CN114911929A - Classification model training method, text mining method, equipment and storage medium - Google Patents

Abstract

The invention discloses a classification model training method, a text mining method, equipment and a medium, and relates to the technical field of computers. The training method uses the acquired plurality of dialogue texts as training data, performs training on the classification model according to the training data, and obtains the classification model that has been trained in stages. Among them, the training data is screened by means of cluster analysis, and the training data is marked with scene categories, so that the amount of training data marked can be greatly reduced. And, according to the difference information determined by the training data, it is judged whether the classification model that has been trained in the said stage needs to continue training. Therefore, when the difference information meets the conditions of scene mining, more scene categories can be mined through continuous training, so that in the process of text mining, more detailed scene categories and related texts with less information can be mined, which is convenient for Statistical analysis of relevant dialogue texts.

Description

Classification model training method, text mining method, equipment and storage medium

technical field

The present invention relates to the field of computer technology, in particular to a classification model training method, text mining method, equipment and storage medium.

Background technique

In some online business scenarios, when business personnel provide business consulting to customers, they will retain a large amount of consulting data. For junior business personnel, the professional words reflected in the consulting data are worth learning. In the prior art, a classification model is generally used to perform statistical analysis on consultation data, so that learning can be performed based on different scene categories. However, the classification model requires a large amount of labeled data in the training process, and lacks the ability to discover new scenes after a fixed scene is classified.

SUMMARY OF THE INVENTION

In view of the above problems, the present invention is proposed to provide a classification model training method, text mining method, device and storage medium that overcome the above problems or at least partially solve the above problems.

According to a first aspect of the present invention, a classification model training method is provided, the method comprising:

Obtain multiple dialogue texts as training data;

Perform training on the classification model according to the training data, and obtain the classification model completed by stage training;

Determine difference information based on the training data, and judge according to the difference information whether the classification model trained in the stage meets the scene mining conditions;

If the difference information meets the scene mining conditions, update the training data and continue training according to the training data and the classification model completed by the stage training;

If the difference information does not meet the scene mining conditions, the classification model trained in the stage is used as the classification model after the training;

Wherein, the classification model is trained according to the training data, and the classification model completed by stage training is obtained, including:

Perform cluster analysis on the training data to determine a plurality of training clusters;

Determine the scene category corresponding to the training cluster, and mark the text in the training cluster as the scene category;

Use the labeled text data to train the classification model, and obtain the classification model that has been trained in stages.

According to a second aspect of the present invention, there is also provided a text mining method, the method comprising:

receiving dialogue information, and obtaining the dialogue text of the first user from the dialogue information;

Inputting the dialogue text into a classification model for classification and identification, and determining the corresponding target scene category, the classification model performs training through the training data to obtain a classification model completed by stage training, and determines the difference information based on the training data, Judging whether the classification model trained in the stage meets the scene mining conditions, and determining whether to update the training data according to the judgment result and continuing the classification model trained in the training stage to obtain;

query the target reply text corresponding to the target scene category;

The target reply text is used as the dialogue text of the second user, and the dialogue text of the first user is fed back.

According to a third aspect of the present invention, an electronic device is also provided, comprising:

one or more processors;

memory;

one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs are configured to perform any of the above the method described.

According to a fourth aspect of the present invention, a computer-readable storage medium is also provided, which stores a computer program used in conjunction with an electronic device, and the computer program can be executed by a processor to perform any of the above-mentioned methods.

In the solution of the present invention, a plurality of acquired dialogue texts are used as training data, and the classification model is trained according to the training data to obtain a classification model that has been trained in stages. Among them, the training data is screened by means of cluster analysis, and the training data is marked with scene categories, so that the amount of marking of the training data can be greatly reduced. And, according to the difference information determined by the training data, it is judged whether the classification model after the training in the said stage needs to continue training. When the difference information meets the conditions for scene mining, more scene categories can be mined by continuing training. In the process of text mining, more detailed scene categories and related texts with less information can be mined, so as to facilitate statistical analysis of related dialogue texts.

