TWI803815B - Interactive natural language processing system and method thereof and computer readable medium - Google Patents

Abstract

An interactive natural language processing system and method thereof for receiving, processing and reacting to text input of a user are provided. When the text input lacks required parameters, the system and the method can ask the user to supplement the parameters. When the text input is ambiguous, the system and the method can ask the user to confirm his or her intention. Therefore, the interactive natural language processing system and the method can still proceed to process the text input when the text input is unclear. The present invention further provides a computer-readable medium for performing the interactive natural language processing method.

Description

Interactive natural language processing system, method and computer readable medium

The present invention relates to a natural language processing technology, and in particular to an interactive natural language processing system, method and computer readable medium.

Users of intelligent voice interaction devices, such as users of smart speakers, often have ambiguous spoken commands. For example, when using a smart device to listen to music, if the user gives an instruction of "Andy Lau", the instruction is vague due to the lack of clear information. Or playing Andy Lau's movies and other possibilities, so it often makes it impossible for the natural language processing system at the back end of the device to understand. Current technology is often unable to understand such ambiguous instructions, so it cannot be processed or processed with poor efficiency and accuracy.

In order to solve the above problems, the present invention provides an interactive natural language processing system, including: an intention recognition module, which is used to determine the corresponding input of the user's first text in the dialogue script. Sentence pattern, to judge whether the first text input has included all the parameters required by the sentence pattern according to the parameter collection setting of the dialogue script, wherein, when the first text input has included all the parameters required by the sentence pattern, then the The intent recognition module triggers the execution of the corresponding function, and when the first text input does not include all the parameters required by the sentence pattern, the intent recognition module triggers parameter collection and feedback to make the user provide a second text input; And a parameter collection module, which is used to collect parameters required by the sentence pattern and lacking in the first text input according to the second text input, so as to trigger the execution of the corresponding function.

In one embodiment, the interactive natural language processing system further includes: a fuzzy semantic questioning module, which is used to answer questions based on key words in the first text input when there is no corresponding sentence pattern in the dialogue script for the first text input. Words trigger fuzzy semantic questions, so that the user can provide a third text input to confirm whether the user's intention is the corresponding intention of the keyword in the dialogue script

The present invention also provides an interactive natural language processing method, including: judging the sentence pattern corresponding to the user's first text input in the dialogue script, and judging whether the first text input includes the sentence pattern according to the parameter collection setting of the dialogue script. All the parameters required by the sentence pattern, wherein, when the first text input has included all the parameters required by the sentence pattern, the execution of the corresponding function will be triggered, and when the first text input does not include all the parameters required by the sentence pattern, Then trigger parameter collection to ask the user to provide a second text input; and collect parameters required by the sentence pattern and lacking in the first text input from the second text input to trigger the execution of the corresponding function.

In one embodiment, the interactive natural language processing method further includes: if the first text input has no corresponding sentence pattern in the dialogue script, triggering a vague semantic question according to the keyword in the first text input, so that The user provides a third text input to confirm whether the user's intention is the intention corresponding to the keyword in the dialogue script.

The present invention further provides a computer-readable medium, which is used in a computing device or a computer and stores instructions to execute the above-mentioned interactive natural language processing method.

The interactive natural language processing system, method and computer-readable medium of the present invention process the text input of the user's command, directly process the user's clear command, use the interactive method to collect the parameters that the user's command lacks, and use the interactive method Confirm the intent of ambiguous commands to quickly and accurately resolve problems where commands are difficult to understand and cannot be executed.

10: Interactive natural language processing system

12: Input interface

14: Output interface

40: Decision Trees

42: root node

44,46: node

100: Pre-processing module

110:Symbol processing module

120: word segmentation module

130: Parameter labeling module

140:Dialog management module

200: Intent recognition module

210: Synonym restoration module

220: Similarity calculation module

230: Sentence pattern competition module

300: parameter collection module

310: Parameter identification module

320: Parameter reply module

400: Fuzzy Semantic Response Module

500: Dialogue script editing module

S21 ~ S33: method steps

FIG. 1 is a block diagram of an interactive natural language processing system according to an embodiment of the present invention.

