CN113792819B - Multi-task model-based node reusable intelligent outbound method and system - Google Patents
Multi-task model-based node reusable intelligent outbound method and system Download PDFInfo
- Publication number
- CN113792819B CN113792819B CN202111276483.4A CN202111276483A CN113792819B CN 113792819 B CN113792819 B CN 113792819B CN 202111276483 A CN202111276483 A CN 202111276483A CN 113792819 B CN113792819 B CN 113792819B
- Authority
- CN
- China
- Prior art keywords
- task
- matching model
- training
- intention
- node
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000012549 training Methods 0.000 claims abstract description 143
- 239000013598 vector Substances 0.000 claims description 68
- 230000006870 function Effects 0.000 claims description 16
- 230000015654 memory Effects 0.000 claims description 15
- 238000007621 cluster analysis Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 description 10
- 238000012545 processing Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 238000013473 artificial intelligence Methods 0.000 description 4
- 238000013145 classification model Methods 0.000 description 4
- 238000003860 storage Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 2
- 238000012790 confirmation Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Machine Translation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a node reusable intelligent outbound method and a system based on a multi-task model, wherein the method comprises the following steps: obtaining the training corpus and taking the training corpus as a training set; obtaining a first flow Cheng Shu from a first outbound scene; obtaining a second flow tree according to a second outbound scenario, and taking each non-leaf node of the first flow Cheng Shu and the second flow tree as a subtask of multitasking; initializing a multi-task intent matching model, the multi-task intent matching model comprising an encoder and a plurality of task layers; obtaining a first multitasking intent matching model; training and updating task layers different from the subtasks of the first stream Cheng Shu in the first multitasking intention matching model through the training set to obtain a second multitasking intention matching model; and carrying out intelligent outbound according to the second multitasking intention matching model. The technical problems that the regular dead plate accuracy is difficult to reuse and cannot be updated in real time in the prior art are solved.
Description
Technical Field
The invention relates to the field of intelligent outbound, in particular to a node reusable intelligent outbound method and system based on a multi-task model.
Background
In recent years, artificial intelligence has evolved dramatically, and in most industries, artificial intelligence has been able to work in part. The telephone calling system can make automatic call and issue notice, and is a perfect combination of computer technology and speech technology. With the increasing cost of operation, telephone-side customer service centers are also following the trend of shifting from "cost center to profit center". When the enterprise has a plurality of service objects to communicate or inform, if the service objects are manually dialed one by one, the method is time-consuming and labor-consuming, and the efficiency is low. The telephone outbound system can be used for automatically notifying, and after the other party is automatically judged to be off-hook, a pre-designed voice or text notification is sent to the other party. After judging that the other party hangs up, the next object can be notified automatically, and the method is very convenient and quick. The intelligent outbound system well solves the problems of high investment cost and low working efficiency at the server, enterprises can preset the required voice outbound flow according to business requirements, and the system can realize the flows of identity confirmation, business consultation, business processing and the like of users.
However, in the process of implementing the technical scheme of the embodiment of the application, the inventor discovers that the above technology has at least the following technical problems:
rule-based intent recognition requires manual configuration of rules for each phone node, but this approach is too dead and has very low recognition accuracy, and if the phone of the customer is slightly changed, the recognition is wrong, so that the problem that the same text is under different nodes of the same phone flow but has different meanings cannot be solved, and the text is difficult to multiplex and cannot be updated in real time.
Disclosure of Invention
Aiming at the defects in the prior art, the method and the system for node multiplexing intelligent outbound based on the multi-task model solve the technical problems that rule-based intent recognition in the prior art needs to manually configure rules for each voice node, but the mode is excessively dead and the recognition accuracy is very low, if a user has slight variation, the recognition is wrong, the problem that the same text is under different nodes of the same voice flow but has different meanings, the text is difficult to multiplex and cannot be updated in real time can not be solved, and the technical effects that corpus and labels required for realizing the free configuration of different nodes of a flow tree through the multi-task intent classification model are achieved, intent labels under different nodes are not mutually influenced, prototype corpus contained under each node can be directly updated to finish the update and configuration of the model, and the reusability of the model is improved are achieved.
In view of the above problems, the embodiments of the present application provide a method and a system for node reusable intelligent outbound based on a multitasking model.
In one aspect, an embodiment of the present application provides a method for node-reusable intelligent outbound based on a multitasking model, where the method includes: performing cluster analysis on all the training corpuses by using a cluster algorithm to obtain the training corpuses conforming to a preset rule in each category, and taking the training corpuses as a training set; obtaining a first flow Cheng Shu from a first outbound scene; obtaining a second flow Cheng Shu according to a second outbound scenario, wherein each non-leaf node of the first flow tree and the second flow tree is used as a subtask for multitasking training; initializing a multi-task intention matching model, wherein the multi-task intention matching model comprises a task layer which corresponds to each subtask in the flow tree respectively; training the initialized multi-task intention matching model by using the corpus of the first flow tree through the training set to obtain a first multi-task intention matching model; training and updating task layers different from the subtasks of the first stream Cheng Shu in the initialized first multi-task intention matching model by using the corpus of the second flow tree in the training set to obtain a second multi-task intention matching model; and carrying out intelligent outbound according to the second multitasking intention matching model.
On the other hand, the embodiment of the application also provides a node reusable intelligent outbound system based on a multi-task model, wherein the system comprises: the first obtaining unit is used for obtaining the training corpus which accords with a preset rule in each category and taking the training corpus as a training set; a second obtaining unit for obtaining a first flow Cheng Shu according to a first outbound scene; a third obtaining unit for obtaining a second stream Cheng Shu from a second outbound scene; the first initializing unit is used for initializing a multi-task intention matching model, and the multi-task intention matching model comprises a task layer which corresponds to each subtask in the flow tree respectively; the fourth obtaining unit is used for training the initialized multi-task intention matching model by using the corpus of the first flow tree in the training set to obtain a first multi-task intention matching model; a fifth obtaining unit, configured to perform training update on a task layer different from the subtasks of the first flow Cheng Shu in the initialized first multitasking intention matching model by using the corpus of the second flow tree in the training set, to obtain a second multitasking intention matching model; and the first outbound unit is used for performing intelligent outbound according to the second multitasking intention matching model.