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

Description of drawings

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

In the attached image:

Fig. 1 is a flow chart of steps of a classification model training method provided by an embodiment of the present invention;

2 is a flowchart of steps of another classification model training method provided by an embodiment of the present invention;

3 is a flowchart of steps of a text mining method provided by an embodiment of the present invention;

4 is a schematic diagram of display content of a display page of a first client provided by an embodiment of the present invention;

5 is a schematic diagram of a display content of a display page of a first client provided by an embodiment of the present invention;

6 is a schematic diagram of another display content of a display page of a first client provided by an embodiment of the present invention;

7 is a schematic diagram of a display content of a display page of a client according to an embodiment of the present invention;

8 is a schematic diagram of another display content of a display page of a client according to an embodiment of the present invention;

9 is a block diagram of a classification model training apparatus provided by an embodiment of the present invention;

FIG. 10 is a block diagram of a text data mining apparatus provided by an embodiment of the present invention.

Detailed ways

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

Referring to FIG. 1, a classification model training method provided by an embodiment of the present invention is shown, and the method may include:

S101. Acquire multiple dialogue texts as training data.

S102. Perform training on the classification model according to the training data, to obtain a classification model that has been trained in stages.

In the embodiment of the present invention, in some speech dialogue scenarios or text dialogue scenarios, the dialogue records in the dialogue process are stored under the authorization of the customer. Thus, a plurality of dialogue texts can be determined based on the collected dialogue records. Among them, the customer can be regarded as the first user, and the business personnel can be regarded as the second user. If the dialogue record is stored in the form of voice data, the dialogue record can be voice-converted to obtain the corresponding dialogue text. After obtaining several dialogue texts, use them as training data, and perform training on the preset classification model until the classification model meets the first training condition, stop the training of the classification model, and stop the trained classification model Identify the classification model trained for the stage. In one example, the first training condition may be that the loss function of the classification model no longer decreases or decreases by a small amount.

Perform training on the classification model according to the training data, and obtain the classification model completed by stage training, including the following steps:

Cluster analysis is performed on the training data to determine a plurality of training clusters.

The scene category corresponding to the training cluster is determined, and the text in the training cluster is marked as the scene category.

Use the labeled text data to train the classification model, and obtain the classification model that has been trained in stages.

In this embodiment of the present invention, the training data can be clustered and analyzed to determine a plurality of training clusters, and a corresponding scene category can be determined from the plurality of training clusters. Wherein, each scene category may be determined according to the textual semantic information in each training cluster. After the scene category corresponding to the training cluster is determined, the text in each training cluster may be marked as the corresponding scene category. Annotated text data is formed by annotating scene categories for each training cluster. As a result, labeling each dialogue text is avoided, which greatly reduces the amount of labeling of training data. Then, the labeled text data is trained on the preset classification model, and the classification model completed in the stage training is obtained.

S103. Determine difference information based on the training data, and judge whether the classification model completed in the stage of training meets the scene mining conditions according to the difference information.

S104. If the difference information meets the scene mining conditions, update the training data according to the training data and the classification model completed by the stage training, and continue the training.

S105. If the difference information does not meet the scene mining conditions, use the classification model trained in the stage as the classification model after training.

In the embodiment of the present invention, after the scene category corresponding to the training cluster is determined, the difference information of the text data in the same training cluster can be determined. Among them, the difference information can be understood as the semantic similarity of the semantic information of different texts. In this way, it can be determined whether the classification model trained in the stage meets the scene mining conditions according to the difference information.

In one example, the scene mining condition is used to evaluate whether the scene category of the classification model trained in the stage needs to be refined. For example, when the semantic similarity of the semantic information of different texts in the same training cluster is high, it means that the boundary between the corresponding two texts is not clear, and the classification effect is not obvious. Therefore, a semantic threshold can be preset, and when the semantic similarity is smaller than the semantic threshold, it is determined that the difference information does not meet the scene mining conditions, that is, the classification effect is obvious, so the classification model trained in the above stage is used. directly as a trained classification model. When the semantic similarity is equal to or greater than the semantic threshold, it is determined that the difference information complies with the scene mining conditions, and the training data needs to be updated at this time, and the updated training data is continued to train the classification model that has been trained at the stage, Until the classification model trained in the stage of continuing training meets the first training condition, the training of the classification model is stopped, and the classification model trained in the stage after the continuous training is stopped is determined as the classification model that has been trained.

