Dialogue state recognition method and device, computer device and storage medium

Abstract

The application relates to a training method and device of a dialogue state recognition model, a computer device, a storage medium and a computer program product. The method comprises the following steps: determining a plurality of state feature words corresponding to a training sample dialogue set; the training sample dialogue set comprises a plurality of sample dialogue sentences; a sample dialogue vocabulary matrix for the training sample dialogue set is constructed; the sample dialogue vocabulary matrix comprises a state feature vector and a state category label corresponding to each sample dialogue sentence; the state feature vector is used for representing the occurrence of each state feature word in the sample dialogue sentence; according to the sample dialogue vocabulary matrix, prior probability information corresponding to each state category label in the training sample dialogue set and class conditional probability information corresponding to each sample dialogue sentence in the training sample dialogue set are determined; and according to the prior probability information and the class conditional probability information, a dialogue state recognition model is constructed. The method can effectively improve the dialogue state recognition efficiency in a dialogue scene.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a dialogue state recognition method, apparatus, computer device, storage medium, and computer program product. Background Technology

[0002] Currently, dialogue state recognition methods are widely used in real life. Dialogue state analysis is an important research direction in the field of natural language, involving theories and methods from fields such as linguistics, statistics, psychology, and artificial intelligence.

[0003] Dialogue state analysis can be applied to service interactions. However, dialogue state recognition based on call interactions in related technologies often relies on subjective judgment. Moreover, it is impossible to make real-time judgments and reminders about the dialogue state between the two parties during the service process. Instead, customer service personnel can only be evaluated afterward by sampling and listening to recordings. This results in a lag in the judgment of service quality, and it cannot promptly detect any negative states that customers may have during communication, nor can it promptly remind customer service personnel to provide high-quality service.

[0004] Therefore, the relevant technologies have the problem of being unable to efficiently identify the dialogue status during call interactions. Summary of the Invention

[0005] Therefore, it is necessary to provide a training method, apparatus, computer device, computer-readable storage medium, and computer program product for a dialogue state recognition model that can improve the efficiency of dialogue state recognition in dialogue scenarios, in order to address the above-mentioned technical problems.

[0006] Firstly, this application provides a method for training a dialogue state recognition model. The method includes:

[0007] Determine several state feature words corresponding to the training sample dialogue set; the training sample dialogue set includes several sample dialogue statements.

[0008] Construct a sample dialogue vocabulary matrix for the training sample dialogue set; the sample dialogue vocabulary matrix includes a state feature vector and a state category label corresponding to each sample dialogue statement; the state feature vector is used to characterize the occurrence of each state feature word in the sample dialogue statement;

[0009] Based on the sample dialogue vocabulary matrix, determine the prior probability information of each state category label in the training sample dialogue set, and the class conditional probability information of each sample dialogue statement in the training sample dialogue set.

[0010] A dialogue state recognition model is constructed based on the prior probability information and the class conditional probability information; the dialogue state recognition model is used to identify the dialogue state category corresponding to the current dialogue statement.

[0011] In one embodiment, the state category label includes a positive class label; the step of constructing a dialogue state recognition model based on the prior probability and the class conditional probability information includes:

[0012] Based on the prior probability information and the class conditional probability information corresponding to each sample dialogue statement, the posterior probability information of each sample dialogue statement belonging to the positive class label is determined.

[0013] The dialogue state recognition model is constructed based on the posterior probability information corresponding to each sample dialogue statement.

[0014] In one embodiment, the state category label further includes a negative category label; determining the prior probability information corresponding to each state category label in the training sample dialogue set based on the sample dialogue vocabulary matrix includes:

[0015] Based on the sample dialogue vocabulary matrix, determine the first prior probability information corresponding to the positive class label in the training sample dialogue set;

[0016] as well as,

[0017] Based on the sample dialogue vocabulary matrix, determine the second prior probability information corresponding to the negative class label in the training sample dialogue set;

[0018] Based on the first prior probability information and the second prior probability information, the prior probability information corresponding to each state category label in the training sample dialogue set is obtained.

[0019] In one embodiment, the state category label further includes a negative class label; determining the class conditional probability information corresponding to each of the sample dialogue statements in the training sample dialogue set includes:

[0020] For any of the sample dialogue statements, the first type of conditional probability information of the sample dialogue statement for the positive class label is determined based on the sample dialogue vocabulary matrix.

[0021] as well as,

[0022] Based on the sample dialogue vocabulary matrix, determine the second type of conditional probability information of any sample dialogue statement for the negative class label;

[0023] Based on the first type of conditional probability information and the second type of conditional probability information, the class conditional probability information corresponding to any sample dialogue statement is obtained.

[0024] In one embodiment, the state category label further includes a negative category label; the prior probability information includes first prior probability information corresponding to the positive category label and second prior probability information corresponding to the negative category label; the class conditional probability information includes first class conditional probability information of the sample dialogue statement for the positive category label and second class conditional probability information of the sample dialogue statement for the negative category label.

[0025] The step of determining the posterior probability information of each sample dialogue statement belonging to the positive class label based on the prior probability information and the class-conditional probability information corresponding to each sample dialogue statement includes:

[0026] For any of the sample dialogue statements, the first joint probability information corresponding to any of the sample dialogue statements is obtained based on the first prior probability information and the corresponding first type conditional probability information.

[0027] And, based on the second prior probability information and the corresponding second type of conditional probability information, the second joint probability information corresponding to any sample dialogue statement is obtained;

[0028] Based on the first joint probability information and the corresponding second joint probability information of any sample dialogue statement, the posterior probability information of any sample dialogue statement belonging to the positive class label is obtained.

[0029] In one embodiment, obtaining the posterior probability information of any sample dialogue statement belonging to the positive class label based on the first joint probability information and the corresponding second joint probability information includes:

[0030] Based on the first joint probability information and the second joint probability information corresponding to any sample dialogue statement, the total probability information corresponding to any sample dialogue statement is obtained.

