficient; determine according to the confidence level and the preset intention A rule is used to determine the dialogue intention corresponding to the current sentence of the current round of dialogue, wherein the parameters involved in the preset intention determination rule include: the historical dialogue intention corresponding to the historical sentence of the current round of dialogue. It can be seen from this that the present invention can accurately identify the dialogue intention of each sentence in combination with the foregoing dialogue intention, further improve the fluency of the response, and improve the user experience.

The above description is only an overview of the technical solutions of the present invention, in order to be able to understand the technical means of the present invention more clearly, it can be implemented according to the content of the description, and in order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand , the following specific embodiments of the present invention are given.

Description of drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are for the purpose of illustrating preferred embodiments only and are not to be considered limiting of the invention. Also, the same components are denoted by the same reference numerals throughout the drawings. In the attached image:

Fig. 1 shows a flowchart of a method for identifying a user's dialog intention provided by the present invention;

2 shows a schematic diagram of an intent state determination rule provided by the present invention;

3 shows a schematic diagram of a first intent determination rule provided by the present invention;

Fig. 4 shows the schematic diagram of the second intention determination rule provided by the present invention;

5 shows a schematic diagram of a third intent determination rule provided by the present invention;

6 shows a schematic structural diagram of a device for identifying user dialog intentions provided by the present invention;

FIG. 7 shows a schematic structural diagram of an electronic device provided by the present invention.

Detailed ways

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure will be more thoroughly understood, and will fully convey the scope of the present disclosure to those skilled in the art.

As shown in FIG. 1 , the present invention provides a method for identifying a user's dialog intention, including: S100, S200, S300 and S400;

S100, obtaining a sentence intent recognition result for the current sentence in the current round of dialogue, wherein the sentence intent recognition result includes: a classification name and a classification score;

Optionally, the category name can represent to a certain extent: the dialogue intention corresponding to the current sentence obtained by the natural language recognition module identifying the sentence, which is not limited in the present invention.

Optionally, the classification score can represent to a certain extent: the credibility of the dialogue intent corresponding to the current sentence obtained by the natural language recognition module for recognizing the sentence. The higher the reliability, the present invention does not limit this.

Optionally, the natural language recognition module of the present invention can recognize a specific sentence to obtain a corresponding sentence intent recognition result, but cannot be combined with the previous text, that is, cannot be combined with the historical sentence in the historical dialogue process of the current round of dialogue to identify the current sentence. The sentence intent recognition results of the words. Therefore, it is necessary to obtain the sentence intent recognition result of the current sentence of the current round of dialogue, so as to further determine the sentence intent recognition result of the current sentence in combination with the historical dialogue intent corresponding to the historical sentence of the current round of dialogue, which can improve the recognition of user dialogues to a certain extent. The intended accuracy is not limited by the present invention.

Optionally, based on the characteristics of providing services through online dialogue, when the user needs the service, it will initiate a dialogue with the subject who implements the present invention. The subject of the invention instructs to end the current round of dialogue. The starting point of the current round of dialogue in the present invention is the moment when the user of the current round of dialogue initiates the dialogue, and the corresponding historical sentences of the current round of dialogue can also be understood as starting from the starting point of the current round of dialogue, to the period when the user inputs the current sentence, All historical sentences input by the user are not limited in the present invention.

Optionally, the present invention does not limit the value range of the classification score, and any feasible manner falls within the protection scope of the present invention. For example, the full score may be 10 points, or the thousand-point system or the hundred-point system may be adopted. For example, the value range of the classification score is: a value range from 0 to 1.

Optionally, the present invention does not specifically limit the division methods of different classification names, any feasible method belongs to the protection scope of the present invention, and any feasible method belongs to the protection scope of the present invention. For example, the category name can include five categories: new category, exclusive category, modified category, concurrent category and invalid category.

The newly added class may specifically include: the user's dialog intention is a new state by default.

The exclusive category can specifically include: (1) the dialogue intention of the core business consulted by the user enters the exclusive state; (2) there is only one exclusive state in the historical dialogue of the current round of calls; (3) the historical dialogue of the current round of calls has exclusive intentions , except for invalid intent, there is no other status, etc.