In summary, the embodiment of the present invention provides a classification model training method. The method uses a plurality of acquired dialogue texts as training data, performs training on a classification model according to the training data, and obtains a classification model that has been trained in stages. Among them, the training data is screened by means of cluster analysis, and the training data is marked with scene categories, so that the amount of marking of the training data can be greatly reduced. And, according to the difference information determined by the training data, it is judged whether the classification model after the training in the said stage needs to continue training. Therefore, when the difference information meets the conditions for scene mining, more scene categories can be mined by continuing training, so that in the process of text mining, more detailed scene categories and related texts with less information can be mined, which is convenient for Statistical analysis of relevant dialogue texts.

Referring to FIG. 2, another classification model training method provided by an embodiment of the present invention is shown, and the method may include:

S201. Acquire multiple dialogue texts as training data.

S202. Perform training on the classification model according to the training data, and obtain a classification model that has been trained in stages.

In the embodiment of the present invention, after several dialogue texts are obtained, they are used as training data to perform training on a preset classification model until the classification model meets the first training condition, the training of the classification model is stopped, and the The classification model after the training is stopped is determined as the classification model after the stage training is completed.

Cluster analysis is performed on the training data, multiple training clusters are determined, and corresponding scene categories are determined from the multiple training clusters. Wherein, a first clustering model may be preset to perform a first cluster analysis on the training data. For example, the first clustering model may use a K-means (k-means clustering algorithm, K-means clustering) algorithm. According to the number of clusters, two dialogue texts with close distances can be determined as a cluster. Therefore, when performing statistical analysis of scene categories on multiple dialogue texts, the number m1 of clusters output by the first clustering model for the first time can be predetermined, wherein the number m1 of clusters can be based on the actual number of scene categories n1 is determined, which is not limited here. However, the number m1 of clusters needs to be greater than the number n1 of scene categories. The number of clusters m may be the number of scene categories n plus a preset threshold. The range of the preset threshold may be set to be between 3 and 10. For example, when it is specified that the initial business scenario includes a sales scenario and an after-sales scenario, the number n1 of the corresponding scenario categories is determined to be 2. Then the number of clusters m1 may be 3, 4, 5, or 12, etc., so that the clustering results can be prevented from being too divergent.

After the first cluster analysis of the first cluster model, m1 training clusters are determined. In one example, after multiple training clusters are determined, a display page can be provided to display the multiple training clusters, so that based on the triggering of the screening control, the scene category can be screened out from the multiple training clusters. class cluster, so that the training class cluster can be classified according to the preset business scenario. For example, 2 training clusters with clear business scenarios are selected from 3 training clusters for category labeling. When performing category marking, an editing control may be provided on the display page for the user to customize the corresponding scene category. Based on the triggering of the edit control, the class labeling of the training cluster is completed. For example, after the category labeling is completed, text data corresponding to the sales scene and text data corresponding to the after-sales scene are obtained.

By labeling the scene category for each training cluster, the texts in the training clusters that are labelled respectively form the labeled text data. As a result, the step of labeling each dialogue text during model training is avoided. Greatly reduces the amount of annotations for training data. Then, the labeled text data is trained on the preset classification model, and the classification model M1 completed by the stage training is obtained. The classification model may be a BERT-base model including N hidden layers. When training the classification model for the first time, the text features output by the Nth hidden layer can be used for classification, and the number of scene categories corresponding to the classification is n1. M1.