[0031] Based on the first joint probability information and the corresponding total probability information corresponding to any sample dialogue statement, the posterior probability information of any sample dialogue statement belonging to the positive class label is obtained.

[0032] Secondly, this application provides a dialogue state recognition method. The method includes:

[0033] In the current dialogue scenario, the current dialogue statement is input into the dialogue state recognition model to obtain the dialogue state category corresponding to the current dialogue statement; the dialogue state recognition model is obtained according to the training method of the above-mentioned dialogue state recognition model.

[0034] Display a status category prompt window; the status category prompt window includes the dialogue status category corresponding to the current dialogue statement.

[0035] Thirdly, this application also provides a training device for a dialogue state recognition model. The device includes:

[0036] The feature word determination module is used to determine several state feature words corresponding to the training sample dialogue set; the training sample dialogue set includes several sample dialogue statements.

[0037] A matrix construction module is used to construct a sample dialogue vocabulary matrix for the training sample dialogue set; the sample dialogue vocabulary matrix includes a state feature vector and a state category label corresponding to each sample dialogue statement; the state feature vector is used to characterize the occurrence of each state feature word in the sample dialogue statement;

[0038] The information determination module is used to determine, based on the sample dialogue vocabulary matrix, the prior probability information corresponding to each state category label in the training sample dialogue set, and the class conditional probability information corresponding to each sample dialogue statement in the training sample dialogue set.

[0039] The model building module is used to construct a dialogue state recognition model based on the prior probability information and the class conditional probability information; the dialogue state recognition model is used to identify the dialogue state category corresponding to the current dialogue statement.

[0040] Fourthly, this application also provides a dialogue state recognition device. The device includes:

[0041] The input module is used to input the current dialogue statement into the dialogue state recognition model in the current dialogue scenario to obtain the dialogue state category corresponding to the current dialogue statement; the dialogue state recognition model is obtained according to the training method of the dialogue state recognition model described above.

[0042] The display module is used to display a status category prompt window; the status category prompt window includes the dialogue status category corresponding to the current dialogue statement.

[0043] Fifthly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a training method for a dialogue state recognition model as described in the first aspect or any embodiment of the first aspect, or a dialogue state recognition method as described in the second aspect.

[0044] Sixthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements a training method for a dialogue state recognition model as described in the first aspect or any embodiment of the first aspect, or a dialogue state recognition method as described in the second aspect.

[0045] In a seventh aspect, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements a training method for a dialogue state recognition model as described in the first aspect or any embodiment of the first aspect, or a dialogue state recognition method as described in the second aspect.

[0046] The training method, apparatus, computer equipment, storage medium, and computer program product of the aforementioned dialogue state recognition model determine several state feature words corresponding to the training sample dialogue set. The training sample dialogue set includes several sample dialogue statements. A sample dialogue vocabulary matrix is ​​constructed for the training sample dialogue set. The sample dialogue vocabulary matrix includes state feature vectors and state category labels corresponding to each sample dialogue statement. The state feature vectors are used to characterize the occurrence of each state feature word in the sample dialogue statement. Then, based on the sample dialogue vocabulary matrix, the prior probability information corresponding to each state category label in the training sample dialogue set, as well as the class conditional probability information corresponding to each sample dialogue statement in the training sample dialogue set, are determined. Thus, based on the prior probability information corresponding to each state category label and the class conditional probability information corresponding to each sample dialogue statement, a state recognition model for state recognition of each dialogue statement can be constructed. Therefore, in real-time dialogue scenarios, the dialogue state recognition model can identify the dialogue state category corresponding to the current dialogue statement in real time, realizing state recognition of each dialogue statement in the dialogue scenario without having to analyze the dialogue state by checking and listening to the recording after the dialogue ends, thereby effectively improving the efficiency of dialogue state recognition in dialogue scenarios. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating a training method for a dialogue state recognition model in one embodiment.

[0048] Figure 2 This is a schematic diagram of a sample dialogue vocabulary matrix for a training sample dialogue set in one embodiment.

[0049] Figure 3 This is a schematic diagram of the state category labels corresponding to each sample dialogue statement in one embodiment;

[0050] Figure 4This is a flowchart illustrating a training method for a dialogue state recognition model in another embodiment;

[0051] Figure 5 This is a flowchart illustrating a method for establishing a dialogue state recognition model and predicting dialogue state categories in real time, as described in one embodiment.

[0052] Figure 6 This is a flowchart illustrating a dialogue state recognition method in one embodiment;

[0053] Figure 7 This is a structural block diagram of a training device for a dialogue state recognition model in one embodiment;

[0054] Figure 8 This is a structural block diagram of a dialogue state recognition device in one embodiment;

[0055] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0057] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0058] In one embodiment, such as Figure 1 As shown, a training method for a dialogue state recognition model is provided. This embodiment illustrates the application of this method to a server, which can be a standalone server or a server cluster consisting of multiple servers. It is understood that this method can also be applied to a terminal, and further to a system including both a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0059] Step S110: Determine several state feature words corresponding to the training sample dialogue set.

[0060] The training sample dialogue set includes several sample dialogue statements.

[0061] Among them, state feature words are words related to the dialogue state in the sample dialogue statements.

[0062] The dialogue status can include, but is not limited to, dialogue mood and dialogue attitude.

[0063] The sample dialogue statements can be from real-world conversations in different business scenarios. For example, the sample dialogue statements could be from a real conversation between a customer service representative and a customer in a banking business scenario.

[0064] In practice, the server can obtain a training sample dialogue set that includes several sample dialogue statements, and extract several words related to the dialogue state from the sample dialogue statements as several state feature words corresponding to the training sample dialogue set.

[0065] Specifically, the server can segment the sample dialogue statements, remove stop words, delete some words that are irrelevant to the dialogue state, and finally perform deduplication to obtain the deduplicated state feature words.