FIG. 2 is a flowchart of an interactive natural language processing method according to an embodiment of the present invention.

Fig. 3 is a schematic diagram of a decision tree according to an embodiment of the present invention.

The implementation of the present invention will be described below through specific specific examples. Those with ordinary knowledge in the technical field can easily understand other advantages and effects of the present invention from the contents disclosed in this specification.

FIG. 1 is a block diagram of an interactive natural language processing system 10 according to an embodiment of the present invention. The interactive natural language processing system 10 is used for processing and responding to user's instructions. In one embodiment, the interactive natural language processing system 10 includes an input interface 12, a pre-processing module 100, an intention recognition module 200, a parameter collection module 300, a fuzzy semantic questioning module 400, and a dialog script The editing module 500 and the output interface 14; the pre-processing module 100 includes a symbol processing module 110, a word segmentation module 120, a parameter labeling module 130 and a dialogue management module 140; the intent recognition module 200 includes a synonym restoration module 210, similarity calculation module 220 and sentence pattern competition module 230; parameter collection module 300 includes parameter identification module 310 and parameter answering module 320; input interface 12, symbol processing module 110, word segmentation module 120, The parameter labeling module 130 and the dialog management module 140 are sequentially connected by communication; the synonym restoration module 210 and the parameter recognition module 310 are all communicatively connected to the dialog management module 140; the synonym restoration module 210, the similarity calculation module 220 and the The sentence pattern competition module 230 is sequentially connected by communication; the parameter reply module 320 is connected by communication to the parameter identification module 310; the fuzzy semantic reply module 400 is connected by communication with the sentence pattern competition module 230 and the parameter reply module 320; the output interface 14 Communicatively connect the fuzzy semantic questioning module 400; the synonym restoration module 210, the similarity calculation module 220, the parameter questioning module 320 and the fuzzy semantic questioning module 400 are all connected to the dialog script editing module 500 by communication.

In one embodiment, the input interface 12 is used to receive the text input of the user's instruction; the pre-processing module 100 is used to perform symbol filtering on the text input to filter out redundant symbols, then perform word segmentation processing, and then divide the word Label the parameters in the results; the intent recognition module 200 is used to restore synonyms to the text input through the pre-processing module 100, and then perform similarity calculations with various sentence patterns in the dialogue script to select multiple candidate sentences After the competition, the sentence pattern competition will be carried out. After the competition, the most suitable sentence pattern for text input and the corresponding semantic meaning will be left as the basis for responding to the user; the parameter collection module 300 is used for lack of parameters in the text input of the command When performing parameter identification, collection and questioning, and then complete the understanding of the intention of the text input; the fuzzy semantic questioning module 400 is used to ask the user when the semantic meaning cannot be judged in the text input, so as to confirm the user's intention; use The personnel of the interactive natural language processing system 10 can edit the dialogue script through the dialogue script editing module 500, which is the basis for the interactive natural language processing system 10 to process the text input of the user, and edit the dialogue script The semantic understanding ability of the interactive natural language processing system 10 can be improved, and the interactive natural language processing system 10 can be customized for different applications; the output interface 14 is used to provide text output in response to the text input of the user's instruction.

Each module in Figure 1 can be software, hardware or firmware; if it is hardware, it can be a processing unit, processor, computer or server with data processing and computing capabilities; if it is software or firmware , may include instructions executable by a processing unit, processor, computer or server.

FIG. 2 is a flowchart of an interactive natural language processing method according to an embodiment of the present invention. The interactive natural language processing method is executed by the interactive natural language processing system 10 . The functions and effects of each module of the interactive natural language processing system 10 will be described below with reference to the flowchart in FIG. 2 .

First, in step S21 , the input interface 12 receives the text input of the user's instruction.

In step S22, the symbol processing module 110 performs symbol filtering on the text input to filter out symbols irrelevant to semantics and parameters, such as brackets, punctuation marks, and blanks, so that the back-end module will not be affected by these symbols.