In an optional embodiment of the invention, in the process of inputting the training data into the first clustering model to perform the first cluster analysis and outputting multiple training clusters, first of all, each dialogue in the training data can be analyzed. The text is divided into words, and several corresponding text segmentation words are determined. Cluster analysis is performed according to the semantic information of text segmentation of each dialogue text. Among them, in some dialogue texts, classifying the semantic information corresponding to some text segments can improve the effect of cluster analysis. For example, the user's intention to ask "how long does it take to express to Beijing" and "how long does it take to express to Shanghai" are the same. Both Shanghai and Beijing can be classified as addresses. Therefore, before performing semantic analysis, named entity recognition is performed on several text segments, so as to determine the target named entity that needs to be classified. Wherein, the target named entity includes at least one of the following: a person's name, an organization name, and a place name. After the target named entity is determined, the target named entity may be replaced with the corresponding target keyword. Among them, the target keywords can be words related to the target named entity.

After all the replacement of the target named entity is completed, the replaced text segments are converted into text word vectors or text TFIDF (term frequency-inverse document frequency, term frequency-inverse text frequency index) values respectively. Then several text word vectors or text TFIDF values are input into the first clustering model, and the above clustering analysis operation is performed.

S203. Determine difference information based on the training data.

S204. Provide a display page to display the difference information, and acquire mining operation information based on the display page.

S205. According to the mining operation information, determine whether the classification model trained in the stage meets the scene mining conditions.

In the embodiment of the present invention, after the scene category corresponding to the training cluster is determined, the difference information of the text data in the same training cluster can be determined. Among them, the difference information can be understood as the semantic similarity of the semantic information of different texts. In this way, it can be determined whether the classification model trained in the stage meets the scene mining conditions according to the difference information.

In an example, a display page may be provided, the difference information is displayed on the display page, and a first selection control is set on the display page, and based on the triggering of the first selection control, the acquisition is obtained. Corresponding mining operation information in the display page. For example, the mining operation information may include Yes or No. When the mining operation information is Yes, it is determined that the classification model trained in the stage meets the scene mining conditions, and step S206 is executed; when the mining operation information is No , it is determined that the classification model trained in the said stage does not meet the scene mining conditions, and step S208 is executed.

S206: Input the marked text data into the classification model completed in the stage training, and perform feature extraction to obtain corresponding feature text.

S207, using the feature text as training data.

In the embodiment of the present invention, the marked text data is input into the classification model after the stage training is completed, and feature extraction is performed. The feature extraction can be understood as the hidden layer in the classification model completed by stage training, and the corresponding feature text is directly output. In an example, when the feature extraction is performed for the first time, the output feature of the first hidden layer is determined as the corresponding feature text, and the feature text is used as new training data. Considering that the text features output by the hidden layer closer to the output layer are more accurate. When performing the first feature extraction, the text feature output by the N-1th hidden layer can be determined as feature text, and the training data can be updated according to the feature text. Then, step S202 is performed to continue training the classification model that has been trained in the stage through the updated training data.

After the feature text is used as new training data, the classification model is trained according to the training data. Since the scene category needs to be further mined, when the first clustering analysis is performed on the training data according to the first clustering model, the number of clusters output by the first clustering model is adjusted to m2, wherein the number of clusters m2 greater than the number of scene categories n1. After the first cluster analysis of the first cluster model, m2 training clusters are determined. The m2 training clusters are displayed on the display page, so that based on the triggering of the screening control, clusters with clear scene categories can be selected from the m2 training clusters, so that the scene categories can be marked according to the training clusters. , the number n2 of scene categories is obtained at this time, and the number n2 of scene categories is greater than the number n1 of scene categories. For example, on the basis of sales scenarios and after-sales scenarios, the scenario categories are further divided into: product consultation scenarios, purchase consent scenarios, after-sales scenarios consent, and refusal to exchange items, etc.

By labeling the scene category for each training cluster, the texts in the training clusters that are labelled respectively form the labeled text data. Then, the labeled text data is used to continue training the classification model M1 completed in the stage training, and the classification model M2 completed in the stage training is obtained, which can classify n2 scene categories, so that the classification model M2 completed in the stage training is compared with the stage training. The completed classification model M1 increases the number of scene categories.