The modification class can specifically include: (1) When the user's intention may be switched, multiple intentions that may be modified; (2) When there is a modification intention in the historical dialogue of this round of calls, except for the invalid intention, there is no other state intention, etc. Happening.

The coexistence class can specifically include: (1) when the user expresses the scheme-type intention of the same slot; (2) when there is a coexistence intention in the dialogue, except for the invalid intention, there are no other state intentions, etc.

Among them, the scheme-based service refers to collecting user information on demand to fill the slot, and pushing the response scheme to the user after filling the slot. Task-based business refers to collecting user information to fill slots, and after multiple rounds of interaction to fill and confirm all slots, execute the task, and push the results after the task is completed.

The invalid category may specifically include: (1) After the user's intention is switched, all the historical conversation intentions of the current round of calls enter the invalid state; (2) The invalid intention can coexist with any state intention, etc.

Optionally, an invalid intent may be understood as an intent that has no history and a relatively low degree of confidence, which is not limited in the present invention.

Optionally, the present invention does not specifically limit the natural language recognition module, and any feasible manner falls within the protection scope of the present invention. For example, the intent recognition model may be: a semantic recognition model trained based on an RNN deep learning algorithm.

S200. Determine a corresponding intention state coefficient according to the classification name;

Optionally, as mentioned above, different categories correspond to different situations, and can be set in advance according to the corresponding situations of the five categories described above, namely, new categories, exclusive categories, modified categories, coexisting categories, and invalid categories. The corresponding intent state determines the rules.

For example, in optional implementation 1, in combination with the implementation shown in FIG. 1 , in some optional implementations, the S200 includes:

According to the classification name, the corresponding intention state coefficient is determined through a preset intention state determination rule, wherein the intention state coefficient matches the classification name.

Optionally, the intent state determination rule can be set in the form of a table. For example, as shown in FIG. 2 , the intent state coefficient corresponding to the current sentence can be determined through the table shown in FIG. 2 , which is not limited in the present invention.

For example, in optional implementation 2, in combination with optional implementation 1, in some optional implementation manners, according to the classification name, the corresponding intention state coefficient is determined through a preset intention state determination rule, wherein , the intention state coefficient matches the classification name, including: step 1.1, step 1.2, step 1.3, step 1.4 and step 1.5;

Step 1.1. If the category name matches the newly added intent in the intent state determination rule, determine the newly added coefficient in the intent state determination rule as the intent state coefficient, wherein the newly added coefficient Corresponding to the newly added intention, the newly added intention represents that the intention involved in the current sentence has never appeared in the historical dialogue of the current round of calls;

Step 1.2: If the category name matches the exclusive intent in the intent state determination rule, determine the exclusive coefficient in the intent state determination rule as the intent state coefficient, wherein the exclusive coefficient is the same as the intent state coefficient. Corresponding to an exclusive intent, where the exclusive intent represents that the confidence level of the intent involved in the current sentence is higher than a certain threshold;

Step 1.3: If the category name matches the modification intention in the intention state determination rule, determine the modification coefficient in the intention state determination rule as the intention state coefficient, wherein the modification coefficient is the same as the intention state coefficient. Corresponding to the modification intention, the modification intention represents that the confidence level of the intention involved in the current sentence is higher than a certain threshold and has nothing to do with each intention involved in the historical dialogue of the current round of calls;

Step 1.4. If the category name matches the coexistence intention in the intention state determination rule, determine the coexistence coefficient in the intention state determination rule as the intention state coefficient, wherein the coexistence coefficient is the same as the intention state coefficient. Corresponding to coexisting intentions, the coexisting intentions represent that the intention involved in the current sentence has a strong correlation with at least one intention of the historical dialogue of the current round of calls;

Step 1.5: If the category name matches the invalid intent in the intent state determination rule, determine the invalid coefficient in the intent state determination rule as the intent state coefficient, where the invalid coefficient is the same as the intent state coefficient. The invalid intent corresponds to the invalid intent representing that the intent involved in the current sentence has an explicit intent switch.