[0066] Among them, state feature words may include, but are not limited to, adjectives, verbs, idioms and other words in the sample dialogue sentences.

[0067] For example, if the training sample dialogue set has three sample dialogue sentences, the first sentence is: "I am very happy, I won an award today"; the second sentence is: "Haha, I am very satisfied."; and the third sentence is: "I am going to complain about you", then the extracted state feature words include "happy", "won an award", "satisfied" and "complain".

[0068] In the process of acquiring the training sample dialogue set, the server can first acquire call recordings of real dialogue scenarios under different business scenarios, and then use Automatic Speech Recognition (ASR) to transcribe the recordings in real time, converting the content of the call speech into text-based sentences as sample dialogue sentences, so as to obtain a training sample dialogue set composed of several sample dialogue sentences.

[0069] Step S120: Construct a sample dialogue vocabulary matrix for the training sample dialogue set.

[0070] The sample dialogue vocabulary matrix includes the state feature vector and state category label corresponding to each sample dialogue statement.

[0071] Among them, the state feature vector is used to characterize the occurrence of each state feature word in the sample dialogue statement.

[0072] In a specific implementation, the server can determine the state feature vector corresponding to each sample dialogue statement based on the occurrence of each state feature word in the sample dialogue statement, so that the state feature vector can represent the occurrence of each state feature word in the sample dialogue statement; in this way, the server can construct a sample dialogue vocabulary matrix for the above training sample dialogue set based on the state feature vector corresponding to each sample dialogue statement and the state category label labeled for each sample dialogue statement, so that the sample dialogue vocabulary matrix includes the state feature vector and state category label corresponding to each sample dialogue statement.

[0073] Specifically, the sample dialogue vocabulary matrix can take the form shown below:

[0074]

[0075] Where, x i,j This indicates the occurrence of the j-th state feature word in the i-th sample dialogue statement; y i Let represent the state category label corresponding to the i-th sample dialogue statement. Here, n is the total number of state feature words corresponding to the training sample dialogue set; m is the total number of sample dialogue statements in the training sample dialogue set; i is greater than or equal to 1 and less than or equal to m; (x i,1 x i,2 x i,3 …x i,n ) can represent the state feature vector corresponding to the i-th sample dialogue statement.

[0076] In the sample dialogue, the occurrence of each state feature word can be represented by "0" and "1". "0" indicates that the state feature word does not appear in the sample dialogue; "1" indicates that the state feature word appears in the sample dialogue.

[0077] For example, continuing from the previous example, the sample dialogue vocabulary matrix for the training sample dialogue set containing the three sample dialogue statements mentioned above is as follows: Figure 2 As shown.

[0078] Step S130: Based on the sample dialogue vocabulary matrix, determine the prior probability information of each state category label in the training sample dialogue set, and the class conditional probability information of each sample dialogue statement in the training sample dialogue set.

[0079] The prior probability information includes prior probability values.

[0080] Among them, class conditional probability information includes class conditional probability values.

[0081] In practice, the server can determine the prior probability information of each state category label in the training sample dialogue set, as well as the class conditional probability information of each sample dialogue statement in the training sample dialogue set, based on the state feature vector and the corresponding state category label of each sample dialogue statement in the sample dialogue vocabulary matrix.

[0082] Step S140: Construct a dialogue state recognition model based on prior probability information and class conditional probability information.

[0083] Among them, the dialogue state recognition model is used to identify the dialogue state category corresponding to the current dialogue statement.

[0084] In practice, the server can construct a Naive Bayes-based dialogue state recognition model based on prior probability information and class-conditional probability information. This model allows for real-time identification of the dialogue state category corresponding to the current dialogue statement within the current dialogue context.

[0085] In the training method of the above-mentioned dialogue state recognition model, several state feature words corresponding to the training sample dialogue set are determined. The training sample dialogue set includes several sample dialogue sentences. A sample dialogue vocabulary matrix is ​​constructed for the training sample dialogue set. The sample dialogue vocabulary matrix includes state feature vectors and state category labels corresponding to each sample dialogue sentence. The state feature vectors are used to characterize the occurrence of each state feature word in the sample dialogue sentence. Then, based on the sample dialogue vocabulary matrix, the prior probability information corresponding to each state category label in the training sample dialogue set and the class conditional probability information corresponding to each sample dialogue sentence in the training sample dialogue set are determined. Thus, based on the prior probability information corresponding to each state category label and the class conditional probability information corresponding to each sample dialogue sentence, a state recognition model for state recognition of each dialogue sentence can be constructed. Therefore, in real-time dialogue scenarios, the dialogue state recognition model can identify the dialogue state category corresponding to the current dialogue sentence in real time, realizing state recognition of each dialogue sentence in the dialogue scenario. It eliminates the need to analyze the dialogue state by checking and listening to the recording after the dialogue ends, thereby effectively improving the efficiency of dialogue state recognition in dialogue scenarios.

[0086] In one embodiment, the state category label includes a positive class label; the dialogue state recognition model is constructed based on prior probability and class conditional probability information, including: determining the posterior probability information of each sample dialogue statement belonging to the positive class label based on prior probability information and class conditional probability information corresponding to each sample dialogue statement; and constructing the dialogue state recognition model based on the posterior probability information corresponding to each sample dialogue statement.

[0087] The posterior probability information includes the posterior probability value.

[0088] In the specific implementation, the state category label includes the positive class label; in the process of constructing the dialogue state recognition model based on the prior probability and class conditional probability information, the server can determine the posterior probability information of each sample dialogue statement belonging to the positive class label based on the prior probability information and the class conditional probability information corresponding to each sample dialogue statement. Then, the server can construct the dialogue state recognition model based on the posterior probability information corresponding to each sample dialogue statement.