Then, the difference information is determined based on the training data of the classification model M2 that is trained in the training stage, and then it is determined according to the difference information whether the classification model M2 completed in the stage training meets the scene mining conditions, and it is determined that the classification model M2 completed in the stage training meets the When the scene mining conditions are met, step S206 is performed; when it is determined that the classification model M2 completed in the training phase does not meet the scene mining conditions, step S208 is performed. For example, when it is determined that the classification model M2 trained in the said stage meets the conditions for scene mining, the scene category needs to be further refined. For example, the product consultation scene is further divided into: price inquiry scene, size inquiry scene, style inquiry scene, etc. And so on, and finally get the trained classification model. Applying the trained classification model to text mining can accurately classify the customer's dialogue text, and determine the target reply text with high matching degree based on the scene category obtained by the accurate classification, so as to improve the service level of business personnel and improve user experience.

S208. Use the classification model that has been trained at the stage as the classification model that has been trained.

In the embodiment of the present invention, it is determined that the difference information does not meet the scene mining conditions, that is, the classification effect is obvious, so the classification model completed in the training phase is directly used as the classification model after the training.

Referring to FIG. 3, a flowchart of steps of a text mining method provided by an embodiment of the present invention is shown, and the method may include:

S301. Receive dialog information, and obtain a dialog text of a first user from the dialog information.

S302. Input the dialogue text into a classification model for classification and recognition, and determine the corresponding target scene category, and the classification model is trained by training data to obtain a classification model completed by stage training, and determine the difference based on the training data. information, judging whether the classification model completed in the training stage meets the scene mining conditions, and determining whether to update the training data according to the judgment result to continue to obtain the classification model trained in the training stage.

S303 , query the target reply text corresponding to the target scene category.

S304 , using the target reply text as the dialogue text of the second user, and feeding back the dialogue text of the first user.

In the embodiment of the present invention, the classification model is obtained after being trained by the above-mentioned classification model training method. After the dialogue information is acquired, the dialogue text of the first user is acquired from the dialogue information. The dialogue text of the first user is input into the classification model for classification and recognition, and a target scene category corresponding to the dialogue text of the first user is obtained. Then, the historical dialogue text of the second user is obtained in the dialogue record, wherein the scene category corresponding to the historical dialogue text of the second user has been marked, wherein the number and type of scene categories corresponding to the historical dialogue text of the second user, It is consistent with the number and type of scene categories classified and identified by the classification model.

Therefore, after determining the target scene category corresponding to the dialogue text of the first user, the historical dialogue text of the second user under the target scene category is queried in the database, and determined as the target reply text, and the target The reply text is used as the dialogue text of the second user for feedback of the dialogue text of the first user. There may be multiple pieces of the target reply text, which may be displayed in the form of at least one cluster. In an example, when there are multiple pieces of the target reply text, one of the target reply texts may be determined as the dialogue text of the second user by clicking on the selection control on the display page, and fed back to the first user. Therefore, even an inexperienced business person can dig out the most suitable words for responding to the dialogue text of the first user from the historical dialogue text of the second user. Therefore, the learning of the business personnel is facilitated, the service level is improved, and the user experience of the first user is also improved.

In another example, if there are a large number of target reply texts under the same target scene category, in order to excavate different reply methods, a preset second clustering model can be used to analyze several target reply texts. The second cluster analysis. Among them, the number of clusters output by the second clustering model changes dynamically with the semantic information of the target reply text. It can use the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm. Therefore, in the second cluster analysis process, it is not necessary to set a specific number of clusters, and several target reply texts with different reply modes are obtained. Therefore, one of the target reply texts can be determined as the dialogue text of the second user by clicking on the selection control on the display page, and fed back to the first user.

