CN113961687A - Multi-turn question-answering system intention classification and named entity identification research method - Google Patents
Multi-turn question-answering system intention classification and named entity identification research method Download PDFInfo
- Publication number
- CN113961687A CN113961687A CN202111238190.7A CN202111238190A CN113961687A CN 113961687 A CN113961687 A CN 113961687A CN 202111238190 A CN202111238190 A CN 202111238190A CN 113961687 A CN113961687 A CN 113961687A
- Authority
- CN
- China
- Prior art keywords
- named entity
- intention classification
- model
- classification
- intention
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
The invention provides a research method for intention classification and named entity identification of a multi-turn question-answering system. Which comprises the following steps: 1) when the intention classification is carried out, the historical information of the context is used as input through a dialogue state tracking module, and an intention classification prediction model is transmitted; 2) during named entity recognition and prediction, the intention recognition result of the round is used as a feature and is transmitted into a named entity prediction model, and 3) the named entity recognition and intention classification are trained through a multi-task model, and the model is improved according to the requirements of the first two points. According to the invention, NLU and DM are directly changed into bidirectional models, the accuracy of intention classification can be improved by context information, named entity identification and intention classification are fused through a multi-task model Bert model, and deployment can be conveniently carried out.
Description
Technical Field
The invention relates to an intelligent question-answering system, in particular to a research method for intention classification and named entity identification of a multi-turn question-answering system, and belongs to the technical field of artificial intelligence.
Background
Referring to fig. 1, a conventional task-based question-answering system adopts a pipeline structure, a natural speech understanding module (NLU) is a main component of the task-based question-answering system, named entity recognition and intention classification are respectively used for extracting entities in a dialog and classifying dialog intentions in the NLU, and the direction from a natural language understanding module (NLU) to a dialog management module (DM) is unidirectional.
In addition, the existing intelligent dialogue platform has two defects when the NLU is realized. First, existing platforms do not consider the intent of the text as a feature when doing named entity recognition. Second, the impact of the contextual information on intent classification is not considered when doing intent classification.
Disclosure of Invention
The invention aims to provide a multi-turn question-answering system intention classification and named entity identification research method.
In order to achieve the purpose, the invention is realized by the following technical scheme:
a multi-turn question-answering system intention classification and named entity identification research method comprises the following steps:
1) when the intention classification is carried out, the historical information of the context is used as input through a dialogue state tracking module, and an intention classification prediction model is transmitted;
2) when the named entity is identified and predicted, the intention identification result of the round is used as a characteristic and is transmitted into a named entity prediction model;
3) named entity recognition and intent classification are trained through a multi-tasking model, and the model is refined as required by the first two points.
The optimal scheme of the research method for intention classification and named entity recognition of the multi-turn question-answering system is characterized in that a multi-task framework of a Bert model is used, a [ CLS ] mark bit is added at the beginning of each input sentence of the Bert, the mark bit serves as a prediction result of intention classification, a conditional random field CRF is added on an output layer of the Bert to serve as prediction output of named entity recognition, and hidden layer information of the intention classification serves as input to serve as a part of named entity recognition prediction.
The preferable scheme of the research method for intention classification and named entity identification of the multi-turn question-answering system specifically comprises the following steps:
s1, taking DST context data in a DM module as one of inputs of a Bert model;
s2, fusing the named entity identification module and the intention classification module by using a Bert multitask;
and S3, transmitting the intention classification result state as an input to a CRF model for prediction.
The invention has the advantages that:
NLU and DM are directly changed into a bidirectional model, the accuracy of intention classification can be improved by context information, named entity identification and intention classification are fused through a multi-task model Bert model, and deployment can be conveniently carried out.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention.
Fig. 1 is a flow chart of an intelligent dialog in an embodiment of the present invention.
Fig. 2 is a diagram of an intelligent dialog improvement in an embodiment of the present invention.
FIG. 3 is a graph of multitasking berts in an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As described in the background art, the conventional task-based question-answering system adopts a pipeline structure, and is unidirectional from a natural language understanding module (NLU) to a dialogue management module (DM), so that the problem of inaccurate intention classification exists; the named entity recognition task (NER) and the Intent classification task (Intent Classifier) are separate and complex to lay out.
In order to solve the problems, the invention adopts the following scheme:
a multi-turn question-answering system intention classification and named entity identification research method comprises the following steps:
1) when the intention classification is carried out, the historical information of the context is used as input through a dialogue state tracking module, and an intention classification prediction model is transmitted;
2) when the named entity is identified and predicted, the intention identification result of the round is used as a characteristic and is transmitted into a named entity prediction model;
3) named entity recognition and intent classification are trained through a multi-tasking model, and the model is refined as required by the first two points.