[0089] The technical solution of this embodiment can accurately determine the posterior probability information of each sample dialogue statement belonging to the positive class label by using prior probability information and class conditional probability information corresponding to each sample dialogue statement. Thus, by constructing a dialogue state recognition model based on the posterior probability information corresponding to each sample dialogue statement, the accuracy of dialogue state recognition of the dialogue state recognition model can be improved.

[0090] In one embodiment, the state category label further includes a negative category label; determining the prior probability information corresponding to each state category label in the training sample dialogue set based on the sample dialogue vocabulary matrix includes: determining the first prior probability information corresponding to the positive category label in the training sample dialogue set based on the sample dialogue vocabulary matrix; and determining the second prior probability information corresponding to the negative category label in the training sample dialogue set based on the sample dialogue vocabulary matrix; and obtaining the prior probability information corresponding to each state category label in the training sample dialogue set based on the first prior probability information and the second prior probability information.

[0091] The status category labels include both positive and negative labels.

[0092] The first prior probability information includes the first prior probability value.

[0093] The second prior probability information includes the second prior probability value.

[0094] Among them, the state category labels for each sample dialogue statement can be labeled according to the established dialogue state category standards.

[0095] Examples of the established dialogue state category standards are shown in Table 1 below:

[0096] Table 1 Examples of Dialogue Status Category Standards

[0097]

[0098] During the model building process, most dialogue statements in real-world dialogue scenarios belong to the neutral class label. Therefore, only dialogue statements with positive and negative class labels can be selected from real-world dialogue scenarios as sample dialogue statements to obtain the training sample set.

[0099] For example, continuing from the previous example, the state category labels corresponding to each sample dialogue statement can be obtained through the dialogue state category criteria established above, such as... Figure 3 As shown, the first sample dialogue contains the feature word "happy," which directly confirms that the first sample dialogue belongs to the positive category. Specifically, as... Figure 3 As shown, the first and second sample dialogue statements belong to the positive category, while the third sample dialogue statement belongs to the negative category.

[0100] In the specific implementation, based on the principle of the Naive Bayes classifier, it is necessary to determine P(ci), which means determining the probability that a sentence belongs to class ci. In this embodiment, class ci is equivalent to the state category label, and P(ci) is equivalent to the prior probability value of each state category label in the training sample dialogue set. Since the state category labels include negative and positive class labels, the prior probability information corresponding to each state category label includes the first prior probability information corresponding to the positive class label and the second prior probability information corresponding to the negative class label; the prior probability value corresponding to each state category label includes the first prior probability value P(c1) corresponding to the positive class label c1 and the second prior probability value P(c2) corresponding to the negative class label c2.

[0101] Therefore, in the process of determining the prior probability information corresponding to each state category label in the training sample dialogue set based on the sample dialogue vocabulary matrix, the server can determine the proportion of sample dialogue statements belonging to the positive category label in the training sample dialogue set based on the state category label corresponding to each sample dialogue statement in the sample dialogue vocabulary matrix, thereby obtaining the first prior probability information corresponding to the positive category label in the training sample dialogue set; and determine the proportion of sample dialogue statements belonging to the negative category label in the training sample dialogue set, thereby obtaining the second prior probability information corresponding to the negative category label in the training sample dialogue set.

[0102] Thus, based on the first prior probability information corresponding to the positive class label and the second prior probability information corresponding to the negative class label, the prior probability information corresponding to each state category label in the training sample dialogue set can be obtained.

[0103] The technical solution of this embodiment determines the first prior probability information of positive class labels in the training sample dialogue set based on the sample dialogue vocabulary matrix; and determines the second prior probability information of negative class labels in the training sample dialogue set based on the sample dialogue vocabulary matrix; and obtains the prior probability information of each state category label in the training sample dialogue set based on the first prior probability information and the second prior probability information.

[0104] Thus, since the sample dialogue vocabulary matrix includes the state category label corresponding to each sample dialogue statement, the proportion of sample dialogue statements belonging to each state category label in the training sample dialogue set can be determined based on the sample dialogue vocabulary matrix. This allows for the accurate determination of the first prior probability information corresponding to the positive category label in the training sample dialogue set, as well as the second prior probability information corresponding to the negative category label in the training sample dialogue set.

[0105] In one embodiment, the state category label further includes a negative category label; determining the class-conditional probability information corresponding to each sample dialogue statement in the training sample dialogue set includes: for any sample dialogue statement, determining the first class-conditional probability information of any sample dialogue statement for the positive category label based on the sample dialogue vocabulary matrix; and determining the second class-conditional probability information of any sample dialogue statement for the negative category label based on the sample dialogue vocabulary matrix; and obtaining the class-conditional probability information corresponding to any sample dialogue statement based on the first class-conditional probability information and the second class-conditional probability information.

[0106] Among them, the first type of conditional probability information includes the first type of conditional probability value.

[0107] Among them, the second type of conditional probability information includes the second type of conditional probability value.

[0108] In the specific implementation, based on the principle of the Naive Bayes classifier, it is necessary to determine P(w1, w2, ..., wn|ci). The meaning of P(w1, w2, ..., wn|ci) is that if a sentence belongs to class ci, then the sentence has the probability values ​​of features w1, w2, ..., wn. In this embodiment, the features w1, w2, ..., wn are equivalent to state feature words. For example, continuing from the previous example, w1 = happy, w2 = won an award, w3 = satisfied, w4 = complain; class ci is equivalent to state category label; P(w1, w2, ..., wn|ci) is equivalent to the class conditional probability value corresponding to the sample dialogue statement. Since the state category label includes negative and positive class labels, the class conditional probability information corresponding to the sample dialogue statement in the training sample dialogue set includes the first class conditional probability information for positive class labels and the second class conditional probability information for negative class labels. The class-conditional probability values ​​corresponding to the sample dialogue statements include the first class-conditional probability value P(w1, w2, ..., wn|c1) for the positive class label c1, and the second class-conditional probability value P(w1, w2, ..., wn|c2) for the negative class label c2.