In step S23, the word segmentation module 120 performs word segmentation processing on the text input to decompose the text input into words with the smallest lexical meaning. For example, "Grape" cannot be divided into "泉" and "糖", because the two characters have no independent meaning after being separated; and "Da'an Forest Park" can be divided into "Da'an", "Forest" and "Park", because These words still have their own meanings after they are separated. In one embodiment, the word segmentation technology is divided into dictionary type, statistical type and hybrid type. Preferably, this embodiment uses the hybrid type technology for word segmentation processing of text input.

Next, in step S24 , the parameter labeling module 130 identifies known parameters in the above words, and labels the parameters in the words to indicate the category of the parameters. In an embodiment, the category of tags (ie, the category of parameters) may include person names, place names, organizations and countries. For example, "Daan Forest The result of the word segmentation of Park is "Da'an", "Forest" and "Park". After the parameters are identified, "Da'an" will be labeled as a place name. The parameter labeling module 130 can also expand the categories of its labels according to user needs. In one embodiment, the parameter identification and labeling techniques are divided into dictionary, statistical and hybrid. Preferably, this embodiment uses the dictionary technique for parameter identification and labeling.

Next, in step S25, the dialog management module 140 determines whether the interactive natural language processing system 10 is in the parameter collection state. The initial interactive natural language processing system 10 is not in the parameter collection state. If the user's text input lacks parameters or the meaning is vague, the parameter collection state will be triggered. If the interactive natural language processing system 10 is not in the parameter collection state, then the flow process enters step S26 to carry out semantic recognition of text input; if the interactive natural language processing system 10 is in the parameter collection state, then the flow process enters step S32 to perform parameter collect.

In step S26 , the synonym restoration module 210 restores synonyms to the text input through the preprocessing module 100 , so as to restore the synonyms in the above words to the original words. In one embodiment, the synonym recovery module 210 may include a thesaurus storing synonyms and their corresponding original words, such as the thesaurus in Table 1. If the user's text input contains the three words "Amit", "De" and "Song" after word segmentation, after the synonyms are restored, the two synonyms "Amit" and "Song" will be restored is the corresponding original word, so it will become the three words "Zhang Huimei", "de" and "song".

Next, in step S27, the similarity calculation module 220 calculates the similarity between the text input restored by the synonym restoration module 210 and each sentence pattern in the dialogue script, and selects sentences whose similarity is higher than the preset threshold value. type, wherein each sentence type is a set containing multiple words. For example, Table 2 includes four sentence patterns in the dialogue script and their corresponding semantics. The semantic meaning of this embodiment is expressed in Chinese, but not limited thereto. In another embodiment, the semantic meaning can be expressed in other languages or identification codes.

In this embodiment, the Pearson correlation coefficient is used to represent the similarity between text input and sentence patterns. The similarity calculation module 220 may include a dictionary, which at least includes all words of all sentence types in the dialogue script. The similarity calculation module 220 will vectorize the text input of the synonym restoration module 210 according to the dictionary to generate a vector x , and vectorize each sentence pattern in the dialogue script according to the dictionary to generate each sentence pattern vector y , and then use the following formulas 1 to 6 to calculate the Pearson correlation coefficient p between the text input and each sentence pattern. If the Pearson correlation coefficient p of any sentence pattern is greater than the preset threshold value, it is considered to be used The possible intention of the reader (hereinafter such a sentence pattern is referred to as a candidate sentence pattern).

In the above formula, n is the number of words in the dictionary, x and y are n- dimensional vectors, and each dimension corresponds to a word in the dictionary. If the text input through the synonym restoration module 210 includes a certain word in the dictionary, the value of the dimension x _i corresponding to the word in the vector x is 1; if the text input through the synonym restoration module 210 does not include in the dictionary A certain word of , then the value of the dimension x _i corresponding to the word in the vector x is 0. If a certain sentence pattern in the dialogue script includes a certain word in the dictionary, the value of the dimension y _i corresponding to the word in the vector y of the sentence pattern is 1; if a certain sentence pattern in the dialogue script does not include the word in the dictionary , then the value of the dimension y _i corresponding to the word in the vector y of the sentence type is 0. u _x is the average of the n dimensions in the vector x . u _y is the average value of n dimensions in vector y .