The embodiment of the invention specifically comprises the following steps:
1. directly changing NLU and DM into bidirectional models, so that an intention classification model can acquire context information of Dialog State Tracking (DST) in classification;
2. adopting a multi-classification task of Bert, fusing a named entity recognition task (NER) and an intention classification task (Intent Classifier) through a model, and classifying the CLS by softmax through the Attention layer of Bert after the fused model passes through the Attention layer of Bert to obtain a final intention;
3. and the hidden state CLS of the intention is used as input and transmitted to a CRF layer, so that when the CRF layer identifies the named entity, the CRF layer outputs the identification result of the named entity in consideration of the intention of the corresponding round.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (3)
1. A multi-turn question-answering system intention classification and named entity recognition research method is characterized in that: the method comprises the following steps:
1) when the intention classification is carried out, the historical information of the context is used as input through a dialogue state tracking module, and an intention classification prediction model is transmitted;
2) when the named entity is identified and predicted, the intention identification result of the round is used as a characteristic and is transmitted into a named entity prediction model;
3) named entity recognition and intent classification are trained through a multi-tasking model, and the model is refined as required by the first two points.
2. The multi-turn question-answering system intention classification and named entity identification research method of claim 1, wherein: the method comprises the steps of using a multitask architecture of a Bert model, adding a [ CLS ] identification bit at the beginning of each sentence input by the Bert, using the identification bit as a prediction result of an intention classification, adding a conditional random field CRF on an output layer of the Bert as a prediction output of named entity recognition, and using hidden layer information of the intention classification as input to serve as a part of named entity recognition prediction.
3. The multi-turn question-answering system intention classification and named entity identification research method of claim 2, wherein: the method specifically comprises the following steps:
s1, taking DST context data in a DM module as one of inputs of a Bert model;
s2, fusing the named entity identification module and the intention classification module by using a Bert multitask;
and S3, transmitting the intention classification result state as an input to a CRF model for prediction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111238190.7A CN113961687A (en) | 2021-10-25 | 2021-10-25 | Multi-turn question-answering system intention classification and named entity identification research method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111238190.7A CN113961687A (en) | 2021-10-25 | 2021-10-25 | Multi-turn question-answering system intention classification and named entity identification research method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113961687A true CN113961687A (en) | 2022-01-21 |
Family
ID=79466455
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111238190.7A Pending CN113961687A (en) | 2021-10-25 | 2021-10-25 | Multi-turn question-answering system intention classification and named entity identification research method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113961687A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116010583A (en) * | 2023-03-17 | 2023-04-25 | 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) | Cascade coupling knowledge enhancement dialogue generation method |
-
2021
- 2021-10-25 CN CN202111238190.7A patent/CN113961687A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116010583A (en) * | 2023-03-17 | 2023-04-25 | 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) | Cascade coupling knowledge enhancement dialogue generation method |
| CN116010583B (en) * | 2023-03-17 | 2023-07-18 | 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) | Cascade coupling knowledge enhancement dialogue generation method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108829722B (en) | Remote supervision Dual-Attention relation classification method and system | |
| CN110321418B (en) | Deep learning-based field, intention recognition and groove filling method | |
| WO2022057712A1 (en) | Electronic device and semantic parsing method therefor, medium, and human-machine dialog system | |
| WO2018028077A1 (en) | Deep learning based method and device for chinese semantics analysis | |
| CN111062217B (en) | Language information processing method and device, storage medium and electronic equipment | |
| CN112100349A (en) | Multi-turn dialogue method and device, electronic equipment and storage medium | |
| CN111402861B (en) | Voice recognition method, device, equipment and storage medium | |
| CN113326367B (en) | Task type dialogue method and system based on end-to-end text generation | |
| CN111221961A (en) | Speaker intention recognition system and method based on S-LSTM model and slot value gate | |
| CN115292463B (en) | Information extraction-based method for joint multi-intention detection and overlapping slot filling | |
| CN111651572A (en) | Multi-domain task type dialogue system, method and terminal | |
| CN109933773A (en) | A kind of multiple semantic sentence analysis system and method | |
| Shalyminov et al. | Few-shot dialogue generation without annotated data: A transfer learning approach | |
| CN112116907A (en) | Speech recognition model establishing method, speech recognition device, speech recognition equipment and medium | |
| CN113961687A (en) | Multi-turn question-answering system intention classification and named entity identification research method | |
| CN113076421A (en) | Social noise text entity relation extraction optimization method and system | |
| CN112597301A (en) | Voice intention recognition method and device | |
| CN115687934A (en) | Intention recognition method and device, computer equipment and storage medium | |
| CN113239694B (en) | Argument role identification method based on argument phrase | |
| CN114860869A (en) | Controllable universal dialogue model with generalized intentions | |
| CN110472655A (en) | A kind of marker machine learning identifying system and method for border tourism | |
| CN113642862A (en) | Method and system for identifying named entities of power grid dispatching instructions based on BERT-MBIGRU-CRF model | |
| CN116306502A (en) | Data annotation optimization system and method for BERT classification task | |
| CN114116975A (en) | Multi-intention identification method and system | |
| CN115240712A (en) | Multi-mode-based emotion classification method, device, equipment and storage medium |
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 |