[0109] Therefore, in the process of determining the class-conditional probability information corresponding to each sample dialogue statement in the training sample dialogue set, for any sample dialogue statement, the server can determine the first class-conditional probability information of that sample dialogue statement for the positive class label based on the state feature vector and state category label corresponding to each sample dialogue statement in the sample dialogue vocabulary matrix, and determine the second class-conditional probability information of that sample dialogue statement for the negative class label based on the state feature vector and state category label corresponding to each sample dialogue statement in the sample dialogue vocabulary matrix.

[0110] In this way, the server can obtain the class conditional probability information corresponding to any sample dialogue statement based on the first type of conditional probability information and the corresponding second type of conditional probability information.

[0111] The technical solution of this embodiment determines, for any sample dialogue statement, a first type of conditional probability information for a positive class label based on the sample dialogue vocabulary matrix; and a second type of conditional probability information for a negative class label based on the sample dialogue vocabulary matrix. In this way, the class conditional probability information corresponding to any sample dialogue statement can be accurately obtained based on the first and second type of conditional probability information. Thus, a dialogue state recognition model based on the Naive Bayes algorithm can be constructed based on the class conditional probability information to more accurately identify the dialogue state category corresponding to the dialogue statement.

[0112] In one embodiment, the state category label further includes a negative category label; the prior probability information includes a first prior probability information corresponding to the positive category label and a second prior probability information corresponding to the negative category label; the class conditional probability information includes a first class conditional probability information of the sample dialogue statement for the positive category label and a second class conditional probability information of the sample dialogue statement for the negative category label.

[0113] Specifically, based on prior probability information and class-conditional probability information corresponding to each sample dialogue statement, the posterior probability information of each sample dialogue statement belonging to the positive class label is determined, including: for any sample dialogue statement, obtaining the first joint probability information corresponding to any sample dialogue statement based on the first prior probability information and the corresponding first class-conditional probability information; obtaining the second joint probability information corresponding to any sample dialogue statement based on the second prior probability information and the corresponding second class-conditional probability information; and obtaining the posterior probability information of any sample dialogue statement belonging to the positive class label based on the first joint probability information and the corresponding second joint probability information.

[0114] The first joint probability information includes the first joint probability value.

[0115] The second joint probability information includes the second joint probability value.

[0116] In the specific implementation, based on the Naive Bayes classifier principle, P(ci|w1,w2,…,wn)=(P(w1,w2,…,wn|ci)×P(ci)) / (P(w1,w2,…,wn)); where P(ci|w1,w2,…,wn) means: the probability value of a sentence with features w1, w2,…,wn belonging to class ci. In this embodiment, the features w1, w2,…,wn are equivalent to state feature words, and class ci is equivalent to state category label. When class ci is a positive class label, P(ci|w1,w2,…,wn) is equivalent to the posterior probability value of the sample dialogue sentence belonging to the positive class label. (P(w1,w2,…,wn)) means: the probability value of a sentence having features w1, w2,…,wn. And according to the law of total probability, P(B)=P(B|A)P(A)+P(B|A')P(A'), (P(w1,w2,…,wn))=[P(w1,w2,…,wn|c1)×P(c1)+P(w1,w2,…,wn|c2)×P(c2)].

[0117] Thus, in the process of determining the posterior probability information of each sample dialogue statement belonging to the positive class label based on the prior probability information and the class-conditional probability information corresponding to each sample dialogue statement, the server can obtain the first joint probability information corresponding to any sample dialogue statement based on the first prior probability information and the first class-conditional probability information corresponding to that sample dialogue statement. Specifically, the service can obtain the first joint probability value P(w1,w2,…,wn|c1)×P(c1) corresponding to any sample dialogue statement by multiplying the first prior probability value P(c1) and the first class-conditional probability value P(w1,w2,…,wn|c1) corresponding to that sample dialogue statement.

[0118] Furthermore, the server can obtain the second joint probability information corresponding to any sample dialogue statement based on the second prior probability information and the second type conditional probability information corresponding to any sample dialogue statement. Specifically, the service can obtain the second joint probability value P(w1,w2,…,wn|c2)×P(c2) corresponding to any sample dialogue statement by multiplying the second prior probability value P(c2) and the second type conditional probability value P(w1,w2,…,wn|c2) corresponding to any sample dialogue statement.

[0119] Thus, the server can obtain the posterior probability information of any sample dialogue statement belonging to the positive class label based on the first joint probability information and the corresponding second joint probability information. Specifically, the server can obtain the posterior probability value P(ci|w1,w2,…,wn|c1)×P(c1) and the corresponding second joint probability value P(w1,w2,…,wn|c2)×P(c2) of any sample dialogue statement.

[0120] The technical solution of this embodiment obtains a first joint probability information corresponding to any sample dialogue statement based on a first prior probability information and a corresponding first type of conditional probability information; and obtains a second joint probability information corresponding to any sample dialogue statement based on a second prior probability information and a corresponding second type of conditional probability information; and obtains a posterior probability information of any sample dialogue statement belonging to the positive class label based on the first joint probability information and the corresponding second joint probability information. In this way, the posterior probability information of each sample dialogue statement in the training sample set belonging to the positive class label can be determined. By using the posterior probability information of each sample dialogue statement belonging to the positive class label, a dialogue state recognition model is constructed, realizing the dialogue state category corresponding to the dialogue statement based on the Naive Bayes principle, which can effectively improve the recognition accuracy of dialogue state category.

[0121] In one embodiment, obtaining the posterior probability information of any sample dialogue statement belonging to the positive class label based on the first joint probability information and the corresponding second joint probability information includes: obtaining the total probability information corresponding to any sample dialogue statement based on the first joint probability information and the corresponding second joint probability information; and obtaining the posterior probability information of any sample dialogue statement belonging to the positive class label based on the first joint probability information and the corresponding total probability information.

[0122] Among them, the total probability information includes the total probability value.