Optionally, step 1.1, step 1.2, step 1.3, step 1.4 and step 1.5 are in a parallel relationship, and there is no necessary sequence of execution, which is not limited in the present invention.

Optionally, the newly added category, exclusive category, modified category, concurrent category and invalid category mentioned in the present invention may correspond to the newly added intent, exclusive intent, modified intent, concurrent intent and void intent in sequence, and may also be understood to be equivalent in sequence. . That is, on the side of the natural language recognition module, new classes, exclusive classes, modified classes, concurrent classes and invalid classes are used to describe, and on the side of the intent state management module, new intents, exclusive intents, modified intents, concurrent intents and Invalid intentions are described correspondingly for the convenience of distinction, which is not limited in the present invention.

Optionally, the present invention does not specifically limit the sizes of newly added coefficients, exclusive coefficients, modified coefficients, coexisting coefficients and invalid coefficients, which can be set according to actual needs, which is not limited by the present invention. For example, set: new coefficient=2.5, exclusive coefficient=3.5, modification coefficient=2.5, coexistence coefficient=3, and invalid coefficient=2.

S300. Determine a corresponding confidence level according to the classification score and the intention state coefficient;

Optionally, as mentioned above, although the classification score can represent to a certain extent: the credibility of the dialogue intention corresponding to the current sentence obtained by the natural language recognition module recognizing the sentence. However, since the natural language recognition module does not perform recognition in combination with the historical dialogue of the current round, the corresponding confidence degree can be further calculated by the confidence degree calculation module, which is not limited in the present invention.

Optionally, the present invention does not limit the process of calculating the confidence level, and any feasible method falls within the protection scope of the present invention. For example, in optional implementation 3, in combination with the implementation shown in FIG. 1 , in some optional implementations, the S300 includes:

According to formula 1: C=log(score _i ×Z)×ratio _i , the confidence level is calculated, wherein the C is the confidence level, the i is the sentence label, and the score _i is the classification score, the Z is a preset scaling factor, and the ratio _i is the intent state coefficient.

S400. Determine a dialogue intention corresponding to the current sentence of the current round of dialogue according to the confidence level and a preset intention determination rule, wherein the parameters involved in the preset intention determination rule include: the current round of dialogue The historical sentence corresponding to the historical dialogue intent.

Optionally, the present invention may record each historical dialogue intention of the current round of dialogue, so that the intention determination rule in S400 determines the dialogue intention of the current sentence with reference to each historical dialogue intention, which is not limited in the present invention.

Optionally, the preset intent determination rules may record different intent determination rules corresponding to different confidence levels. An intent determination rule can further record different dialogue intents corresponding to different situations in detail, so as to facilitate determining the dialogue intent corresponding to the current sentence of the current round of dialogue, which is not limited in the present invention.

For example, in optional implementation 4, in combination with the implementation shown in FIG. 1 , in some optional implementations, the S400 includes: step 2.1, step 2.2, step 2.3, and step 2.4;

Step 2.1. Compare the size between the confidence level and a first preset threshold, and compare the size between the confidence level and a second preset threshold, wherein the first preset threshold is greater than the second preset threshold preset threshold;

Step 2.2, if the confidence level is not less than the first preset threshold, determine the dialog intent according to a preset first intent determination rule;

Optionally, the present invention does not specifically limit the process of determining the dialog intent by the intent state management module according to a preset first intent determination rule. For details, refer to the first intent determination rule. For example, according to the first intent determination rule shown in FIG. 3 , the dialog intent corresponding to the current sentence can be determined, which is not limited in the present invention.

Optionally, the table shown in FIG. 3 may be a first intent determination rule with a confidence level of not less than 10. Through the first intent determination rule, the dialog intent corresponding to the current sentence can be returned, that is, the "return dialog intent" column shown in FIG. 3 , which is not limited in the present invention.

Step 2.3, if the confidence level is less than the first preset threshold and greater than the second preset threshold, determine the user dialog intent according to a preset second intent determination rule;

Optionally, the present invention does not specifically limit the process of determining the dialog intent by the intent state management module according to a preset second intent determination rule. For details, please refer to the second intent determination rule. For example, according to the second graph determination rule shown in FIG. 4 , the dialog intention corresponding to the current sentence can be determined, which is not limited in the present invention.

