Query system and generation method based on NLP engine and online data recall

Abstract

The present invention relates to an intelligent customer service system, and specifically to an inquiry system and a generation method based on an NLP engine and online data recall, which can not only improve the system's intent recognition capability according to industry characteristics, but also effectively expand the system's application scope; it supports users to create custom intent models and independently import industry basic data, and at the same time uses a training engine to train the intent model to obtain a corresponding intent model, thereby achieving an accurate understanding of user intent; for intent models with low answer matching, the model can be optimized by manually secondary labeling of online conversation data, adjusting nodes and slots in the intent model, and performing secondary model training on the intent model, so as to effectively improve the accuracy of understanding user intent; the technical solution provided by the present invention can effectively overcome the defects of the prior art that the user intent cannot be accurately understood and the application scope is narrow.

Description

Technical Field

[0001] The present invention relates to an intelligent customer service system, and in particular to an inquiry system and a generation method based on an NLP engine and online data recall. Background Art

[0002] With the in-depth development in logistics, finance, insurance and other service industries, enterprises have accumulated a large amount of data in actual operations. In the service industry, users often ask various professional questions. How to maintain and manage industry knowledge content and help agents quickly find relevant answers to customer inquiries has become an urgent problem to be solved in various service industries.

[0003] Robot customer service is increasingly favored by various service industries due to its long-term online and low-cost features. Online robot customer service responds faster than human customer service and can handle multiple users at the same time. It does not have the emotional problems of human customer service and can provide customers with a better inquiry experience.

[0004] At present, the intelligent customer service systems on the market cannot reasonably abstract various scenarios and apply them to various service industries. For example, different customers may ask the same question in different ways. How to reduce questions with different questions to the same standard answer, effectively understand customer intentions, and organize the words and recommend answers based on customer intentions cannot be solved well in the current intelligent customer service systems. In addition, for each service industry, the questions raised by customers may involve different business scenarios and business knowledge, which also limits the scope of application of intelligent customer service systems. Summary of the invention

[0005] 1. Technical issues to be solved

[0006] In view of the above-mentioned shortcomings of the prior art, the present invention provides a query system and a generation method based on an NLP engine and online data recall, which can effectively overcome the defects of the prior art that the user intention cannot be accurately understood and the application scope is narrow.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] The query system based on the NLP engine and online data recall includes an intent management unit, a data annotation module, a model training module, and an intent model call recording module;

[0010] The intent management unit is used for users to create custom intent models and enter basic industry data, match the model architecture and annotation data of the intent model, and import them into the model training module;

[0011] The model training module uses the training engine to train the intent model based on the model architecture of the intent model and the corresponding annotated data. It also uses the training engine to match answers to online conversation data and optimize the model based on the matching score.

[0012] The data annotation module manually annotates the recalled online conversation data and inputs the annotated conversation data into the model training module;

[0013] The intent model call recording module is used to record and store the relevant usage data generated after the intent model is called, and generate corresponding call logs for users to query.

[0014] Preferably, the intention management unit includes an intention model management module, a node management module, a slot management module and an import module;

[0015] The intent model management module allows users to create, modify, and delete custom intent models and enter industry basic data. It also allows users to add industry intent models when creating new custom intent models.

[0016] The node management module allows users to create, modify, and delete nodes that are used to mark the true meaning of industry basic data;

[0017] The slot management module allows users to create, modify, and delete slots used to mark important metadata in industry basic data;

[0018] Import the module, match the model architecture and annotation data corresponding to the custom intent model and industry intent model, and import the model training module.

[0019] Preferably, when the user creates a new custom intent model through the intent model management module and enters industry basic data, the import module matches the nodes, slots and annotation data formed by each node and slot in the custom intent model, and imports them into the model training module.

[0020] Preferably, when a user adds an industry intent model through the intent model management module, the corresponding nodes, slots and annotation data under the industry intent model are automatically deployed, and the import module matches the nodes, slots and annotation data in the industry intent model and imports them into the model training module.