[0123] In the specific implementation, according to the total probability formula P(B)=P(B|A)P(A)+P(B|A')P(A'), then (P(w1,w2,…,wn))=[P(w1,w2,…,wn|c1)×P(c1)+P(w1,w2,…,wn|c2)×P(c2)]. Thus, in the process of obtaining the posterior probability information of any sample dialogue statement belonging to the positive class label based on the first joint probability information and the corresponding second joint probability information, the server can obtain the total probability information corresponding to any sample dialogue statement based on the first joint probability information and the corresponding second joint probability information. Specifically, the server can obtain the corresponding total probability value P(w1,w2,…,wn) based on the sum of the corresponding first joint probability value P(w1,w2,…,wn|c1)×P(c1) and the corresponding second joint probability value P(w1,w2,…,wn|c2)×P(c2).

[0124] Then, the server can obtain the posterior probability information of any sample dialogue statement belonging to the positive class label based on the first joint probability information and the corresponding total probability information. Specifically, based on the Naive Bayes classifier principle, P(ci|w1,w2,…,wn)=(P(w1,w2,…,wn|ci)×P(ci)) / (P(w1,w2,…,wn)), when ci=c1, that is, P(c1|w1,w2,…,wn) is equivalent to the posterior probability value of the sample dialogue statement belonging to the positive class label. Then, based on the quotient between the first joint probability value and the corresponding total probability value of any sample dialogue statement, the posterior probability value of any sample dialogue statement belonging to the positive class label can be determined.

[0125] The technical solution of this embodiment can accurately obtain the total probability information corresponding to any sample dialogue statement by using the first joint probability information and the corresponding second joint probability information. Then, based on the first joint probability information and the corresponding total probability information, the posterior probability information of any sample dialogue statement belonging to the positive class label can be accurately obtained. In this way, a dialogue state recognition model based on the Naive Bayes principle can be constructed by using the posterior probability information of each sample dialogue statement belonging to the positive class label, so as to accurately identify the dialogue state corresponding to each dialogue statement according to the dialogue state recognition model.

[0126] In another embodiment, such as Figure 4 As shown, a training method for a dialogue state recognition model is provided. Taking the application of this method to a server as an example, the method includes the following steps:

[0127] Step S410: Determine several state feature words corresponding to the training sample dialogue set.

[0128] Step S420: Construct a sample dialogue vocabulary matrix for the training sample dialogue set.

[0129] Step S430: Based on the sample dialogue vocabulary matrix, determine the first prior probability information corresponding to the positive class label in the training sample dialogue set, and the second prior probability information corresponding to the negative class label in the training sample dialogue set.

[0130] Step S440: Based on the sample dialogue vocabulary matrix, determine the first type of conditional probability information for any sample dialogue statement with respect to positive class labels, and the second type of conditional probability information for any sample dialogue statement with respect to negative class labels.

[0131] Step S450: Based on the first prior probability information, the second prior probability information, the first type of conditional probability information, and the corresponding second type of conditional probability information, determine the posterior probability information of each sample dialogue statement belonging to the positive class label.

[0132] Step S460: Construct a dialogue state recognition model based on the posterior probability information corresponding to the dialogue statements of each sample.

[0133] It should be noted that the specific limitations of the above steps can be found in the specific limitations of the training method for a dialogue state recognition model described above.

[0134] For the ease of understanding of those skilled in the art, Figure 5 This document provides a flowchart illustrating the process of establishing a dialogue state recognition model and predicting dialogue state categories in real time. For example... Figure 5 As shown, it includes the following steps:

[0135] First, the untrained dialogue state recognition model is structurally initialized. Then, positive and negative samples are read from the training sample dialogue set; positive samples are dialogue statements belonging to the positive category, and negative samples are dialogue statements belonging to the negative category. Next, the sample dialogue statements are segmented, stop words are removed, and some words irrelevant to the dialogue state are deleted. Finally, deduplication is performed to obtain deduplicated state feature words, and a positive / negative dictionary is added to the deduplicated state feature words to construct a state feature word set. Then, a dialogue state recognition model is generated based on the state feature word set and the sample dialogue statements. Finally, the dialogue statements are classified using the dialogue state recognition model. Specifically, the prior probabilities corresponding to positive and negative samples are calculated first, followed by the class-conditional probability. Finally, the posterior probability is calculated and normalized to give a 0-1 probability value for being judged as a positive dialogue state. The final dialogue state classification is obtained by dividing the state classification probability interval, and the final recognition result and probability value are returned, ending the process.

[0136] In the process of training the dialogue state recognition model using a training sample set, the training sample set can be iterated: Naive Bayes-based dialogue state analysis classifies dialogue statements by calculating probability values, and incremental iterative training is performed on the generated training set to reduce the final error rate.

[0137] This requires defining the probability intervals for each state category label. Based on the actual business scenario (e.g., a banking scenario) and the set of dialogue statements, the probability intervals are defined according to the following criteria. These intervals can be adjusted depending on the specific data. An example of the probability interval definition criteria is shown in Table 2 below.

[0138] Table 2. Criteria for Classifying State Probability

[0139]

[0140] The model's performance can also be evaluated. Specifically, a confusion matrix is ​​constructed based on the prediction performance on the prediction set, as shown in Table 3 below:

[0141] Table 3 Confusion Matrix

[0142]

[0143] Based on this, accuracy can be calculated. Since the focus in dialogue state recognition in real-world dialogue scenarios is on the accuracy of recognizing positive and negative dialogue states, weights are added to the accuracy calculation process.

[0144]

[0145] in Here, Accuracy refers to the accuracy rate.

[0146] It should be noted that the specific limitations of the above steps can be found in the specific limitations of the training method for a dialogue state recognition model described above.