Optionally, the table shown in FIG. 4 may be a second intention determination rule with a confidence level of less than 10 and greater than 5. Through the second intent determination rule, the dialog intent corresponding to the current sentence can be returned, that is, the "return dialog intent" column shown in FIG. 4 , which is not limited in the present invention.

Step 2.4: If the confidence level is not greater than the second preset threshold, determine the user dialog intent according to a preset third intent determination rule.

Optionally, the present invention does not specifically limit the process of determining the dialog intent by the intent state management module according to a preset third intent determination rule. For details, refer to the third intent determination rule. For example, according to the third intent determination rule shown in FIG. 5 , the dialog intent corresponding to the current sentence can be determined, which is not limited in the present invention.

Optionally, the table shown in FIG. 5 may be a third intent determination rule with a confidence level not greater than 5. Through the third intent determination rule, the dialog intent corresponding to the current sentence can be returned, that is, the "return dialog intent" column shown in FIG. 5 , which is not limited in the present invention.

Optionally, step 2.2, step 2.3 and step 2.4 are in a parallel relationship, and there is no necessary sequence of execution, which is not limited in the present invention.

Optional implementation 5, in combination with the implementation shown in FIG. 1, in some optional implementations, before the S100, the method further includes: step 3.1, step 3.2, step 3.3, and step 3.4;

Step 3.1. The natural language recognition module in this system invokes a preset intention recognition model to perform intention recognition on the current sentence, thereby determining the sentence intention recognition result corresponding to the current sentence, wherein the intention recognition model is: : The model trained based on the RNN deep learning algorithm;

Step 3.2, the natural language recognition module sends the sentence intention recognition result to the dialogue management module in this system;

Optionally, the dialogue management module can adopt different response methods based on different dialogue intentions: 1. Scheme-based business: collect user information as needed to fill in the slot, and push the response scheme to the user after filling the slot; 2. Task-based Business: Collect user information to fill slots, and after multiple rounds of interaction to fill and confirm all slots, execute the task, and push the result after the task is completed, which is not limited in the present invention.

Step 3.3, the dialogue management module sends the sentence intention recognition result to the dialogue intention decision-making module in this system, so as to call the dialogue intention decision-making module;

Optionally, the dialogue intention decision module may be embedded in the dialogue management module as a sub-module of the dialogue management module, or may be called by the dialogue management module as an independent module, which is not limited in the present invention.

Optionally, the dialog intent decision module may complete the process of receiving the sentence intent recognition result and sending it to the intent state management module, and the subsequent process of pushing the result to the user, which is not limited in the present invention.

Step 3.4, the dialogue intention decision module sends the sentence intention recognition result to the intention state management module in the system to call the intention state management module;

Optionally, the dialogue intention decision module sends the sentence intention recognition result to the intention state management module, which can be understood as the intention state management module obtains the sentence intention recognition result, which is not limited in the present invention.

Optionally, after the intent state management module obtains the sentence intent recognition result, it triggers the intent state management module to execute a corresponding process, including the process of calling the confidence calculation module, which is not limited in the present invention.

Wherein, the intention state management module, the dialogue intention decision module and the confidence calculation module are all modules called internally by the dialogue management module, and the dialogue intention decision module calls the intention state management module, so The intention state management module calls the confidence calculation module.

Optional implementation 6, in conjunction with the implementation shown in FIG. 1 , in some optional implementations, the method further includes:

According to the dialogue intention corresponding to the current sentence of the current round of dialogue, the historical dialogue intention corresponding to the historical sentence of the current round of dialogue is updated.

Optionally, as described above, the preset intent determination rule involves historical dialog intent, that is, the intent state management module determines the current dialog based on the confidence level and the preset intent determination rule. When the dialogue intent corresponding to the current sentence is used, the historical dialogue intent is used. Therefore, the accuracy of the historical dialogue intention will affect the final execution effect of the present invention to a certain extent. Therefore, the historical dialogue intention corresponding to the historical sentence of the current round of dialogue can be further updated according to the dialogue intention corresponding to the current sentence. For details, please refer to Fig. 3. The column of “dialog intent update logic” in FIG. 4 and FIG. 5 is not limited in the present invention.