In another aspect, an embodiment of the present application further provides a node-reusable intelligent outbound system based on a multitasking model, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the steps of the method according to any one of the first aspects when the processor executes the program.
One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:
performing cluster analysis on all the training corpuses by using a cluster algorithm to obtain the training corpuses conforming to a preset rule in each category, and taking the training corpuses as a training set; obtaining a first flow Cheng Shu from a first outbound scene; obtaining a second flow Cheng Shu according to a second outbound scenario, wherein each non-leaf node of the first flow tree and the second flow tree is used as a subtask for multitasking training; initializing a multi-task intention matching model, wherein the multi-task intention matching model comprises a task layer which corresponds to each subtask in the flow tree respectively; training the initialized multi-task intention matching model by using the corpus of the first flow tree through the training set to obtain a first multi-task intention matching model; training and updating task layers different from the subtasks of the first stream Cheng Shu in the initialized first multi-task intention matching model by using the corpus of the second flow tree in the training set to obtain a second multi-task intention matching model; and carrying out intelligent outbound technical effects according to the second multitasking intention matching model.
The foregoing description is only an overview of the present application, and is intended to be implemented in accordance with the teachings of the present application in order that the same may be more clearly understood and to make the same and other objects, features and advantages of the present application more readily apparent.
Drawings
Other features, objects and advantages of the present invention will become more apparent upon reading of the detailed description of non-limiting embodiments, made with reference to the accompanying drawings in which:
fig. 1 is a schematic flow chart of a node reusable intelligent outbound method based on a multitasking model according to an embodiment of the present application;
FIG. 2 is a diagram showing an embodiment of the present application for optimizing the multi-task intent-matching model by minimizing the difference between the maximum intra-class distance and the minimum inter-class distance to obtain a first multi-task intent-matching model;
FIG. 3 is a diagram illustrating an embodiment of the present application for training and updating the task layer T 5,T6 corresponding to the node 5,node6 to obtain a second multi-task intention matching model;
FIG. 4 is a diagram illustrating an intelligent outbound call according to the first prediction result according to an embodiment of the present application;
FIG. 5 is a diagram illustrating the assignment of training data and corresponding intent labels for each sub-task according to an embodiment of the present application;
fig. 6 is a schematic diagram of a data modeling system for improving intelligent marketing efficiency according to an embodiment of the present application.
Fig. 7 is a schematic structural diagram of an exemplary electronic device according to an embodiment of the present application.
Reference numerals illustrate: the device comprises a first obtaining unit 11, a second extracting unit 12, a third obtaining unit 13, a first initializing unit 14, a fourth obtaining unit 15, a fifth obtaining unit 16, a first outbound unit 17, an electronic device 50, a processor 51, a memory 52, an input device 53, and an output device 54.
Detailed Description
The embodiment of the application solves the problems that rule-based intent recognition needs to be manually configured for each telephone node, but the mode is excessively dead and the recognition accuracy is very low, and recognition errors are caused if the telephone of a client is slightly changed, the problem that the same text is under different nodes of the same telephone process but has different meanings and is difficult to multiplex and cannot be updated in real time in the prior art by providing the data modeling method and the system for improving the intelligent marketing efficiency, achieves the technical effects that corpus and labels required by realizing the free configuration of different nodes of a process tree through a multi-task intent classification model, causes intent labels under different nodes not to influence each other, can directly update prototype corpus contained under each node to finish the update and configuration of the model, and improves the reusability of the model.
Hereinafter, technical solutions in exemplary embodiments of the present application will be clearly and in detail described with reference to the accompanying drawings. It should be apparent that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein. All other embodiments, which can be made by a person of ordinary skill in the art based on the embodiments of the application without making any inventive effort, are intended to fall within the scope of the application.
Summary of the application
In recent years, artificial intelligence has evolved dramatically, and in most industries, artificial intelligence has been able to work in part. The telephone calling system can make automatic call and issue notice, and is a perfect combination of computer technology and speech technology. With the increasing cost of operation, telephone-side customer service centers are also following the trend of shifting from "cost center to profit center". When the enterprise has a plurality of service objects to communicate or inform, if the service objects are manually dialed one by one, the method is time-consuming and labor-consuming, and the efficiency is low. The telephone outbound system can be used for automatically notifying, and after the other party is automatically judged to be off-hook, a pre-designed voice or text notification is sent to the other party. After judging that the other party hangs up, the next object can be notified automatically, and the method is very convenient and quick. The intelligent outbound system well solves the problems of high investment cost and low working efficiency at the server, enterprises can preset the required voice outbound flow according to business requirements, and the system can realize the flows of identity confirmation, business consultation, business processing and the like of users. However, in the prior art, rule-based intent recognition needs to manually configure rules for each telephone operation node, but the method is excessively dead and has low recognition accuracy, and if a customer telephone operation is slightly changed, errors are recognized, so that the technical problems that the same text is under different nodes of the same telephone operation flow but has different meanings, is difficult to multiplex and cannot be updated in real time cannot be solved.
Aiming at the technical problems, the general thought of the technical proposal provided by the self please is as follows:
The embodiment of the application provides a node reusable intelligent outbound method and a system based on a multi-task model, wherein the method is applied to the node reusable intelligent outbound system based on the multi-task model, and the method comprises the following steps: performing cluster analysis on all the training corpuses by using a cluster algorithm to obtain the training corpuses conforming to a preset rule in each category, and taking the training corpuses as a training set; obtaining a first flow Cheng Shu from a first outbound scene; obtaining a second flow Cheng Shu according to a second outbound scenario, wherein each non-leaf node of the first flow tree and the second flow tree is used as a subtask for multitasking training; initializing a multi-task intention matching model, wherein the multi-task intention matching model comprises a task layer which corresponds to each subtask in the flow tree respectively; training the initialized multi-task intention matching model by using the corpus of the first flow tree through the training set to obtain a first multi-task intention matching model; training and updating task layers different from the subtasks of the first stream Cheng Shu in the initialized first multi-task intention matching model by using the corpus of the second flow tree in the training set to obtain a second multi-task intention matching model; and carrying out intelligent outbound according to the second multitasking intention matching model.