[0021] Preferably, the model training module creates a model training task for the model architecture of the intent model and the corresponding annotation data, and uses a training engine to train the corresponding intent model.

[0022] Preferably, the using a training engine to train a corresponding intent model includes:

[0023] The model training module calls the NLP engine to train the corresponding intent model in elasticsearch based on the provided data information.

[0024] Preferably, the model training module uses a training engine to match and score the answers to the online conversation data in each intent model. If the highest matching score is less than a score threshold, the model training module optimizes the intent model with the highest matching score.

[0025] Preferably, the model training module performs model optimization on the intent model with the highest matching score, including:

[0026] The model training module creates model optimization tasks for online conversation data, and uses the training engine to optimize the intent model based on the secondary annotated conversation data generated by the data annotation module and the adjustment of nodes and slots in the intent model.

[0027] Preferably, the intention model call recording module records and stores relevant usage data generated after the intention model is called, and the relevant usage data includes: the user's actual question, the intention model recommended answer, and the matching score between the recommended answer and the actual question.

[0028] The method for generating a query system based on an NLP engine and online data recall includes the following steps:

[0029] S1. The user creates a custom intent model in the intent management unit, as well as the node and slot information in the custom intent model, and imports the industry basic data into the corresponding nodes and slots;

[0030] S2, the model training module creates a model training task for the model architecture of the intent model and the corresponding annotation data, and uses the training engine to train the corresponding intent model;

[0031] S3. The model training module uses the training engine to match and score the answers of the online conversation data in each intent model. If the highest matching score is less than the score threshold, the model training module optimizes the intent model with the highest matching score.

[0032] S4: The model training module creates a model optimization task for the online conversation data, and optimizes the intent model using the training engine based on the secondary annotated conversation data generated by the data annotation module and the adjustment of nodes and slots in the intent model;

[0033] S5. The intent model call recording module records and stores the relevant usage data generated after the intent model is called, and generates a corresponding call log for users to query.

[0034] (III) Beneficial effects

[0035] Compared with the prior art, the query system and generation method based on NLP engine and online data recall provided by the present invention have the following beneficial effects:

[0036] 1) By inputting the industry basic data that has been generated and accumulated over a long period of time in various industries, and adding industry intent models that conform to the actual business scenarios and business knowledge of the enterprise, it can not only improve the system's intent recognition ability based on industry characteristics, but also effectively expand the application scope of the system;

[0037] 2) Support users to create custom intent models and import industry basic data independently. At the same time, use the training engine to train the intent model to obtain the corresponding intent model and achieve accurate understanding of user intent;

[0038] 3) For intent models with low answer matching, the model can be optimized by manually re-annotating online conversation data, adjusting the nodes and slots in the intent model, and performing secondary model training on the intent model to effectively improve the accuracy of understanding user intent. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0040] Figure 1 It is a system schematic diagram of the present invention;

[0041] Figure 2 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Query systems based on NLP engines and online data recall, such as Figure 1 As shown, it includes an intent management unit, a data annotation module, a model training module and an intent model call recording module;

[0044] The intent management unit is used for users to create custom intent models and enter basic industry data, match the model architecture and annotation data of the intent model, and import them into the model training module;

[0045] The model training module uses the training engine to train the intent model based on the model architecture of the intent model and the corresponding annotated data. It also uses the training engine to match answers to online conversation data and optimize the model based on the matching score.

[0046] The data annotation module manually annotates the recalled online conversation data and inputs the annotated conversation data into the model training module;

[0047] The intent model call recording module is used to record and store the relevant usage data generated after the intent model is called, and generate corresponding call logs for users to query.

[0048] The intent management unit includes an intent model management module, a node management module, a slot management module and an import module;

[0049] The intent model management module allows users to create, modify, and delete custom intent models and enter industry basic data. It also allows users to add industry intent models when creating new custom intent models.