Then, in step S28, when the similarity calculation module 220 finds out a plurality of candidate sentence patterns, the sentence pattern competition module 230 will use a decision tree to compete with the plurality of candidate sentence patterns, and leave the most suitable sentence pattern . For example, the decision tree 40 shown in FIG. 3, the decision tree 40 is a binary tree, wherein the root node 42 corresponds to a condition, and there are two branches "yes" and "no" below it, corresponding to text input and/or a sentence type meets this condition. There is another node 44 under the branch "yes", and the node 44 corresponds to another condition, and there are also two branches "yes" and "no" under it, corresponding to text input and/or whether the sentence pattern meets the condition respectively, according to And so on. For example, the conditions corresponding to the nodes 42, 44 and 46 are respectively "the similarity between the text input and the sentence pattern is greater than the threshold value", "the text input does not use synonyms" and "the text input does not trigger the parameter collection state". Sentence competition module 230 can use text input and dialogue Each sentence pattern in the script begins to compare from the root node 42 of the decision tree 40, if the text input and/or the sentence pattern does not meet the conditions of the root node 42, then the comparison ends; if the text input and/or the sentence pattern meet the The condition of the root node 42 is then continued down, compared with the condition of the node 44, and so on. Finally, the candidate sentence pattern and its corresponding semantics that have gone through the most branches of "yes" are selected as the corresponding sentence pattern and semantics of the user's text input in the dialogue script.

Next, in step S29 , the sentence pattern competition module 230 determines whether the parameter collection state of the interactive natural language processing system 10 needs to be triggered. Specifically, the sentence pattern competition module 230 checks whether the parameters identified in step S24 in the user's text input include all the parameters required for the parameter collection setting of the sentence pattern corresponding to the text input in the dialogue script. If yes, it means that the user's text input has complete semantics, and the parameter collection state does not need to be triggered; if not, it means that the user's text input does not have complete semantics due to lack of parameters, and the sentence pattern competition module 230 triggers parameter collection state, therefore, when processing the user's next text input, the flow will enter step S32 from step S25 instead of entering step S26. The parameter collection setting in the dialogue script includes the category of each parameter required by each sentence pattern, and the sentence pattern competition module 230 will compare the category of the parameter required by the sentence pattern corresponding to the text input and the category of the parameter in the text input , to determine whether the recognized parameters in the text input include all parameters required by the sentence pattern corresponding to the text input.

Next, in step S30, the vague semantic questioning module 400 determines whether to trigger the vague semantic questioning. If the similarity calculation module 220 finds at least one candidate sentence pattern in step S27 (the similarity of at least one sentence pattern is greater than the threshold value), it means that the user's text input has a clear semantic meaning, and there is no need to trigger the vague semantic question. Conversely, if the similarity calculation module 220 does not find candidate sentence patterns in step S27 (the similarity of each sentence pattern does not exceed the threshold value), it means that the semantic meaning of the user's text input is vague, then the vague semantic answering module 400 Trigger vague semantic questions. In this case, the user's next text Word input will be treated as a parameter, therefore, the fuzzy semantic answering module 400 will also trigger the parameter collection state.

Next, in step S31 , the output interface 14 provides text output in response to the user's text input. If the previous sentence pattern competition module 230 triggers the parameter collection state in step S29, the text output provided by the output interface 14 is a parameter collection reply, so that the user can provide parameters. Otherwise, if the fuzzy semantic question module 400 triggers the fuzzy semantic question in step S30 before, the text output provided by the output interface 14 is a fuzzy semantic question to make the user confirm his intention. If neither the parameter collection state nor the fuzzy semantic question is triggered before, the text output provided by the output interface 14 is the processed and normalized user instruction.