Having described the basic principles of the present application, embodiments of the present application are described below with reference to the accompanying drawings. As one of ordinary skill in the art can know, with the development of technology and the appearance of new scenes, the technical solution provided by the embodiment of the present application is applicable to similar technical problems.
Example 1
As shown in fig. 1, an embodiment of the present application provides a method and a system for node-reusable intelligent outbound based on a multi-task model, where the method is applied to a node-reusable intelligent outbound system based on a multi-task model, and the method includes:
Step S100: performing cluster analysis on all the training corpuses by using a cluster algorithm to obtain the training corpuses conforming to a preset rule in each category, and taking the training corpuses as a training set;
Specifically, the cluster analysis is also called cluster analysis, which is a statistical analysis method for researching, sampling, indexing or classifying problems, and has more similarity between modes in one cluster than modes not in the same cluster based on similarity, namely, before the whole training corpus is obtained, the domain direction to which the system is applied is firstly determined, after the determination, the whole training corpus is obtained, the obtained corpus is subjected to word segmentation processing, the long text is divided into a plurality of words, then the text is divided into words, the words are further converted into vectors, all the words contained in the training corpus are constructed into a word list, the list does not contain repeated vocabulary entries, a vector is built for each text, the dimension of the vector is the same as the dimension of the vocabulary entry list, the value of the vector is the number of times each vocabulary entry in the vocabulary entry list appears in the text, then the vocabulary entries are distributed to the most similar clusters according to the similarity with the cluster centers, the training corpus meeting the preset rule in each category is obtained after the distribution is completed, the training corpus is used as a training set, the obtained training set is used as model training, basic data for speech node setting and intention recognition are carried out, and the obtaining of the training set is the basis for building the multi-task model-based node reusable intelligent outbound system.
Step S200: obtaining a first flow Cheng Shu from a first outbound scene;
step S300: obtaining a second flow Cheng Shu according to a second outbound scenario, wherein each non-leaf node of the first flow tree and the second flow tree is used as a subtask for multitasking training;
Specifically, the first outbound scene refers to selecting any application scene in the application field of the multi-task model-based node reusable intelligent outbound system as the first outbound scene, the second scene is similar to the first scene, the selecting of any application scene in the application field of the multi-task model-based node reusable intelligent outbound system, which is different from the first outbound scene, the obtaining of a first flow tree according to the first outbound scene refers to making a corresponding flow tree according to the specific business handling flow of the first outbound scene, the obtaining of a second flow tree according to the second outbound scene refers to making a corresponding flow tree according to the specific business handling flow of the second outbound scene, wherein the flow tree comprises leaf nodes and non-leaf nodes, the leaf nodes contained in the flow tree are endpoints of the corresponding business flows, and the non-leaf nodes are sub-business in the corresponding business flows under the corresponding application scene, the non-leaf nodes are also sub-business flows to be built, and the multi-task nodes can be more trained under the corresponding business scene, and the multi-task model can be achieved under the corresponding business tree, and the multi-task model.
Step S400: initializing a multi-task intention matching model, wherein the multi-task intention matching model comprises N n task layers T n which respectively correspond to each subtask in the flow tree;
Specifically, the multitasking intention matching model refers to a given multiple service scenarios, wherein some tasks or all tasks have similarity but have low similarity, the task intention matching model aims to assist each other by knowledge contained in the multiple service scenarios to improve performance of completing each service scenario, the initialization multitasking intention matching model refers to firstly making a corresponding multitasking intention matching model according to the first outbound scenario and the second outbound scenario before training the multitasking intention matching model, wherein the multitasking intention matching model comprises N n subtasks, namely N n task layers, the multitasking intention matching model is set based on specific application scenarios, and after training by the training set, the task intention matching model is more intelligent and can perform corresponding operation processing according to service content processed by the task intention matching model.
Step S500: training the initialized multi-task intention matching model by using the corpus of the first flow tree through the training set to obtain a first multi-task intention matching model;
Specifically, training the initialized multi-task intention matching model by using the corpus of the first flow tree in the training set refers to randomly selecting any plurality of objects from the training set which is processed as an initial clustering center, calculating the rest unselected objects, distributing the rest unselected objects to the most similar clusters according to the similarity between the rest unselected objects and the selected objects, namely, distributing the rest unselected objects to the most similar clusters according to the distance between the rest unselected objects and the clustering center, classifying the obtained corpus in the training set according to the similarity, and if the corpus is found to have a cluster center which is more similar to the distance from the center, re-classifying according to a similarity principle until the classification result reaches relative optimal, extracting features of the corpus according to the obtained classification result, further obtaining corresponding feature vectors, and matching different feature vectors corresponding to different task layers to different texts to obtain the first multi-task intention matching model.
Step S600: training and updating task layers different from the subtasks of the first stream Cheng Shu in the initialized first multi-task intention matching model by using the corpus of the second flow tree in the training set to obtain a second multi-task intention matching model;
Specifically, the corpus of the second process tree is used in the training set to perform training update on task layers different from the subtasks of the first stream Cheng Shu in the initialized first multitask intention matching model, namely, a non-leaf node is randomly selected in the second process tree to serve as a subtask of the training, then a data set corresponding to the subtask is used as input data of the training task, intention of a plurality of categories is distinguished from the training set, a plurality of texts are selected to serve as data required for training, the input training text is preprocessed to obtain the feature vectors, distances among the feature vectors are calculated, the maximum value and the minimum value of the distances among the feature vectors in the category of the feature vectors are obtained according to the distances among the feature vectors, namely, the maximum inter-category distance and the minimum inter-category distance are optimized, a model is reselected by calculating the difference value of the maximum inter-category distance and the minimum inter-category distance, a new subtask node and corresponding training data are reselected until an ideal difference value is obtained, the optimization is finished, the intention of the plurality of categories can be obtained, the intention of the task is not required to be processed in a scene by the task scene is extracted through the first multitask intention matching model, and the intention is not processed in the general scene.