[0050] The node management module allows users to create, modify, and delete nodes that are used to mark the real meaning of industry basic data (such as querying addresses, placing orders, etc.);

[0051] The slot management module allows users to create, modify, and delete slots used to mark important metadata (such as addresses, names, etc.) in industry basic data;

[0052] Import the module, match the model architecture and annotation data corresponding to the custom intent model and industry intent model, and import the model training module.

[0053] Users can enter the model architecture and corresponding annotation data for training the intent model in the intent management unit in the following two ways:

[0054] 1) When a user creates a custom intent model through the intent model management module and enters industry basic data, the import module matches the nodes, slots and annotation data formed by each node and slot in the custom intent model and imports them into the model training module;

[0055] 2) When a user adds an industry intent model through the intent model management module, the corresponding nodes, slots and annotation data under the industry intent model are automatically deployed. The import module matches the nodes, slots and annotation data in the industry intent model and imports them into the model training module.

[0056] In the technical solution of the present application, users can create custom intent models in the intent management unit and enter the basic industry data that has been generated and accumulated for a long time in various industries. At the same time, they can also add industry intent models that conform to the actual business scenarios and business knowledge of the enterprise in the intent management unit, and automatically deploy the corresponding nodes, slots and annotation data under the industry intent model. This can not only improve the system's intent recognition capabilities based on industry characteristics, but also effectively expand the application scope of the system.

[0057] The model training module creates model training tasks based on the model architecture of the intent model and the corresponding annotated data, and uses the training engine to train the corresponding intent model.

[0058] Among them, the corresponding intent model is trained using the training engine, including:

[0059] The model training module calls the NLP engine to train the corresponding intent model in elasticsearch based on the provided data information.

[0060] The model training module uses the training engine to match and score the answers to the online conversation data in each intent model. If the highest matching score is less than the score threshold, the model training module optimizes the intent model with the highest matching score.

[0061] The model training module optimizes the intent model with the highest matching score, including:

[0062] The model training module creates model optimization tasks for online conversation data, and uses the training engine to optimize the intent model based on the secondary annotated conversation data generated by the data annotation module and the adjustment of nodes and slots in the intent model.

[0063] In the technical solution of the present application, after the intent model training is completed, the intent model of that version can be released. The system supports the training and optimization of multiple versions of the model. Each training and optimization will be assigned a version number. The system can switch and deploy the intent model between multiple released versions.

[0064] The intent model call recording module records and stores relevant usage data generated after the intent model is called. The relevant usage data includes: the user's actual question, the answer recommended by the intent model, and the matching score between the recommended answer and the actual question.

[0065] In the technical solution of this application, the model training module automatically optimizes the intent model based on the matching scores of the online conversation data in each intent model. Users can retrieve the corresponding call log of the intent model through conditional queries (such as matching score intervals, etc.). When users are not satisfied with the matching scores, they can also manually create model optimization tasks for online conversation data through the model training module, and manually re-annotate the recalled online conversation data through the data annotation module.

[0066] The generation method of the query system based on NLP engine and online data recall, such as Figure 2 As shown, the following steps are included:

[0067] S1. The user creates a custom intent model in the intent management unit, as well as the node and slot information in the custom intent model, and imports the industry basic data into the corresponding nodes and slots;

[0068] S2, the model training module creates a model training task for the model architecture of the intent model and the corresponding annotation data, and uses the training engine to train the corresponding intent model;

[0069] S3. The model training module uses the training engine to match and score the answers of the online conversation data in each intent model. If the highest matching score is less than the score threshold, the model training module optimizes the intent model with the highest matching score.

[0070] S4: The model training module creates a model optimization task for the online conversation data, and optimizes the intent model using the training engine based on the secondary annotated conversation data generated by the data annotation module and the adjustment of nodes and slots in the intent model;

[0071] S5. The intent model call recording module records and stores the relevant usage data generated after the intent model is called, and generates a corresponding call log for users to query.