In an example, referring to FIG. 4 and FIG. 5 , a display page is respectively provided on the first client of the first user and the second client of the second user to display the dialog message. Wherein, in the display page corresponding to the second user, a classification control (such as a circle in the middle of the page) corresponding to each dialogue text of the first user is set, and according to the triggering of the classification control, a classification instruction is generated to send to the server. The dialog text corresponding to the classification control is input into the classification model for classification and identification, and after the target scene category is determined and the target reply text is determined, the target reply text is sent to the second client. The corresponding target reply text is displayed on the display page corresponding to the second user. 6, each target reply text is matched with a reply selection control (such as the arrow in the middle part of the page), and based on the triggering of the reply selection control, the target reply text corresponding to the reply selection control is used as the second user's Conversation text, sent to the first client.

In another example, the first client and the second client may be the same client. Referring to FIG. 7 , the user may enter a dialog text on the display page of the client, and the dialog text is the dialog text of the first user by default. Based on the triggering of the classification control on the display page, the dialogue text of the first user is input into the classification model for classification and recognition, and the target scene category corresponding to the dialogue text of the first user is obtained. Then, the historical dialogue text of the second user is obtained in the dialogue record, wherein the scene category corresponding to the historical dialogue text of the second user has been marked, wherein the number and type of scene categories corresponding to the historical dialogue text of the second user, It is consistent with the number and type of scene categories classified and identified by the classification model.

Therefore, after determining the target scene category corresponding to the dialogue text of the first user, the historical dialogue text of the second user under the target scene category is queried in the database, and determined as the target reply text, and the target The reply text is used as the dialogue text of the second user, and is used for feedback of the dialogue text of the first user, and the dialogue text of the second user is displayed in the user's client terminal, referring to FIG. 8 . Therefore, in the business training and learning scenario of the new employee, the new employee can customize the dialogue text of the first user, so as to learn different vocabulary based on the target reply text obtained from the dialogue text of different first users. These target response texts can provide material for new employee training or for old employees to learn from each other.

It should be noted that, for the sake of simple description, the method embodiments are expressed as a series of action combinations, but those skilled in the art should know that the embodiments of the present application are not limited by the described action sequence, because According to the embodiments of the present application, certain steps may be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present application.

Referring to FIG. 9, an apparatus for training a classification model provided by an embodiment of the present invention is shown, and the apparatus may include:

The data acquisition module 901 is used for acquiring multiple dialogue texts as training data.

The model stage training module 902 is configured to perform training on the classification model according to the training data to obtain the classification model after stage training.

The condition judgment module 903 is configured to determine difference information based on the training data, and judge whether the classification model trained in the stage meets the scene mining conditions according to the difference information.

The first training module 904 is configured to update the training data and continue the training according to the training data and the classification model completed by the stage training if the difference information meets the scene mining conditions.

The second training module 905 is configured to, if the difference information does not meet the scene mining conditions, use the classification model trained in the stage as the trained classification model.

Wherein, the model stage training module is also used for:

Cluster analysis is performed on the training data to determine a plurality of training clusters.

The scene category corresponding to the training cluster is determined, and the text in the training cluster is marked as the scene category.

Use the labeled text data to train the classification model, and obtain the classification model that has been trained in stages.

An optional inventive embodiment, determining the difference information based on the training data, includes:

Analyze the text of the training clusters corresponding to different scene categories to determine the difference information.

In an optional inventive embodiment, the first training module may include:

The feature extraction sub-module is used to input the marked text data into the classification model completed in the stage training, and perform feature extraction to obtain the corresponding feature text.

The data update submodule is used for using the feature text as training data.

In an optional embodiment of the invention, the model stage training module is further used for:

The training data is input into a first clustering model to perform a first clustering analysis, and a plurality of training clusters is output, wherein the number of the training clusters is preset by the first clustering model.

In an optional inventive embodiment, the model stage training module may include:

The word segmentation sub-module is used to divide the training data into words, and determine several corresponding text segmentations.

The word segmentation conversion sub-module is used to convert several text segmentations into text word vectors or text TFIDF values respectively.

The cluster output sub-module is used to input several text word vectors or text TFIDF values into the first clustering model to perform the first cluster analysis, and output several training clusters.