Step S700: and carrying out intelligent outbound according to the second multitasking intention matching model.
Specifically, the second multi-task intention matching model is implemented on the basis of the first multi-task intention matching model, the second multi-task intention matching model can configure required corpora for different business scenes, namely different task nodes, business processes under corresponding scenes are completed, the model is updated once after business is executed once, and periodical updating is not needed manually, so that the second multi-task intention matching model has higher reusability, and intelligent outbound can be performed according to the second multi-task intention matching model.
Further, as shown in fig. 2, step S500 further includes:
step S501: obtaining a first subtask task i in the first flow tree;
Step S502: randomly selecting N C intention categories from the first subtask task i, and selecting N S texts from each intention category as a first training set;
Step S503: inputting all training texts in the first training set into a corresponding task layer T i in the multi-task intention matching model to obtain a first feature vector matrix;
step S504: calculating Euclidean distance between each pair of feature vectors in the first feature vector matrix;
step S505: for each of the categories, calculating a maximum intra-category distance and a minimum inter-category distance;
step S506: obtaining a difference between the maximum intra-class distance and the minimum inter-class distance for each of the classes;
step S507: and optimizing the multi-task intention matching model by minimizing the difference between the maximum intra-class distance and the minimum inter-class distance to obtain a first multi-task intention matching model.
Specifically, obtaining a first subtask task i in the first flow tree, namely randomly selecting a non-leaf node from the first flow Cheng Shu as a first task, namely the first subtask task i, randomly selecting N C intention categories from the first subtask task i refers to selecting N C data from data included in task i as an intention classification for the first subtask task i, selecting N S intention labels in each intention category as a first training set, namely selecting N S intention labels for each selected N C intention categories as training data, namely selecting N S texts as a first training set, inputting the first training set into an encoder, namely obtaining N C×NS vectors with a length d, inputting all training texts in the first training set into a corresponding task layer T i in the multi-task intention matching model, namely inputting all training texts in the first training set into the task layer T i corresponding to the first subtask task i to obtain N C×NS feature vectors with the length d which are weighted by the task layer T i, obtaining a first feature vector matrix, calculating Euclidean distance between each feature vector according to the obtained first feature vector matrix, namely calculating distance between each feature vector and other imagination under the same intention category, respectively obtaining the maximum intra-category distance and the minimum inter-category distance in each category for the calculated Euclidean distance between each vector in each category, and then, according to the obtained maximum intra-class distance and the obtained minimum inter-class distance, obtaining the difference between the maximum intra-class distance and the minimum inter-class distance in each class, and optimizing the direction of optimizing the minimum intra-class distance and the minimum inter-class distance as the multitasking intention matching model to obtain a first multitasking intention matching model, wherein the first multitasking intention matching model is obtained by training according to any business scene and is suitable for business processing under the scene.
Further, step S507 further includes:
step S5071: the optimization function of the first multitasking intention matching model is:
Wherein, Optimizing a target difference value for the multitasking intent matching model;
Step S5072: a is the number of the text vector selected in the intention category;
Step S5073: the positive sample text vector number corresponding to p text vector a;
step S5074: n is the negative sample text vector number corresponding to the text vector a;
step S5075: i is the category to which the intention category belongs;
Step S5077: d is the distance between vectors;
N C is the number of intention categories
N S is the number of samples contained for each intent category
In particular, the optimization function of the first multitasking intent-matching model refers to a function constructed by taking the maximum intra-class distance and the minimum inter-class distance in the class as targets for shortening the maximum intra-class distance and the minimum inter-class distance according to the first subtask,Optimizing target difference values for the multitasking intention matching model, wherein a and p, N are respectively the text numbers selected in the intention categories, namely, the target parameters to be optimized for the optimizing function are the differences between the distances between text vectors with the text vector numbers of a and p and the distances between text vectors of a and N, i is the category to which the intention category belongs, D is the distance between vectors, N C is the number of the intention categories, and N S is the number of samples contained in each intention category
That is, the optimization function of the first multi-tasking intent-to-match model may reduce the distance between samples belonging to one category and increase the distance between samples belonging to a different category.
Further, as shown in fig. 3, step S600 further includes:
Step S601: obtaining a non-leaf node 1,node2,node3,node4 of the first flow tree;
step S602: obtaining a non-leaf node 1,node2,node5,node6 of the second flow tree;
Step S603: an encoder of the first multi-tasking intent matching model is used as an encoder of the second multi-tasking intent matching model. And training and updating the task layer T 5,T6 corresponding to the node 5,node6 to obtain a second multi-task intention matching model.
Specifically, obtaining a non-leaf node 1,node2,node3,node4 of the first flow tree, obtaining a non-leaf node 1,node2,node5,node6 of the second flow tree, wherein the non-leaf nodes 1 and 2 of the first flow tree and the second flow tree are identical, and the first multi-task intention matching model is trained through the training set by the first flow tree, that is, the first multi-task intention matching model comprises a feature matrix and 4 and task layers T 1,T2,T3,T4, further, the second multi-task intention matching model uses an encoder of the first multi-task intention matching model to encode, obtain the task layers T 5,T6 corresponding to the nodes 5,node6, and train and update the task layers T 5,T6 to obtain a second multi-task intention matching model, that is, the task layers of the second multi-task intention matching model comprise T 1,T2,T3,T4,T5,T6, the coverage of the model is more comprehensive, and the business content is more detailed.
Further, as shown in fig. 4, step S700 further includes:
step S701: for each non-leaf node i in the first flow Cheng Shu and/or the second flow tree;
Step S702: inputting the unrecognized corpus and the node i into a multi-task intention matching model to obtain a first prediction result;
Step S703: and carrying out intelligent outbound according to the first prediction result.