As shown in FIG. 6 , the present invention provides an apparatus for identifying user dialog intentions, including: an intention state management module 100 and a confidence level calculation module 200;

The intention state management module 100 includes: a receiving unit 110, a coefficient determination unit 120 and an intention determination unit 130;

The receiving unit 110 is configured to execute and obtain a sentence intent recognition result for the current sentence in the current round of dialogue, wherein the sentence intent recognition result includes: a classification name and a classification score;

The coefficient determination unit 120 is configured to determine the corresponding intention state coefficient according to the classification name;

The confidence level calculation module 200 is configured to determine the corresponding confidence level according to the classification score and the intention state coefficient;

The intention determination unit 130 is configured to execute according to the confidence level and a preset intention determination rule to determine the dialogue intention corresponding to the current sentence of the current round of dialogue, wherein the preset intention determination rule involves The parameters include: the historical dialogue intention corresponding to the historical sentence of the current round of dialogue.

With reference to the embodiment shown in FIG. 6 , in some optional embodiments, the coefficient determination unit 120 includes: a coefficient determination subunit;

The coefficient determination subunit is configured to determine the corresponding intention state coefficient through a preset intention state determination rule according to the classification name, wherein the intention state coefficient matches the classification name.

With reference to the previous embodiment, in some optional embodiments, the coefficient determination subunit is specifically configured to execute:

If the category name matches the newly added intent in the intent state determination rule, determine the newly added coefficient in the intent state determination rule as the intent state coefficient, wherein the newly added coefficient is the same as the Corresponding to the newly added intent, the newly added intent indicates that the intent involved in the current sentence has never appeared in the historical dialogue of the current round of calls;

If the category name matches an exclusive intent in the intent state determination rule, determine an exclusive coefficient in the intent state determination rule as the intent state coefficient, where the exclusive coefficient corresponds to the exclusive intent , the confidence of the exclusive intent representing the intent involved in the current sentence is higher than a certain threshold;

If the category name matches the modification intent in the intent state determination rule, determine a modification coefficient in the intent state determination rule as the intent state coefficient, where the modification coefficient corresponds to the modification intent , the modification intent represents that the confidence level of the intent involved in the current sentence is higher than a certain threshold and has no relationship with each intent involved in the historical dialogue of the current round of calls;

If the category name matches the coexistence intent in the intent state determination rule, determine the coexistence coefficient in the intent state determination rule as the intent state coefficient, wherein the coexistence coefficient corresponds to the coexistence intent , the concurrent intent indicates that the intent involved in the current sentence has a strong correlation with at least one intent of the historical dialogue of the current round of conversation;

If the category name matches an invalid intent in the intent state determination rule, determine an invalid coefficient in the intent state determination rule as the intent state coefficient, wherein the invalid coefficient corresponds to the invalid intent , the invalid intent signifies that the intent involved in the current sentence has an explicit intent switch.

With reference to the embodiment shown in FIG. 6, in some optional embodiments, the confidence calculation module 200 is specifically configured to execute:

According to formula 1: C=log(score _i ×Z)×ratio _i , the confidence level is calculated, wherein the C is the confidence level, the i is the sentence label, and the score _i is the classification score, the Z is a preset scaling factor, and the ratio _i is the intent state coefficient.

With reference to the embodiment shown in FIG. 6 , in some optional embodiments, the intention determination unit 130 includes: a comparison subunit, a first determination subunit, a second determination subunit, and a third determination subunit;

The comparison subunit is configured to perform a comparison between the confidence level and a first preset threshold, and a comparison between the confidence level and a second preset threshold, wherein the first preset threshold Setting the threshold value to be greater than the second preset threshold value;

The first determination subunit is configured to execute, if the confidence level is not less than the first preset threshold, determine the dialog intent according to a preset first intent determination rule;

The second determination subunit is configured to execute, if the confidence level is less than the first preset threshold and greater than the second preset threshold, determine the user according to a preset second intention determination rule conversational intent;

The third determination subunit is configured to execute, if the confidence level is not greater than the second preset threshold, determine the user dialog intention according to a preset third intention determination rule.