In another embodiment, the output interface 14 of step S31 can execute the function corresponding to the semantic meaning of the text input, such as playing the music or song specified by the text input, when the parameter collection state is not triggered and the fuzzy semantic question is not triggered. or stories in response to user text input.

On the other hand, in step S32, the parameter identification module 310 will identify the parameters in the text input processed by the pre-processing module 100, if the text input this time contains parameters, and the category of the parameters conforms to the last time If there is a type of parameter lacking in the text input by the user, the parameter recognition module 310 collects the parameter to supplement the text input by the user last time to make its semantics more complete.

Next, in step S33, the parameter query module 320 judges whether all the parameters lacking in the last text input have been collected. The parameter query module 320 can learn whether there are still missing parameters in the last text input according to the parameter collection settings in the dialogue script. If all the missing parameters have been collected, then the parameter answering module 320 ends the parameter collection state of the interactive natural language processing system 10. Therefore, when processing the user's next text input, the flow will enter step S26 from step S25, and No Go to step S32. If the missing parameters have not been collected yet, the parameter query module 320 triggers a parameter collection query.

Next, in step S30, the vague semantic questioning module 400 still determines whether to trigger the vague semantic questioning. In step S31, the output interface 14 provides text output in response to the user's text input. If the previous parameter query module 320 triggers the parameter collection query in step S33, the text output provided by the output interface 14 is the parameter collection query, so that the user can provide parameters. Otherwise, if the fuzzy semantic question module 400 triggers the fuzzy semantic question in step S30 before, the text output provided by the output interface 14 is a fuzzy semantic question to make the user confirm his intention. If neither the parameter collection question nor the fuzzy semantic question is triggered before, the text output provided by the output interface 14 is the processed and normalized user instruction.

In another embodiment, the output interface 14 of step S31 can execute the function corresponding to the semantic meaning of the text input to respond to the text input of the user when the parameter collection state and the fuzzy semantic question are not triggered before.

In another embodiment, the input interface 12 may include a speech-to-text function to convert the user's voice commands into text input, and the output interface 14 may include a text-to-speech function to collect the above-mentioned parameters and return them to the question. And fuzzy semantic questions are converted from text to voice and then output.

The following embodiments are examples of how the interactive natural language processing system 10 recognizes user intent and collects parameters.

First, in step S21, the text input received by the input interface 12 is "I want to listen to the music of the Three Little Pigs". In step S22, the symbol processing module 110 performs symbol filtering on the text input. In step S23, the word segmentation module 120 divides the text input into six words [I, Want, Listen, The Three Little Pigs, , Music]. In step S24, the parameter labeling module 130 labels the parameters in the words, and the result For [I, want, listen, three little pigs@story@music@idiom, of, music], where the word [three little pigs] is recognized as a parameter, and three categories are labeled as story, music and idiom labels. In step S25, since the parameter collection state is not triggered, the dialog management module 140 calls the intention identification module 200 for subsequent processing, in other words, the process enters step S26 from step S25.

In step S26 , the synonym restoration module 210 restores synonyms for [I, want, listen, three little pigs@story@music@idiom, of,music] based on the thesaurus shown in Table 1. Because there are no synonyms in the text input, the result is unchanged. In step S27, the similarity calculation module 220 vectorizes each sentence pattern in the text input and dialogue script according to the dictionary, wherein the dialogue script is as shown in the following Table 3, and the sentence patterns in each column in Table 3 correspond to the same column The semantic meaning, parameter collection setting and fuzzy semantic question setting in Table 3, and the intention in Table 3 refers to the user's intention, which can cover semantic meaning and sentence pattern. The vectors of each sentence pattern and text input and the dictionary on which it is based are shown in the following table 4. The dictionary is [playing, Zhang Huimei, de, song, The Three Little Pigs, listening, music, Pinocchio, stories, idioms] , a total of 10 words.

Then, the similarity calculation module 220 uses the aforementioned formulas 1 to 6 to calculate the variance, standard deviation, covariance and Pearson correlation coefficient of each sentence pattern and text input, as shown in Table 5 below.