[0147] In one embodiment, such as Figure 6 As shown, a dialogue state recognition method is provided. Taking the application of this method to a server as an example, the method includes the following steps:

[0148] Step S610: In the current dialogue scenario, input the current dialogue statement into the dialogue state recognition model to obtain the dialogue state category corresponding to the current dialogue statement.

[0149] The dialogue state recognition model is obtained according to the training method of the dialogue state recognition model as described above.

[0150] In practice, the server can acquire the voice of the current dialogue scenario in real time, convert the content of the voice into the current dialogue statement in text form, input it into the trained dialogue state recognition model, obtain the probability value of the current dialogue statement belonging to the positive class label, determine the interval in which the probability value of the current dialogue statement falls according to the divided state classification probability interval, and determine the dialogue state category corresponding to the current dialogue according to the state category label corresponding to the above interval.

[0151] Step S620: Display the status category prompt window.

[0152] In practice, the server can display a status category prompt window, which includes the dialogue status category corresponding to the current dialogue statement.

[0153] In some embodiments, a system can be provided based on the trained dialogue state recognition model, enabling real-time prediction and application in various scenarios. For example, during real-time customer service, a pop-up window can display the predicted dialogue state categories of the customer and customer service representatives, providing real-time reminders to the customer service personnel. Simultaneously, operations managers can monitor the service status and quality of their subordinate customer service representatives in real time, allowing for timely adjustments and optimizations. This can create intelligent application scenarios such as "service anomaly warnings," providing digital support for improving service quality and efficiency and risk control. It also encourages customer service representatives to adjust their service status accordingly, and in case of emergency service situations, it can link with on-site operations managers for timely intervention, continuously advancing service management.

[0154] The technical solution of this embodiment, in the current dialogue scenario, inputs the current dialogue statement into a dialogue state recognition model to obtain the dialogue state category corresponding to the current dialogue statement; the dialogue state recognition model is obtained according to the training method of the dialogue state recognition model as described above; a state category prompt window is displayed; the state category prompt window includes the dialogue state category corresponding to the current dialogue statement; thus, in a real-time dialogue scenario, the dialogue state category corresponding to the current dialogue statement can be identified in real time through the dialogue state recognition model, realizing the state recognition and display of each dialogue statement in the dialogue scenario, without the need to analyze the dialogue state by checking and listening to the recording after the dialogue ends, thereby effectively improving the efficiency of dialogue state recognition in the dialogue scenario.

[0155] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0156] Based on the same inventive concept, this application also provides a training apparatus for a dialogue state recognition model to implement the training method for the dialogue state recognition model described above. The solution provided by this apparatus is similar to the implementation described in the above method. Therefore, the specific limitations of the one or more training apparatus embodiments for the dialogue state recognition model provided below can be found in the limitations of the training method for the dialogue state recognition model described above, and will not be repeated here.

[0157] In one embodiment, such as Figure 7 As shown, a training device for a dialogue state recognition model is provided, comprising: a feature word determination module 710, a matrix construction module 720, an information determination module 730, and a model construction module 740, wherein:

[0158] The feature word determination module 710 is used to determine several state feature words corresponding to the training sample dialogue set; the training sample dialogue set includes several sample dialogue statements.

[0159] The matrix construction module 720 is used to construct a sample dialogue vocabulary matrix for the training sample dialogue set; the sample dialogue vocabulary matrix includes a state feature vector and a state category label corresponding to each of the sample dialogue statements; the state feature vector is used to characterize the occurrence of each of the state feature words in the sample dialogue statements.

[0160] The information determination module 730 is used to determine, based on the sample dialogue vocabulary matrix, the prior probability information corresponding to each state category label in the training sample dialogue set, and the class conditional probability information corresponding to each sample dialogue statement in the training sample dialogue set.

[0161] The model building module 740 is used to build a dialogue state recognition model based on the prior probability information and the class conditional probability information; the dialogue state recognition model is used to identify the dialogue state category corresponding to the current dialogue statement.

[0162] In one embodiment, the state category label includes a positive class label; the model building module 740 is specifically used to determine the posterior probability information of each sample dialogue statement belonging to the positive class label based on the prior probability information and the class conditional probability information corresponding to each sample dialogue statement; and to construct the dialogue state recognition model based on the posterior probability information corresponding to each sample dialogue statement.

[0163] In one embodiment, the state category label further includes a negative category label; the model building module 740 is specifically configured to determine, based on the sample dialogue vocabulary matrix, a first prior probability information corresponding to the positive category label in the training sample dialogue set; and, based on the sample dialogue vocabulary matrix, determine a second prior probability information corresponding to the negative category label in the training sample dialogue set; and, based on the first prior probability information and the second prior probability information, obtain the prior probability information corresponding to each state category label in the training sample dialogue set.

[0164] In one embodiment, the state category label further includes a negative category label; the model building module 740 is specifically configured to, for any sample dialogue statement, determine, based on the sample dialogue vocabulary matrix, a first type of conditional probability information for the positive category label of any sample dialogue statement; and, based on the sample dialogue vocabulary matrix, determine a second type of conditional probability information for the negative category label of any sample dialogue statement; and, based on the first type of conditional probability information and the second type of conditional probability information, obtain the class conditional probability information corresponding to any sample dialogue statement.

[0165] In one embodiment, the state category label further includes a negative category label; the prior probability information includes first prior probability information corresponding to the positive category label and second prior probability information corresponding to the negative category label; the class conditional probability information includes first class conditional probability information of the sample dialogue statement for the positive category label and second class conditional probability information of the sample dialogue statement for the negative category label; the model construction module 740 is specifically configured to, for any sample dialogue statement, obtain first joint probability information corresponding to the any sample dialogue statement based on the first prior probability information and the corresponding first class conditional probability information; and obtain second joint probability information corresponding to the any sample dialogue statement based on the second prior probability information and the corresponding second class conditional probability information; and obtain posterior probability information of the any sample dialogue statement belonging to the positive category label based on the first joint probability information and the corresponding second joint probability information.