Specifically, according to each obtained non-leaf node i in the first flow tree and/or the second flow tree, that is, the obtained non-leaf node i may belong to a subtask shared by the first flow Cheng Shu and the second flow tree or belong to a subtask in the first flow tree or the second flow tree, then the obtained non-leaf node i and the unrecognized corpus are input into the multi-task intent matching model to perform intent matching, the multi-task intent matching model calculates and then obtains the belonging intent of the unrecognized corpus, that is, obtains the first prediction result, and the multi-task intent matching model performs outbound according to the obtained first prediction result to perform business process follow-up.
Further, as shown in fig. 5, step S701 further includes:
Step S7011: inputting all texts under the node i into the second multi-task intention matching model to obtain a first feature vector;
Step S7012: inputting the unrecognized corpus into the second multitasking intention recognition model to obtain a second feature vector;
Step S7013: randomly selecting a plurality of corpora for each category under the non-leaf nodes to serve as prototype corpora, wherein the prototype corpora comprises text 1,text2,…,textn and corresponding intention labels y 1,y2,…,yn;
Step S7014: and calculating the Euclidean distance between the first feature vector and the second feature vector, taking a label y j corresponding to the feature vector with the shortest Euclidean distance as the label of the unrecognized corpus, and returning to the label.
Specifically, all texts under the node i are input into the second multi-task intention matching model to obtain a first feature vector, the first feature vector comprises the feature vector of all input texts under the task node, then the unrecognized corpus is input into the second multi-task intention recognition model to obtain a second feature vector, meanwhile, all the corpora contained in the non-leaf node are selected to be used as prototype corpora of the second multi-task intention recognition model, the prototype corpora comprises text 1,text2,…,textn and corresponding intention labels y 1,y2,…,yn, euclidean distance between the first feature vector and the second feature vector is calculated according to the obtained prototype corpora comprising texts and the intention labels corresponding to the prototype corpora, then the label with the shortest Euclidean distance is used as the label of the input unrecognized corpus, the obtained label is returned to the second multi-task intention matching model to be calculated, the task intention matching model can be continuously calculated by returning the label of the unrecognized corpus to the second multi-task intention matching model, and the task intention matching model can be obtained.
Compared with the prior art, the invention has the following beneficial effects:
Performing cluster analysis on all the training corpuses by using a cluster algorithm to obtain the training corpuses conforming to a preset rule in each category, and taking the training corpuses as a training set; obtaining a first flow Cheng Shu from a first outbound scene; obtaining a second flow Cheng Shu according to a second outbound scenario, wherein each non-leaf node of the first flow tree and the second flow tree is used as a subtask for multitasking training; initializing a multi-task intention matching model, wherein the multi-task intention matching model comprises a task layer which corresponds to each subtask in the flow tree respectively; training the initialized multi-task intention matching model by using the corpus of the first flow tree through the training set to obtain a first multi-task intention matching model; training and updating task layers different from the subtasks of the first stream Cheng Shu in the initialized first multi-task intention matching model by using the corpus of the second flow tree in the training set to obtain a second multi-task intention matching model; according to the second multi-task intention matching model, intelligent outbound is performed, corpus and labels required by the free configuration of different nodes of a flow tree are achieved through the multi-task intention classification model, so that intention labels under different nodes are not affected by each other, prototype corpus contained under each node can be directly updated to complete the updating and configuration of the model, and the technical effect of reusability of the model is improved.
Example two
Based on the same inventive concept as the data modeling method based on the intelligent marketing scenario in the foregoing embodiment, the present invention further provides a node reusable intelligent outbound system based on a multitasking model, as shown in fig. 6, the system includes:
A first obtaining unit 11, where the first obtaining unit 11 is configured to obtain the training corpus in each category, where the training corpus meets a predetermined rule, and use the training corpus as a training set;
a second obtaining unit 12, where the second obtaining unit 12 is configured to obtain a first flow Cheng Shu according to a first outbound scene;
a third obtaining unit 13, the third obtaining unit 13 being configured to obtain a second stream Cheng Shu according to a second outbound scene;
A first initializing unit 14, where the first initializing unit 14 is configured to initialize a multitasking intention matching model, where the multitasking intention matching model includes N n task layers T n, which respectively correspond to each of the subtasks in the flow tree;
A fourth obtaining unit 15, where the fourth obtaining unit 15 is configured to train the initialized multi-task intent matching model by using the corpus of the first flow tree in the training set to obtain a first multi-task intent matching model;
A fifth obtaining unit 16, where the fifth obtaining unit 16 is configured to perform training update on task layers different from the subtasks of the first flow Cheng Shu in the initialized first multitasking intention matching model by using the corpus of the second flow tree in the training set, to obtain a second multitasking intention matching model;
the first outbound unit 17 is configured to make an intelligent outbound according to the second multitasking intention matching model.
Further, the system further comprises:
a sixth obtaining unit, configured to obtain a first subtask task i in the first flow tree;
The first selecting unit is used for randomly selecting N C intention categories from the first subtask task i, and N S texts are selected from each intention category to serve as a first training set;
a seventh obtaining unit, configured to input all training texts in the first training set into a task layer T i corresponding to the multi-task intention matching model, to obtain a first feature vector matrix;
The first computing unit is used for computing Euclidean distance between each pair of feature vectors in the first feature vector matrix;
A second calculation unit configured to calculate, for each of the categories, a maximum intra-category distance and a minimum inter-category distance;
An eighth obtaining unit configured to obtain a difference between the maximum intra-class distance and the minimum inter-class distance for each of the classes;
The first optimizing unit is used for optimizing the multi-task intention matching model by minimizing the difference between the distance in the maximum class and the distance between the minimum class, so as to obtain a first multi-task intention matching model.
Further, the system further comprises:
A first function unit, wherein the optimization function of the first function unit for the first multitasking intention matching model is as follows:
Wherein, Optimizing a target difference value for the multitasking intent matching model;
a first numbering unit, configured to a number a text vectors selected in the intent category;
the second numbering unit is used for numbering the text vector selected in the intention category by p;
the third numbering unit is used for numbering n tasks;
a first category unit, wherein the first category unit is used for c being a category to which the intention category belongs;
a first intention unit for y to be an intention tag;
The first distance unit is used for D being the distance between vectors;
and a second category unit, where the second category unit is used for class i as the intention category to which the feature vector belongs.