With reference to the embodiment shown in FIG. 6, in some optional embodiments, the apparatus further includes: a natural language recognition module, a dialogue management module, and a dialogue intention decision module;

The natural language recognition module includes: a calling submodule and a sending subunit;

The calling sub-module is configured to execute, before the intention state management module 100 obtains the sentence intention recognition result output by the natural language recognition module for the current sentence of the current round of dialogue, call the preset intention recognition model to the current sentence. Perform intention recognition, thereby determining the sentence intention recognition result corresponding to the current sentence, wherein the intention recognition model is: a model obtained by training based on an RNN deep learning algorithm;

The sending subunit is configured to execute sending the sentence intent recognition result to the dialogue management module;

The dialogue management module is configured to execute sending the sentence intention recognition result to the dialogue intention decision module, so as to call the dialogue intention decision module;

The dialogue intention decision module is configured to execute sending the sentence intention recognition result to the intention state management module 100, so as to call the intention state management module 100;

The intention state management module 100, the dialogue intention decision module and the confidence calculation module 200 are all modules called internally by the dialogue management module, and the dialogue intention decision module calls the intention state management module 100, the intention state management module 100 calls the confidence calculation module 200.

With reference to the embodiment shown in FIG. 6 , in some optional embodiments, the apparatus further includes: an intention updating unit;

The intention updating unit is configured to update the historical dialogue intention corresponding to the historical sentence of the current round of dialogue according to the dialogue intention corresponding to the current sentence of the current round of dialogue.

The present invention provides a computer-readable storage medium on which a program is stored, and when the program is executed by a processor, implements any one of the above-mentioned methods for identifying a user's dialog intention.

As shown in FIG. 7, the present invention provides an electronic device 70, the electronic device 70 includes at least one processor 701, and at least one memory 702 and a bus 703 connected to the processor 701; wherein, the processing The processor 701 and the memory 702 communicate with each other through the bus 703; the processor 701 is configured to call the program instructions in the memory 702 to execute the method for identifying the user's dialog intention described in any of the above method.

In this application, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that any such relationship exists between these entities or operations. an actual relationship or sequence. Moreover, the terms "comprising", "comprising" or any other variation thereof are intended to encompass a non-exclusive inclusion such that a process, method, article or device comprising a list of elements includes not only those elements, but also includes not explicitly listed or other elements inherent to such a process, method, article or apparatus. Without further limitation, an element qualified by the phrase "comprising a..." does not preclude the presence of additional identical elements in a process, method, article or apparatus that includes the element.

Each embodiment in this specification is described in a related manner, and the same and similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, as for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the partial description of the method embodiment.

The above description of the disclosed embodiments enables any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims (10)