Since the similarity calculation module 220 will leave candidate sentence patterns whose similarity (i.e. Pearson correlation coefficient) is higher than the threshold value, if the threshold value is set to 0.5, then only one sentence pattern will become a candidate sentence pattern, that is, the similarity is 0.612372 second sentence pattern in dialogue script.

Then, in step S28, since there is only one candidate sentence pattern, there will be no sentence pattern competition, and it can be directly determined that the user's intention is to play music, but according to the dialogue script in Table 3, it can be found that the intention lacks the parameter of the category being a singer , so the parameter collection state will be triggered in step S29.

Next, in step S30, since the user's intention is already clear, the ambiguous semantic question will not be triggered. Finally, in step S31, the text output of the output interface 14 collects the question "whose song do you want to listen to" for the parameters of the playing music semantics. After the user sees or hears the interactive natural language processing system 10 replying "whose song do you want to listen to", he can input "Zhang Huimei" to the interactive natural language processing system 10 as the next text input, and then the interactive natural language processing system The system 10 will execute the method flow of FIG. 2 again.

First, in step S21, the text input received by the input interface 12 is "Zhang Huimei". In step S22, the symbol processing module 110 performs symbol filtering on the text input. In step S23, the word segmentation module 120 performs word segmentation processing on the filtered text input. Zhang Huimei is a complete singer name, so it will not be further subdivided. The word segmentation result is [Zhang Huimei]. In step S24, the parameter labeling module 130 labels the parameters in the word segmentation result, and the result is [张惠妹@歌词], wherein the word [张惠妹] is recognized as a parameter, and the category is labeled singer. In step S25, since the parameter collection state of the interactive natural language processing system 10 has been triggered, the dialogue management module 140 calls the parameter collection module 300 for subsequent processing. In other words, the process enters step S32 from step S25.

Next, in step S32, the parameter recognition module 310 knows according to the dialogue script that the parameters whose category is a singer need to be collected at the moment, and the processing result of the pre-processing module 100 is [张惠妹@歌手], So the parameter identification module 310 will collect parameters [张惠妹@歌词]. In step S33, the parameter questioning module 320 knows from the dialog script that all missing parameters have been collected. In other words, the semantic meaning of the last text input has been completed, so the parameter collection questioning will not be triggered, but the parameter collection state will end. The next flow will enter step S26 from step S25.

Next, in step S30, since all missing parameters have been collected and the semantic meaning of the text input is clear, the vague semantic question will not be triggered. Finally, in step S31, the text output provided by the output interface 14 is the user intention of [play music (listen to Zhang Huimei's music)]. In another embodiment, the output interface 14 can play Zhang Huimei's song in response to the user's text input.

The following embodiment is an example of how the interactive natural language processing system 10 performs fuzzy semantic questioning.

First, in step S21, the text input received by the input interface 12 is "Zhang Huimei". In step S22, the symbol processing module 110 performs symbol filtering on the text input. In step S23, the word segmentation module 120 performs word segmentation processing on the filtered text input. Zhang Huimei is a complete singer name, so it will not be further subdivided. The word segmentation result is [Zhang Huimei]. In step S24, the parameter labeling module 130 labels the parameters in the word segmentation result, and the result is [张惠妹@歌词], wherein the word [张惠妹] is recognized as a parameter, and the category is labeled singer. In step S25, since the parameter collection state of the interactive natural language processing system 10 is not triggered, the dialog management module 140 calls the intention recognition module 200 for subsequent processing. In other words, the process enters step S26 from step S25.

Next, in step S26, please refer to Table 1, since there is no synonym in the text input, the synonym recovery module 210 will not replace it. In step S27, the similarity calculation module 220 vectorizes each sentence pattern in the text input and dialogue script according to the dictionary, wherein the dictionary of this embodiment And dialogue scripts are all the same as the previous embodiment, the dialogue scripts are shown in Table 3, and the vectors of each sentence pattern and text input and the dictionary on which they are based are shown in Table 6 below.