In an optional inventive embodiment, the cluster-like output submodule may further include:

The recognition unit is used for performing named entity recognition on several text segments, and determining target named entities in several text segments, and the target named entities include at least one of the following: a person's name, an institution name and a place name.

The replacement unit is used to replace the target named entity with the target keyword, and convert several text word segmentations after the replacement into text TFIDF values respectively.

In an optional inventive embodiment, the method may further include:

The information display module is configured to provide a display page to display the difference information, and obtain mining operation information based on the display page.

A condition determination module is configured to determine, according to the mining operation information, whether the classification model trained in the stage meets the scene mining conditions.

Referring to FIG. 10, a text mining apparatus provided by an embodiment of the present invention is shown, and the apparatus may include:

The dialogue receiving module 1001 is configured to receive dialogue information, and obtain the dialogue text of the first user from the dialogue information.

The scene recognition module 1002 is used to input the dialogue text into the classification model for classification and recognition, and determine the corresponding target scene category. The training data is used to determine the difference information, and it is judged whether the classification model completed in the training stage meets the scene mining conditions, and whether to update the training data according to the judgment result is obtained by continuing the classification model trained in the training stage.

The text query module 1003 is configured to query the target reply text corresponding to the target scene category.

The text feedback module 1004 is configured to use the target reply text as the dialogue text of the second user, and feed back the dialogue text of the first user.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments may be referred to each other.

It is easy for those skilled in the art to think that any combination of the above-mentioned embodiments is feasible, so any combination of the above-mentioned embodiments is an embodiment of the present invention, but due to space limitations, this description is hereby limited to Not detailed.

In the description provided herein, numerous specific details are set forth. It will be understood, however, that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.

Similarly, it is to be understood that in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together into a single embodiment, figure, or its description. This disclosure, however, should not be construed as reflecting an intention that the invention as claimed requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the Detailed Description are hereby expressly incorporated into this Detailed Description, with each claim standing on its own as a separate embodiment of this invention.

Those skilled in the art will understand that the modules in the device in the embodiment can be adaptively changed and arranged in one or more devices different from the embodiment. The modules or units or components in the embodiments may be combined into one module or unit or component, and further they may be divided into multiple sub-modules or sub-units or sub-assemblies. All features disclosed in this specification (including accompanying claims, abstract and drawings) and any method so disclosed may be employed in any combination, unless at least some of such features and/or procedures or elements are mutually exclusive. All processes or units of equipment are combined. Each feature disclosed in this specification (including accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.

An electronic device comprising:

one or more processors;

memory;

one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs are configured to perform the above-described embodiments the method described.

A computer-readable storage medium storing a computer program used in conjunction with an electronic device, where the computer program can be executed by a processor to complete the methods described in the above embodiments.

It should be understood by those skilled in the art that the embodiments of the embodiments of the present invention may be provided as a method, an apparatus, or a computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media having computer-usable program code embodied therein, including but not limited to disk storage, CD-ROM, optical storage, and the like.

Embodiments of the present invention are described with reference to flowcharts and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present invention. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing terminal equipment to produce a machine that causes the instructions to be executed by the processor of the computer or other programmable data processing terminal equipment Means are created for implementing the functions specified in the flow or flows of the flowcharts and/or the blocks or blocks of the block diagrams.

These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing terminal equipment to operate in a particular manner, such that the instructions stored in the computer readable memory result in an article of manufacture comprising instruction means, the The instruction means implement the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing terminal equipment, so that a series of operational steps are performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby executing on the computer or other programmable terminal equipment The instructions executed on the above provide steps for implementing the functions specified in the flowchart or blocks and/or the block or blocks of the block diagrams.

Although preferred embodiments of the embodiments of the present invention have been described, additional changes and modifications to these embodiments may be made by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the embodiments of the present invention.