1. a method for identifying a user's dialogue intention, comprising: Obtain the sentence intent recognition result for the current sentence in this round of dialogue, wherein the sentence intent recognition result includes: classification name and classification score; According to the classification name, determine the corresponding intention state coefficient; Determine a corresponding confidence level according to the classification score and the intention state coefficient; Determine the dialogue intention corresponding to the current sentence of the current round of dialogue according to the confidence level and the preset intention determination rule, wherein the parameters involved in the preset intention determination rule include: the history of the current round of dialogue The historical dialogue intent corresponding to the sentence. 2. The method according to claim 1, wherein the determining the corresponding intention state coefficient according to the classification name comprises: According to the classification name, the corresponding intention state coefficient is determined through a preset intention state determination rule, wherein the intention state coefficient matches the classification name. 3 . The method according to claim 2 , wherein, according to the classification name, determining the corresponding intention state coefficient through a preset intention state determination rule, comprising: 3 . If the category name matches the newly added intent in the intent state determination rule, determine the newly added coefficient in the intent state determination rule as the intent state coefficient, wherein the newly added coefficient is the same as the Corresponding to the newly added intent, the newly added intent indicates that the intent involved in the current sentence has never appeared in the historical dialogue of the current round of calls; If the category name matches an exclusive intent in the intent state determination rule, determine an exclusive coefficient in the intent state determination rule as the intent state coefficient, where the exclusive coefficient corresponds to the exclusive intent , the confidence of the exclusive intent representing the intent involved in the current sentence is higher than a certain threshold; If the category name matches the modification intent in the intent state determination rule, determine a modification coefficient in the intent state determination rule as the intent state coefficient, where the modification coefficient corresponds to the modification intent , the modification intention represents that the confidence level of the intention involved in the current sentence is higher than a certain threshold and has nothing to do with the intentions involved in the historical dialogue of the current round of calls; If the category name matches the coexistence intent in the intent state determination rule, determine the coexistence coefficient in the intent state determination rule as the intent state coefficient, wherein the coexistence coefficient corresponds to the coexistence intent , the concurrent intent indicates that the intent involved in the current sentence has a strong correlation with at least one intent of the historical dialogue of the current round of conversation; If the category name matches an invalid intent in the intent state determination rule, determine an invalid coefficient in the intent state determination rule as the intent state coefficient, wherein the invalid coefficient corresponds to the invalid intent , the invalid intent signifies that the intent involved in the current sentence has an explicit intent switch. 4. The method according to claim 1, wherein the determining a corresponding confidence level according to the classification score and the intention state coefficient comprises: According to formula 1: C=log(score _i ×Z)×ratio _i , the confidence level is calculated, wherein the C is the confidence level, the i is the sentence label, and the score _i is the classification score, the Z is a preset scaling factor, and the ratio _i is the intent state coefficient. 5. The method according to claim 1, characterized in that, determining the dialogue intention corresponding to the current sentence of the current round of dialogue according to the confidence level and a preset intention determination rule, comprising: comparing the magnitude between the confidence level and a first preset threshold, and comparing the magnitude between the confidence level and a second preset threshold, wherein the first preset threshold is greater than the second preset threshold ; If the confidence level is not less than the first preset threshold, determining the dialog intent according to a preset first intent determination rule; If the confidence level is less than the first preset threshold and greater than the second preset threshold, determining the user dialog intent according to a preset second intent determination rule; If the confidence level is not greater than the second preset threshold, the user's dialog intent is determined according to a preset third intent determination rule. 6. The method of claim 1, wherein the method further comprises: According to the dialogue intention corresponding to the current sentence of the current round of dialogue, the historical dialogue intention corresponding to the historical sentence of the current round of dialogue is updated. 7. A device for identifying user dialog intentions, comprising: an intention state management module and a confidence level calculation module; The intention state management module includes: a receiving unit, a coefficient determination unit, and an intention determination unit; The receiving unit is configured to execute and obtain a sentence intent recognition result for the current sentence in the current round of dialogue, wherein the sentence intent recognition result includes: a classification name and a classification score; The coefficient determination unit is configured to determine the corresponding intention state coefficient according to the classification name; The confidence calculation module is configured to perform determining a corresponding confidence according to the classification score and the intention state coefficient; The intention determination unit is configured to execute according to the confidence and a preset intention determination rule to determine the dialogue intention corresponding to the current sentence of the current round of dialogue, wherein the preset intention determination rule involves The parameters include: historical dialogue intentions corresponding to the historical sentences of the current round of dialogue. 8. The apparatus according to claim 7, wherein the coefficient determination unit comprises: a coefficient determination subunit; The coefficient determination subunit is configured to determine the corresponding intention state coefficient through a preset intention state determination rule according to the classification name, wherein the intention state coefficient matches the classification name. 9 . A computer-readable storage medium having a program stored thereon, wherein when the program is executed by a processor, the method for recognizing a user's dialog intention according to any one of claims 1 to 6 is implemented. 10 . 10. An electronic device, characterized in that the electronic device comprises at least one processor, and at least one memory and a bus connected to the processor; wherein, the processor and the memory are completed through the bus Mutual communication; the processor is configured to invoke the program instructions in the memory to execute the method for recognizing the user's dialog intention according to any one of claims 1 to 6.

Images