Then, the similarity calculation module 220 uses the aforementioned formulas 1 to 6 to calculate the variance, standard deviation, covariance and Pearson correlation coefficient of each sentence pattern and text input, as shown in Table 7 below.

Then, the similarity calculation module 220 checks whether there is a candidate sentence pattern whose similarity (Pearson correlation coefficient) exceeds the threshold value. The threshold value of this embodiment is also 0.5. Obviously, there is no candidate sentence whose similarity exceeds the threshold value. type. Therefore, in steps S28 and S29, since there is no candidate sentence pattern, the sentence pattern competition module 230 will not have any effect.

Then, in step S30, since the user's text input has no intention, the fuzzy semantic answer module 400 will inquire whether there is a keyword of the fuzzy semantic answer in the dialogue script in the text input. The semantic questioning module 400 will find that the word [Zhang Huimei] is a keyword, so it triggers the fuzzy semantic questioning and parameter collection status, so the text output in step S31 will be the questioning sentence corresponding to the keyword in the dialogue script "you Do you want to listen to her song?" to allow users to confirm whether their intent is the intent corresponding to the keyword in the dialogue script. The user only needs to answer "yes" as the text input of the next process, and the text input "yes" will be regarded as a parameter, whereby the interactive natural language processing system 10 can determine that the user's intention is determined by the keyword The corresponding intent, which is the first intent in the dialogue script. In step S32 of the next process, the parameter identification module 310 can collect the parameter and trigger a preset response. For example, in one embodiment, the output interface 14 can be triggered to play Zhang Huimei's song in response to the user's text input.

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.

To sum up, the interactive natural language processing system, method and computer readable medium proposed by the present invention use the following three behaviors to understand the user's intention. First, when the user directly conveys a clear instruction, the system and method proposed by the present invention can correctly interpret the instruction, Then find the correct user intent. Second, when the user transmits an instruction lacking parameters, the system and method proposed by the present invention will perform an interactive question to make the user provide the lacking parameters. Thirdly, when the user sends an ambiguous instruction, the system and method proposed by the present invention will ask the keyword in the instruction to confirm the user's intention. The above three behaviors can be used to correctly understand the user's instructions to quickly and accurately solve the problem of incomprehensible instructions, thereby improving user satisfaction.

The above-mentioned embodiments are only illustrative to illustrate the principles and functions of the present invention, and are not intended to limit the present invention. Anyone with ordinary knowledge in the technical field 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 described later.

10: Interactive natural language processing system

12: Input interface

14: Output interface

100: Pre-processing module

110:Symbol processing module

120: word segmentation module

130: Parameter labeling module

140:Dialog management module

200: Intent recognition module

210: Synonym restoration module

220: Similarity calculation module

230: Sentence pattern competition module

300: parameter collection module

310: Parameter identification module

320: Parameter reply module

400: Fuzzy Semantic Response Module

500: Dialogue script editing module

Claims (10)