[0166] In one embodiment, the model building module 740 is specifically used to obtain the total probability information corresponding to any sample dialogue statement based on the first joint probability information and the corresponding second joint probability information; and to obtain the posterior probability information of any sample dialogue statement belonging to the positive class label based on the first joint probability information and the corresponding total probability information.

[0167] In another embodiment, such as Figure 8 As shown, a dialogue state recognition device is provided, including: an input module 810 and a display module 820; wherein:

[0168] The input module 810 is used to input the current dialogue statement into the dialogue state recognition model in the current dialogue scenario to obtain the dialogue state category corresponding to the current dialogue statement; the dialogue state recognition model is obtained according to the training method of the dialogue state recognition model as described above.

[0169] Display module 820 is used to display a status category prompt window; the status category prompt window includes the dialogue status category corresponding to the current dialogue statement.

[0170] The training device and various modules of the dialogue state recognition device described above can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0171] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 9 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a training method for a dialogue state recognition model and / or a dialogue state recognition method. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0172] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0173] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0174] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0175] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0176] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0177] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0178] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0179] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A training method for a dialogue state recognition model, characterized in that, The method includes: Determine several state feature words corresponding to the training sample dialogue set; the training sample dialogue set includes several sample dialogue statements, and the state feature words are adjectives, verbs or idioms related to the dialogue state extracted from the sample dialogue statements; Construct a sample dialogue vocabulary matrix for the training sample dialogue set; the sample dialogue vocabulary matrix includes a state feature vector and a state category label corresponding to each sample dialogue statement; the state feature vector is used to characterize the occurrence of each state feature word in the sample dialogue statement, and the state category label includes positive and negative category labels used to characterize the dialogue sentiment; Based on the sample dialogue vocabulary matrix, determine the prior probability information corresponding to each state category label in the training sample dialogue set, and the class-conditional probability information corresponding to each sample dialogue statement in the training sample dialogue set. The prior probability information includes the first prior probability information corresponding to the positive class label and the second prior probability information corresponding to the negative class label. The class-conditional probability information includes the first class-conditional probability information of the sample dialogue statement for the positive class label and the second class-conditional probability information of the sample dialogue statement for the negative class label. For any of the sample dialogue statements, based on the first prior probability information and the first type of conditional probability information, a first joint probability information corresponding to the sample dialogue statement is obtained; based on the second prior probability information and the second type of conditional probability information, a second joint probability information corresponding to the sample dialogue statement is obtained; based on the first joint probability information and the second joint probability information, the total probability information corresponding to the sample dialogue statement is obtained; based on the first joint probability information and the total probability information, the posterior probability information that the sample dialogue statement belongs to the positive class label is obtained; and based on the posterior probability information corresponding to each sample dialogue statement, a dialogue state recognition model is constructed; the dialogue state recognition model is used to identify the dialogue state category corresponding to the current dialogue statement converted from the call speech in real time during the real-time dialogue between customer service and customer.

2. The method according to claim 1, characterized in that, The step of determining the prior probability information corresponding to each state category label in the training sample dialogue set based on the sample dialogue vocabulary matrix includes: Based on the sample dialogue vocabulary matrix, determine the first prior probability information corresponding to the positive class label in the training sample dialogue set; as well as, Based on the sample dialogue vocabulary matrix, determine the second prior probability information corresponding to the negative class label in the training sample dialogue set; Based on the first prior probability information and the second prior probability information, the prior probability information corresponding to each state category label in the training sample dialogue set is obtained.

3. The method according to claim 1, characterized in that, Determining the class-conditional probability information corresponding to each of the sample dialogue statements in the training sample dialogue set includes: For any of the sample dialogue statements, the first type of conditional probability information of the sample dialogue statement for the positive class label is determined based on the sample dialogue vocabulary matrix. as well as, Based on the sample dialogue vocabulary matrix, determine the second type of conditional probability information of any sample dialogue statement for the negative class label; Based on the first type of conditional probability information and the second type of conditional probability information, the class conditional probability information corresponding to any sample dialogue statement is obtained.

4. A dialogue state recognition method, characterized in that, The method includes: In the current dialogue scenario, the current dialogue statement is input into the dialogue state recognition model to obtain the dialogue state category corresponding to the current dialogue statement; the dialogue state recognition model is obtained according to the training method of the dialogue state recognition model as described in any one of claims 1 to 3. Display a status category prompt window; the status category prompt window includes the dialogue status category corresponding to the current dialogue statement.

5. A training device for a dialogue state recognition model, characterized in that, The apparatus is used to implement the training method of the dialogue state recognition model according to any one of claims 1-3, the apparatus comprising: The feature word determination module is used to determine several state feature words corresponding to the training sample dialogue set; the training sample dialogue set includes several sample dialogue statements. A matrix construction module is used to construct a sample dialogue vocabulary matrix for the training sample dialogue set; the sample dialogue vocabulary matrix includes a state feature vector and a state category label corresponding to each sample dialogue statement; the state feature vector is used to characterize the occurrence of each state feature word in the sample dialogue statement; The information determination module is used to determine, based on the sample dialogue vocabulary matrix, the prior probability information corresponding to each state category label in the training sample dialogue set, and the class conditional probability information corresponding to each sample dialogue statement in the training sample dialogue set. The model building module is used to construct a dialogue state recognition model based on the prior probability information and the class conditional probability information; the dialogue state recognition model is used to identify the dialogue state category corresponding to the current dialogue statement.

6. A dialogue state recognition device, characterized in that, The device includes: The input module is used to input the current dialogue statement into the dialogue state recognition model in the current dialogue scenario to obtain the dialogue state category corresponding to the current dialogue statement; the dialogue state recognition model is obtained according to the training method of the dialogue state recognition model as described in any one of claims 1 to 3. The display module is used to display a status category prompt window; the status category prompt window includes the dialogue status category corresponding to the current dialogue statement.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images