Finally, it should also be noted that in this document, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply these entities or that there is any such actual relationship or sequence between operations. Moreover, the terms "comprising", "comprising" or any other variation thereof are intended to encompass non-exclusive inclusion such that a process, method, article or terminal device that includes a list of elements includes not only those elements, but also a non-exclusive list of elements. other elements, or also include elements inherent to such a process, method, article or terminal equipment. Without further limitation, an element defined by the phrase "comprises a..." does not preclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

A classification model training method, a text mining method, a classification model training device and a text mining device provided by the present invention have been described in detail above. Specific examples are used in this paper to explain the principles and implementation of the present invention. The description of the above embodiment is only used to help understand the method of the present invention and its core idea; meanwhile, for those of ordinary skill in the art, according to the idea of the present invention, both in the specific embodiment and application scope will be There are changes. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims (10)

1. a classification model training method, is characterized in that, described method comprises: Obtain multiple dialogue texts as training data; Perform training on the classification model according to the training data, and obtain the classification model completed by stage training; Determine difference information based on the training data, and judge whether the classification model trained in the stage meets the scene mining conditions according to the difference information; If the difference information meets the scene mining conditions, update the training data and continue training according to the training data and the classification model completed by the stage training; If the difference information does not meet the scene mining conditions, the classification model trained in the stage is used as the classification model after the training; Wherein, the classification model is trained according to the training data, and the classification model completed by stage training is obtained, including: Perform cluster analysis on the training data to determine a plurality of training clusters; Determine the scene category corresponding to the training cluster, and mark the text in the training cluster as the scene category; Use the labeled text data to train the classification model, and obtain the classification model that has been trained in stages. 2. The classification model training method according to claim 1, wherein determining difference information based on the training data, comprising: Analyze the text of the training clusters corresponding to different scene categories to determine the difference information. 3. The classification model training method according to claim 1, wherein the updating training data according to the classification model completed by the training data and stage training, comprises: Input the marked text data into the classification model completed in the stage training, and perform feature extraction to obtain the corresponding feature text; The feature text is used as training data. 4. The classification model training method according to claim 1, wherein the said training data is subjected to cluster analysis to determine a plurality of training clusters, comprising: The training data is input into a first clustering model to perform a first clustering analysis, and a plurality of training clusters is output, wherein the number of the training clusters is preset by the first clustering model. 5. The classification model training method according to claim 4, wherein the training data is input into the first clustering model to carry out the first cluster analysis, and a plurality of training clusters are output, comprising: The training data is divided into words, and several corresponding text segmentations are determined; Convert several text segments into text word vectors or text TFIDF values respectively; Several text word vectors or text TFIDF values are input into the first clustering model to perform the first cluster analysis, and several training clusters are output. 6. classification model training method according to claim 5, is characterized in that, described converting several text segmentations into text TFIDF values respectively, comprises: Perform named entity recognition on several text segments, and determine target named entities in several text segments, and the target named entities include at least one of the following: a person's name, an organization name, and a place name; The target named entity is replaced by the target keyword, and the replaced text segments are converted into text TFIDF values respectively. 7. The classification model training method according to claim 2, wherein the method further comprises: providing a display page to display the difference information, and acquiring mining operation information based on the display page; According to the mining operation information, it is determined whether the classification model trained in the stage meets the scene mining conditions. 8. A text mining method, wherein the method comprises: receiving dialogue information, and obtaining the dialogue text of the first user from the dialogue information; Inputting the dialogue text into a classification model for classification and identification, and determining the corresponding target scene category, the classification model performs training through the training data to obtain a classification model completed by stage training, and determines the difference information based on the training data, Judging whether the classification model trained in the stage meets the scene mining conditions, and determining whether to update the training data according to the judgment result and continuing the classification model trained in the training stage to obtain; query the target reply text corresponding to the target scene category; The target reply text is used as the dialogue text of the second user, and the dialogue text of the first user is fed back. 9. An electronic device comprising: one or more processors; memory; one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs are configured to perform claim 1 The method of any one of -8. 10. A computer-readable storage medium storing a computer program for use in conjunction with an electronic device, the computer program being executable by a processor to perform the method of any one of claims 1-8.

Images