Further, the system further comprises:
A ninth obtaining unit, configured to obtain a non-leaf node 1,node2,node3,node4 of the first flow tree;
a tenth obtaining unit, configured to obtain a non-leaf node 1,node2,node5,node6 of the second flow tree;
An eleventh obtaining unit for using an encoder of the first multi-tasking intention matching model as an encoder of the second multi-tasking intention matching model. And training and updating the task layer T 5,T6 corresponding to the node 5,node6 to obtain a second multi-task intention matching model.
Further, the system further comprises:
A second selection unit for each non-leaf node i in the first flow Cheng Shu and/or the second flow tree;
A twelfth obtaining unit, configured to input the unrecognized corpus and the node i into a multi-task intent matching model, to obtain a first prediction result;
the first prediction unit is used for performing intelligent outbound according to the first prediction result.
Further, the system further comprises:
The first input unit is used for inputting all texts under the node i into the second multi-task intention matching model to obtain a first feature vector;
the second input unit is used for inputting the unrecognized corpus into the second multi-task intention recognition model to obtain a second feature vector;
And the third selection unit is used for randomly selecting a plurality of corpora for each category under the non-leaf nodes to serve as prototype corpora. The prototype corpus comprises texts, text 1,text2,…,textn and corresponding intention labels y 1,y2,…,yn;
And the third calculation unit is used for calculating the Euclidean distance between the first feature vector and the second feature vector, taking the label y j corresponding to the feature vector with the shortest Euclidean distance as the label of the unrecognized corpus, and returning the label.
The foregoing various modifications and specific examples of a node-reusable intelligent outbound method based on a multi-task model in the first embodiment of fig. 1 are equally applicable to a node-reusable intelligent outbound system based on a multi-task model in this embodiment, and by the foregoing detailed description of a node-reusable intelligent outbound method based on a multi-task model, those skilled in the art can clearly know that a node-reusable intelligent outbound system based on a multi-task model in this embodiment, so that details of this embodiment will not be described herein for brevity.
Example III
An electronic device of an embodiment of the present application is described below with reference to fig. 7.
Fig. 7 illustrates a schematic structural diagram of an electronic device according to an embodiment of the present application.
Based on the inventive concept of the node reusable intelligent outbound method based on the multi-task model in the foregoing embodiments, the present application further provides a node reusable intelligent outbound system based on the multi-task model, and an electronic device according to an embodiment of the present application is described below with reference to fig. 7. The electronic device may be a removable device itself or a stand-alone device independent thereof, having stored thereon a computer program which when executed by a processor performs the steps of any of the methods described hereinbefore.
As shown in fig. 7, the electronic device 50 includes one or more processors 51 and memory 52.
The processor 51 may be a Central Processing Unit (CPU) or other form of processing unit having data processing and/or instruction execution capabilities, and may control other components in the electronic device 50 to perform desired functions.
Memory 52 may include one or more computer program products that may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory may include, for example, random Access Memory (RAM) and/or cache memory (cache), and the like. The non-volatile memory may include, for example, read Only Memory (ROM), hard disk, flash memory, and the like. One or more computer program instructions may be stored on the computer readable storage medium that may be executed by the processor 51 to implement the methods of the various embodiments of the present application described above and/or other desired functions.
In one example, the electronic device 50 may further include: an input device 53 and an output device 54, which are interconnected by a bus system and/or other forms of connection mechanisms (not shown).
The embodiment of the invention provides a node reusable intelligent outbound method based on a multitasking model, which comprises the following steps: performing cluster analysis on all the training corpuses by using a cluster algorithm to obtain the training corpuses conforming to a preset rule in each category, and taking the training corpuses as a training set; obtaining a first flow Cheng Shu from a first outbound scene; obtaining a second flow Cheng Shu according to a second outbound scenario, wherein each non-leaf node of the first flow tree and the second flow tree is used as a subtask for multitasking training; initializing a multi-task intention matching model, wherein the multi-task intention matching model comprises a task layer which corresponds to each subtask in the flow tree respectively; training the initialized multi-task intention matching model by using the corpus of the first flow tree through the training set to obtain a first multi-task intention matching model; training and updating task layers different from the subtasks of the first stream Cheng Shu in the initialized first multi-task intention matching model by using the corpus of the second flow tree in the training set to obtain a second multi-task intention matching model; according to the second multi-task intention matching model, intelligent outbound is performed, corpus and labels required by the free configuration of different nodes of a flow tree are achieved through the multi-task intention classification model, so that intention labels under different nodes are not affected by each other, prototype corpus contained under each node can be directly updated to complete the updating and configuration of the model, and the technical effect of reusability of the model is improved.
From the above description of the embodiments, it will be apparent to those skilled in the art that the present application may be implemented by means of software plus necessary general purpose hardware, or of course by means of special purpose hardware including application specific integrated circuits, special purpose CPUs, special purpose memories, special purpose components, etc. Generally, functions performed by computer programs can be easily implemented by corresponding hardware, and specific hardware structures for implementing the same functions can be varied, such as analog circuits, digital circuits, or dedicated circuits. But a software program implementation is a preferred embodiment for many more of the cases of the present application. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a readable storage medium, such as a floppy disk, a usb disk, a removable hard disk, a ROM, a RAM, a magnetic disk or an optical disk of a computer, etc., comprising several instructions for causing a computer device to execute the method according to the embodiments of the present application.
In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product.
The computer program product includes one or more computer instructions. When loaded and executed on a computer, produces a flow or function in accordance with embodiments of the present application, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may be stored in or transmitted from a computer-readable storage medium, which may be magnetic media, (e.g., floppy disks, hard disks, magnetic tape), optical media (e.g., DVD), or semiconductor media (e.g., solid state disk (Solid STATE DISK, SSD)), among others.
It should be appreciated that reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present application. Thus, the appearances of the phrases "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. It should be understood that, in various embodiments of the present application, the sequence numbers of the foregoing processes do not mean the order of execution, and the order of execution of the processes should be determined by the functions and internal logic thereof, and should not constitute any limitation on the implementation process of the embodiments of the present application.