An interactive natural language processing system, including: an intention recognition module, used to calculate the similarity between the user's first text input and each sentence pattern in the dialogue script, so as to judge the first text input in the Whether there is a corresponding sentence pattern or the corresponding sentence pattern in the dialogue script, wherein, if the similarity of no sentence pattern in the dialogue script is greater than the threshold value, then the first text input has no corresponding sentence pattern in the dialogue script , and the intent identification module is reused to determine whether the first text input includes the sentence pattern required by the first text input according to the parameter collection setting of the dialog script when the first text input has a corresponding sentence pattern in the dialogue script All the parameters, wherein, when the first text input includes all the parameters required by the sentence pattern, the intention recognition module triggers the execution of the corresponding function, and when the first text input does not include all the parameters required by the sentence pattern , then the intent recognition module triggers the parameter collection response, so that the user provides a second text input; the parameter collection module is used to collect the sentence pattern needed according to the second text input and the first text input lacks parameter, to trigger the execution of the corresponding function; and a fuzzy semantic answering module, used to inquire whether the first text input exists in the dialogue script when the first text input has no corresponding sentence pattern in the dialogue script keyword, and when the keyword exists in the first text input, output the answer sentence corresponding to the keyword to the user, so that the user can provide a third text input, and then confirm the user's intention Whether it is the intent corresponding to the keyword in the dialogue script, wherein the question sentence does not include any sentence pattern in the dialogue script. The interactive natural language processing system as described in Claim 1 further includes: a pre-processing module, used to decompose the first text input into plural words with the minimum lexical meaning, for the intention recognition module to judge the second A word is input before the corresponding sentence pattern in the dialogue script, and the synonyms in the words are restored to the corresponding original words. The interactive natural language processing system as described in Claim 1, further comprising: a pre-processing module, used to decompose the first text input into plural words with minimum lexical meaning, for the intent recognition module to use the words according to the calculating the similarity between the first text input and the sentence patterns in the dialogue script. The interactive natural language processing system as described in claim 3, wherein, if the similarity of the only sentence pattern in the dialogue script is greater than the threshold value, then the sentence pattern with the similarity greater than the threshold value is the The sentence pattern corresponding to the first text input in the dialogue script, if the similarity of plural sentence patterns in the dialogue script is greater than the threshold value, the intent recognition module uses a decision tree to select one of the sentence patterns Or as the first character input corresponding to the sentence pattern in the dialogue script. The interactive natural language processing system as described in Claim 1 further includes: a pre-processing module, which is used to decompose the first text input and the second text input into plural words with the smallest lexical meaning, and recognize these The parameters in the word and the category of each parameter, wherein, the intention recognition module compares the category of the parameters required by the sentence pattern in the parameter collection setting and the category of the parameters of the first text input to determine the Whether the first text input has included all parameters required by the sentence pattern, and the parameter collection module is based on the category of the parameters of the second text input, collects the sentence pattern required by the second text input and the second text input A missing parameter for text input. The interactive natural language processing system as described in claim 5, wherein the parameter collection module is reused to judge the sentence according to the parameter collection setting after collecting the parameters required by the sentence pattern and lacking in the first text input Whether the parameters required by the type and lacking in the first text input have all been collected, if all have been collected, the parameter collection module will trigger the execution of the corresponding function, otherwise, the parameter collection module will trigger the parameter collection response . The interactive natural language processing system as described in claim 1, wherein the parameter collection module is used to confirm that the user's intention is the intention corresponding to the keyword in the dialogue script when the third text input , to trigger the execution of the corresponding function. An interactive natural language processing method, including: calculating the similarity between the user's first text input and each sentence pattern in the dialogue script, so as to judge whether the first text input has a corresponding correspondence in the dialogue script according to the similarity sentence pattern or the corresponding sentence pattern, wherein, if the similarity of no sentence pattern in the dialogue script is greater than the threshold value, then the first text input has no corresponding sentence pattern in the dialogue script; if the first text input is in When there is a corresponding sentence pattern in the dialogue script, according to the parameter collection setting of the dialogue script, it is judged whether the first text input has included all the parameters required by the sentence pattern; if the first text input has included the sentence pattern required All the parameters of the sentence pattern will trigger the execution of the corresponding function; if the first text input does not include all the parameters required by the sentence pattern, the parameter collection response will be triggered so that the user can provide the second text input; from the second text Input the parameters needed to collect the sentence patterns and the lack of the first text input to trigger the execution of the corresponding function; and if the first text input does not have a corresponding sentence pattern in the dialogue script, query whether the first text input There is a keyword in the dialogue script, and when the keyword exists in the first text input, output the answer sentence corresponding to the keyword to the user, so that the user can provide a third text input, and then confirm Whether the user's intention is the intention corresponding to the keyword in the dialogue script, wherein the question sentence does not include any sentence pattern in the dialogue script. The interactive natural language processing method as described in Claim 8, further comprising: Decomposing the first text input into plural words with minimum lexical meaning, so as to restore synonyms in these words to corresponding original words before judging the sentence pattern corresponding to the first text input in the dialogue script. A computer-readable medium, used in a computing device or a computer, stores instructions to execute the interactive natural language processing method described in any one of Claims 8-9.

Images