[0072] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. The query system based on NLP engine and online data recall is characterized by: It includes an intent management unit, a data annotation module, a model training module, and an intent model call recording module; The intent management unit is used for users to create custom intent models and enter basic industry data, match the model architecture and annotation data of the intent model, and import them into the model training module; The model training module uses the training engine to train the intent model based on the model architecture of the intent model and the corresponding annotated data. It also uses the training engine to match answers to online conversation data and optimize the model based on the matching score. The data annotation module manually annotates the recalled online conversation data and inputs the annotated conversation data into the model training module; The intent model call recording module is used to record and store the relevant usage data generated after the intent model is called, and generate the corresponding call log for users to query; The intention management unit includes an intention model management module, a node management module, a slot management module and an import module; The intent model management module allows users to create, modify, and delete custom intent models and enter basic industry data. It also allows users to add industry intent models when creating new custom intent models. The node management module allows users to create, modify, and delete nodes that are used to mark the true meaning of industry basic data; The slot management module allows users to create, modify, and delete slots used to mark important metadata in industry basic data; Import the module, match the model architecture and annotation data corresponding to the custom intent model and industry intent model, and import the model training module.

2. The query system based on NLP engine and online data recall according to claim 1, characterized in that: When the user creates a new custom intent model and enters industry basic data through the intent model management module, the import module matches the nodes, slots and annotation data formed by each node and slot in the custom intent model, and imports them into the model training module.

3. The query system based on NLP engine and online data recall according to claim 1, characterized in that: When a user adds an industry intent model through the intent model management module, the corresponding nodes, slots and annotation data under the industry intent model are automatically deployed. The import module matches the nodes, slots and annotation data in the industry intent model and imports them into the model training module.

4. The query system based on NLP engine and online data recall according to claim 1, characterized in that: The model training module creates a model training task for the model architecture of the intent model and the corresponding annotation data, and uses the training engine to train the corresponding intent model.

5. The query system based on NLP engine and online data recall according to claim 4, characterized in that: The using the training engine to train the corresponding intent model includes: The model training module calls the NLP engine to train the corresponding intent model in elasticsearch based on the provided data information.

6. The query system based on NLP engine and online data recall according to claim 4, characterized in that: The model training module uses a training engine to match and score the answers to the online conversation data in each intent model. If the highest matching score is less than the score threshold, the model training module optimizes the intent model with the highest matching score.

7. The query system based on NLP engine and online data recall according to claim 6, characterized in that: The model training module optimizes the intent model with the highest matching score, including: The model training module creates model optimization tasks for online conversation data, and uses the training engine to optimize the intent model based on the secondary annotated conversation data generated by the data annotation module and the adjustment of nodes and slots in the intent model.

8. The query system based on NLP engine and online data recall according to claim 1, characterized in that: The intention model call recording module records and stores relevant usage data generated after the intention model is called, and the relevant usage data includes: the user's actual question, the answer recommended by the intention model, and the matching score between the recommended answer and the actual question.

9. A method for generating an inquiry system based on an NLP engine and online data recall, characterized in that: The following steps are involved: S1. The user creates a custom intent model in the intent management unit, as well as the node and slot information in the custom intent model, and imports the industry basic data into the corresponding nodes and slots; S2, the model training module creates a model training task for the model architecture of the intent model and the corresponding annotation data, and uses the training engine to train the corresponding intent model; S3. The model training module uses the training engine to match and score the answers of the online conversation data in each intent model. If the highest matching score is less than the score threshold, the model training module optimizes the intent model with the highest matching score. S4: The model training module creates a model optimization task for the online conversation data, and optimizes the intent model using the training engine based on the secondary annotated conversation data generated by the data annotation module and the adjustment of nodes and slots in the intent model; S5. The intent model call recording module records and stores the relevant usage data generated after the intent model is called, and generates a corresponding call log for users to query.

Images