In addition, the terms "system" and "network" are often used interchangeably herein. The term "and/or" is herein merely an association relationship describing an associated object, meaning that there may be three relationships, e.g., a and/or B, may represent: a exists alone, A and B exist together, and B exists alone. In addition, the character "/" herein generally indicates that the front and rear associated objects are an "or" relationship.
It should be understood that in embodiments of the present application, "B corresponding to a" means that B is associated with a, from which B may be determined. It should also be understood that determining B from a does not mean determining B from a alone, but may also determine B from a and/or other information.
Those of ordinary skill in the art will appreciate that the elements and algorithm steps described in connection with the embodiments disclosed herein may be embodied in electronic hardware, in computer software, or in a combination of the two, and that the elements and steps of the examples have been generally described in terms of function in the foregoing description to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
In summary, the foregoing description is only a preferred embodiment of the present application, and is not intended to limit the scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.
Claims (8)
1. A method of node-reusable intelligent outbound based on a multitasking model, wherein the method comprises:
Performing cluster analysis on all the training corpuses by using a cluster algorithm to obtain the training corpuses conforming to a preset rule in each category, and taking the training corpuses as a training set;
Obtaining a first flow Cheng Shu from a first outbound scene;
Obtaining a second flow Cheng Shu according to a second outbound scenario, wherein each non-leaf node of the first flow tree and the second flow tree is used as a subtask for multitasking training;
initializing a multi-task intent matching model, wherein the multi-task intent matching model comprises N n task layers T n, and each subtask in the first flow tree and the second flow tree is respectively corresponding to each subtask;
Training the initialized multi-task intention matching model by using the corpus of the first flow tree through the training set to obtain a first multi-task intention matching model;
Training and updating task layers different from the subtasks of the first stream Cheng Shu in the initialized first multi-task intention matching model by using the corpus of the second flow tree in the training set to obtain a second multi-task intention matching model;
and carrying out intelligent outbound according to the second multitasking intention matching model.
2. The method of claim 1, wherein the training the initialized multi-tasking intent matching model by the training set using the corpus of the first flow tree to obtain a first multi-tasking intent matching model comprises:
Obtaining a first subtask task i in the first flow tree; randomly selecting N C intention categories from the first subtask task i, and selecting N S texts from each intention category as a first training set;
inputting all training texts in the first training set into a corresponding task layer T i in the multi-task intention matching model to obtain a first feature vector matrix;
calculating Euclidean distance between each pair of feature vectors in the first feature vector matrix;
For each of the categories, calculating a maximum intra-category distance and a minimum inter-category distance;
Obtaining a difference between the maximum intra-class distance and the minimum inter-class distance for each of the classes;
And optimizing the multi-task intention matching model by minimizing the difference between the maximum intra-class distance and the minimum inter-class distance to obtain a first multi-task intention matching model.
3. The method of claim 2, wherein the optimizing the multi-tasking intent-to-match model by minimizing a difference between the maximum intra-class distance and the minimum inter-class distance to obtain a first multi-tasking intent-to-match model comprises:
the optimization function of the first multitasking intention matching model is:
Wherein, Optimizing a target difference value for the multitasking intent matching model;
a is the number of the text vector selected in the intention category;
the positive sample text vector number corresponding to p text vector a;
n is the negative sample text vector number corresponding to the text vector a;
i is the category to which the intention category belongs;
d is the distance between vectors;
N C is the number of intention categories;
N S is the number of samples contained for each intent category.
4. The method of claim 1, wherein the training updating, by the training set, the task layer of the initialized first multitasking intent-matching model that is different from the subtasks of the first flow Cheng Shu using the corpus of the second flow tree to obtain a second multitasking intent-matching model, comprises:
obtaining a non-leaf node 1,node2,node3,node4 of the first flow tree;
Obtaining a non-leaf node 1,node2,node5,node6 of the second flow tree;
And training and updating the task layer T 5,T6 corresponding to the node 5,node6 by using the encoder of the first multi-task intention matching model as the encoder of the second multi-task intention matching model to obtain a second multi-task intention matching model.
5. The method of claim 1, wherein the method further comprises:
For each non-leaf node i in the first flow Cheng Shu and/or the second flow tree,
Inputting the unrecognized corpus and the node i into a multi-task intention matching model to obtain a first prediction result;
and carrying out intelligent outbound according to the first prediction result.
6. The method of claim 5, wherein the method further comprises:
inputting all texts under the node i into the second multi-task intention matching model to obtain a first feature vector;
Inputting the unrecognized corpus into the second multitasking intention recognition model to obtain a second feature vector;
Randomly selecting a plurality of corpora for each category under the non-leaf nodes to serve as prototype corpora, wherein the prototype corpora comprise texts, text 1,text2,…,textn and corresponding intention labels y 1,y2,…,yn;
calculating the Euclidean distance between the first feature vector and the second feature vector, taking a label y j corresponding to the feature vector with the shortest Euclidean distance as the label of the unrecognized corpus, and returning to the label;
If the number of the non-leaf nodes of the flow tree is N n, the number of the subtasks is N n, and training data which is used for each subtask is allocated, wherein the training data comprises text, text 1,text2,…,textn and corresponding intention label y 1,y2,…,yn.
7. A multi-tasking model based node reusable intelligent outbound system, wherein the system comprises:
The first obtaining unit is used for obtaining training corpus which accords with a preset rule in each category and taking the training corpus as a training set;
a second obtaining unit for obtaining a first flow Cheng Shu according to a first outbound scene;
a third obtaining unit, configured to obtain a second flow Cheng Shu according to a second outbound scenario, where each non-leaf node of the first flow tree and the second flow tree is used as a subtask of multitasking;
The first initializing unit is used for initializing a multi-task intention matching model, and the multi-task intention matching model comprises N n task layers T n which respectively correspond to each subtask in the first flow tree and the second flow tree;
The fourth obtaining unit is used for training the initialized multi-task intention matching model by using the corpus of the first flow tree in the training set to obtain a first multi-task intention matching model;
A fifth obtaining unit, configured to perform training update on a task layer different from the subtasks of the first flow Cheng Shu in the initialized first multitasking intention matching model by using the corpus of the second flow tree in the training set, to obtain a second multitasking intention matching model;
and the first outbound unit is used for performing intelligent outbound according to the second multitasking intention matching model.
8. A node reusable intelligent outbound system based on a multitasking model comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method of any of claims 1-6 when the program is executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111276483.4A CN113792819B (en) | 2021-10-29 | 2021-10-29 | Multi-task model-based node reusable intelligent outbound method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111276483.4A CN113792819B (en) | 2021-10-29 | 2021-10-29 | Multi-task model-based node reusable intelligent outbound method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113792819A CN113792819A (en) | 2021-12-14 |
| CN113792819B true CN113792819B (en) | 2024-05-14 |
Family
ID=78878376
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111276483.4A Active CN113792819B (en) | 2021-10-29 | 2021-10-29 | Multi-task model-based node reusable intelligent outbound method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113792819B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110233946A (en) * | 2019-06-17 | 2019-09-13 | 三角兽（北京）科技有限公司 | Execute outbound call service method, electronic equipment and computer readable storage medium |
| CN110276074A (en) * | 2019-06-20 | 2019-09-24 | 出门问问信息科技有限公司 | Distributed training method, device, equipment and the storage medium of natural language processing |
| CN111696576A (en) * | 2020-05-21 | 2020-09-22 | 升智信息科技(南京)有限公司 | Intelligent voice robot talk test system |
| CN112185358A (en) * | 2020-08-24 | 2021-01-05 | 维知科技张家口有限责任公司 | Intention recognition method, model training method, device, equipment and medium |
| CN113094481A (en) * | 2021-03-03 | 2021-07-09 | 北京智齿博创科技有限公司 | Intention recognition method and device, electronic equipment and computer readable storage medium |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9992335B2 (en) * | 2016-10-28 | 2018-06-05 | Microsoft Technology Licensing, Llc | Caller assistance system |
| US10839214B2 (en) * | 2018-03-13 | 2020-11-17 | International Business Machines Corporation | Automated intent to action mapping in augmented reality environments |
-
2021
- 2021-10-29 CN CN202111276483.4A patent/CN113792819B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110233946A (en) * | 2019-06-17 | 2019-09-13 | 三角兽（北京）科技有限公司 | Execute outbound call service method, electronic equipment and computer readable storage medium |
| CN110276074A (en) * | 2019-06-20 | 2019-09-24 | 出门问问信息科技有限公司 | Distributed training method, device, equipment and the storage medium of natural language processing |
| CN111696576A (en) * | 2020-05-21 | 2020-09-22 | 升智信息科技(南京)有限公司 | Intelligent voice robot talk test system |
| CN112185358A (en) * | 2020-08-24 | 2021-01-05 | 维知科技张家口有限责任公司 | Intention recognition method, model training method, device, equipment and medium |
| CN113094481A (en) * | 2021-03-03 | 2021-07-09 | 北京智齿博创科技有限公司 | Intention recognition method and device, electronic equipment and computer readable storage medium |
Non-Patent Citations (2)
| Title |
|---|
| Joint Multiple Intent Detection and Slot Filling Via Self-Distillation;Lisong Chen等;arXiv:2108.08042v1;20210818;全文 * |
| 智能机器外呼系统的设计与实现;张庆恒;中国优秀硕士学位论文全文数据库 信息科技辑;20190815(第08期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113792819A (en) | 2021-12-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109582793B (en) | Model training method, customer service system, data labeling system and readable storage medium | |
| JP7266674B2 (en) | Image classification model training method, image processing method and apparatus | |
| US10262272B2 (en) | Active machine learning | |
| CN108076154A (en) | Application message recommends method, apparatus and storage medium and server | |
| CN110569870A (en) | deep acoustic scene classification method and system based on multi-granularity label fusion | |
| CN111783873A (en) | Incremental naive Bayes model-based user portrait method and device | |
| CN113488023A (en) | Language identification model construction method and language identification method | |
| WO2020114109A1 (en) | Interpretation method and apparatus for embedding result | |
| US12001174B2 (en) | Determination of task automation using an artificial intelligence model | |
| CN111310462A (en) | User attribute determination method, device, equipment and storage medium | |
| Chamoso et al. | Social computing for image matching | |
| CN113792819B (en) | Multi-task model-based node reusable intelligent outbound method and system | |
| WO2020095655A1 (en) | Selection device and selection method | |
| CN115905293A (en) | Switching method and device of job execution engine | |
| CN115146653A (en) | Dialogue script construction method, device, equipment and storage medium | |
| CN111241826B (en) | Entity name recognition method, device, equipment and storage medium | |
| CN114372148A (en) | Data processing method based on knowledge graph technology and terminal equipment | |
| AU2021301463A1 (en) | Method and system for generating an ai model using constrained decision tree ensembles | |
| CN111382246A (en) | Text matching method, matching device and terminal | |
| WO2023119672A1 (en) | Inference method, inference device, and inference program | |
| CN112328709B (en) | Entity labeling method and device, server and storage medium | |
| Loong et al. | Image‐based structural analysis for education purposes: A proof‐of‐concept study | |
| CN117392658B (en) | Attention mechanism optimization-based light-weight vehicle license plate color recognition method | |
| CN114826921B (en) | Dynamic network resource allocation method, system and medium based on sampling subgraph | |
| CN116029492B (en) | Order sending method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information |
Address after: 100000 floors 1-3, block a, global creative Plaza, No. 10, Furong street, Chaoyang District, Beijing Applicant after: Bairong Zhixin (Beijing) Technology Co.,Ltd. Address before: 100000 floors 1-3, block a, global creative Plaza, No. 10, Furong street, Chaoyang District, Beijing Applicant before: Bairong Zhixin (Beijing) credit investigation Co.,Ltd. |
|
| CB02 | Change of applicant information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |