CN113420137A - Method, device and medium for implementing intelligent question-answering system based on end-to-end framework - Google Patents
Method, device and medium for implementing intelligent question-answering system based on end-to-end framework Download PDFInfo
- Publication number
- CN113420137A CN113420137A CN202110725000.8A CN202110725000A CN113420137A CN 113420137 A CN113420137 A CN 113420137A CN 202110725000 A CN202110725000 A CN 202110725000A CN 113420137 A CN113420137 A CN 113420137A
- Authority
- CN
- China
- Prior art keywords
- information
- reply
- input
- reply information
- module
- 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
- 238000000034 method Methods 0.000 title claims abstract description 38
- 230000007246 mechanism Effects 0.000 claims abstract description 24
- 238000004140 cleaning Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 4
- 238000004886 process control Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011144 upstream manufacturing Methods 0.000 description 1
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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a method, a device and a medium for realizing an intelligent question-answering system based on an end-to-end framework, wherein the scheme comprises the following steps: according to first input information input into the Encoder module, the Decode module matches corresponding first reply information for the first input information, so that the flow control module stores the first input information and the first reply information into the context storage module to form historical reply information; setting the content of one head of the multi-head at content mechanism as slot position information through a multi-head attribute ion mechanism, obtaining an intention corresponding to the slot position information based on the historical reply information and second input information, and filling the slot position; and determining that the slot filling is completed, matching corresponding second reply information for the second input information, and determining that the second reply information is final reply information. The problem of the intelligent question-answering system of end-to-end framework model poor in flexibility when processing the situation that slot filling needs to interact with the user for many times is solved.
Description
Technical Field
The embodiment of the specification relates to the field of intelligent question answering, in particular to a method, equipment and medium for realizing an intelligent question answering system based on an end-to-end framework.
Background
In the field of intelligent question answering, the multi-turn question answering system implementation architecture is mainly divided into two types: a pipelined architecture and an end-to-end architecture. The pipeline type framework has the advantages that all modules are separated from each other, the interpretability is strong, the positioning and the modification can be rapidly carried out after problems are met, and the front multi-turn question-answering system is mainly realized based on the pipeline type framework. However, because the pipeline architecture is implemented, the modules are separated from each other, and therefore errors occurring in the upstream module are easily transmitted to the downstream, which affects user experience.
The end-To-end architecture adopts a Seq-To-Seq framework, the whole architecture has no definite module division, and the whole optimization is carried out in a back propagation mode, so that the defects of the pipeline architecture can be effectively overcome. However, when the end-to-end structure involves a situation that slot filling needs to interact with a user for multiple times, the end-to-end structure cannot be designed for a specific slot value in an intermediate link, and compared with a pipeline type structure, the end-to-end structure is very sluggish in processing the problem.
Disclosure of Invention
The embodiment of the specification provides an intelligent question-answering system implementation method, equipment and a medium based on an end-to-end framework, and is used for solving the following technical problems in the prior art:
an end-to-end architecture model in a multi-turn question-answering system relates to the problems of weak flexibility and poor interactivity when slot filling needs to interact with a user for multiple times.
In order to solve the above technical problem, the embodiments of the present specification are implemented as follows:
in one aspect, an embodiment of the present specification provides an implementation method of an intelligent question-answering system based on an end-to-end framework, including: according to first input information input into the Encoder module, the Decode module matches corresponding first reply information for the first input information, so that the flow control module stores the first input information and the first reply information into the context storage module to form historical reply information; setting the content of one head of the multi-head attention mechanism as slot position information through a multi-head attention mechanism, obtaining an intention corresponding to the slot position information based on the historical reply information and second input information, and filling the slot position; and determining that the slot filling is completed, matching corresponding second reply information for the second input information, and determining that the second reply information is final reply information.
In one embodiment, the matching, by the Decoder module, corresponding first reply information for the first input information according to the first input information input into the Encoder module specifically includes: according to first input information input into an Encoder module, the Encoder module encodes the first input information to obtain first encoded information; and acquiring the first coding information, decoding the first coding information to obtain first decoding information, and acquiring first reply information matched with the first decoding information from a preset reply data set.
In one embodiment, before matching the second input information with the corresponding second reply information, the method further comprises: acquiring additional information in an additional information module, wherein the additional information at least comprises a conversation subject, a conversation scene and a user image.
In one embodiment, after determining that the second reply message is a final reply message, the method further comprises: and responding to a data cleaning request, and deleting specified question and answer related information at regular time, wherein the question and answer related information comprises the historical reply information, the second input information, the second reply information and the additional information.
In one embodiment, the matching of the second input information with the corresponding second reply information and the determination that the second reply information is the final reply information specifically include: acquiring second reply information matched with the second input information from the preset reply data set prestored in the context storage module, and determining the second reply information as final reply information; or generating the second reply information based on the second input information, and determining the second reply information as final reply information.
In one embodiment, after the acquiring additional information in the additional information module and before the matching of the second input information with the corresponding second reply information, the method further includes: and judging the task type of the intention through the multi-head attention mechanism according to second input information and the additional information input into the Decoder module.
In one embodiment, the second reply information matched with the second input information is obtained from the preset answer data set prestored in the context storage module, and the second reply information is determined to be final reply information; or generating the second reply information based on the second input information, and determining that the second reply information is the final reply information, which specifically includes: if the task type is a specific task, acquiring information matched with the specific task from the context storage module as second reply information, and determining the second reply information as final reply information; if the task type is a non-specific task, judging whether a candidate answer matched with the non-specific task exists; if so, taking the candidate answer as the second reply message, and determining the second reply message as a final reply message; if not, generating the second reply information based on the second input information, and determining the second reply information as final reply information.
In one embodiment, the determining whether there is a candidate answer matching the non-specific task specifically includes: and acquiring a candidate answer set pre-stored in the context module, and judging whether the candidate answer set has the candidate answer with the matching degree with the non-specific task higher than a preset threshold value.
On the other hand, the embodiment of the present specification provides an intelligent question-answering system implementation device based on an end-to-end framework, including: at least one processor, and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to: according to first input information input into the Encoder module, the Decode module matches corresponding first reply information for the first input information, so that the flow control module stores the first input information and the first reply information into the context storage module to form historical reply information; setting the content of one head of the multi-head attention mechanism as slot position information through a multi-head attention mechanism, obtaining an intention corresponding to the slot position information based on the historical reply information and second input information, and filling the slot position; and determining that the slot filling is completed, matching corresponding second reply information for the second input information, and determining that the second reply information is final reply information.
The embodiment of the specification can achieve the following beneficial effects by adopting at least one technical scheme: the problem of the intelligent question-answering system of end-to-end framework model poor in flexibility when processing the situation that slot filling needs to interact with the user for many times is solved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the disclosure and are incorporated in and constitute a part of this disclosure, illustrate embodiments of the disclosure and together with the description serve to explain the disclosure and not to limit the disclosure in any way. In the drawings:
fig. 1 is a schematic flowchart of an implementation method of an intelligent question-answering system based on an end-to-end framework according to an embodiment of the present application;
fig. 2 is a schematic structural diagram of an implementation method of an intelligent question-answering system based on an end-to-end framework according to an embodiment of the present application;
fig. 3 is a schematic structural diagram of an implementation device of an intelligent question answering system based on an end-to-end framework according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present disclosure more clear, the following description of the present disclosure will be made in detail and completely with reference to the embodiments of the present disclosure and the accompanying drawings. It is to be understood that the embodiments described are only a few embodiments of the present application and not all embodiments. All other embodiments obtained by a person skilled in the art without making any inventive step based on the embodiments in the description belong to the protection scope of the present application.
The embodiment of the application provides an intelligent question-answering system implementation method, equipment and a medium based on an end-to-end framework, and the following specific description is given:
fig. 1 is a schematic flow chart of an implementation method of an intelligent question-answering system based on an end-to-end framework according to an embodiment of the present application, and as shown in fig. 1, the implementation method of the intelligent question-answering system based on the end-to-end framework according to the embodiment of the present application may include the following steps:
s101: according to first input information input into the Encoder module, the Decode module matches corresponding first reply information for the first input information, so that the flow control module stores the first input information and the first reply information into the context storage module to form historical reply information.
In the intelligent question-answering system, firstly, a user initiates a question, the question initiated by the user is input into an Encoder module as first input information, the Decoder module can match a corresponding reply for the question proposed by the user, and the reply can be an answer to the question proposed by the user or one or more matched questions of the user can be asked against the question proposed by the user. For example, user a asks robot number 1: "how do the weather today? "robot No. 1 asks user a in reverse: "asking for weather in which city? ".
In some service scenes, multiple rounds of questions and answers may be needed between a user and a robot to enable the robot to provide the user with an accurate reply, a flow control module controls input information and reply information related in the multiple rounds of questions and answers, the input information and the reply information related in the multiple rounds of questions and answers are stored in a context storage module, and the reply information before the robot provides the user with an accurate reply can be called as historical reply information.
It should be noted that the context storage module mainly stores useful information in the conversation process. The method comprises the information of the user's question, the robot's answer, the user-defined scene, the user's figure portrait and the topic of the conversation in each round.
Further, according to the first input information input into the Encoder module, the Decoder module matches corresponding first reply information for the first input information, specifically including: according to first input information input into an Encoder module, the Encoder module encodes the first input information to obtain first encoded information; and acquiring the first coding information, decoding the first coding information to obtain first decoding information, and acquiring first reply information matched with the first decoding information from a preset reply data set.
S102: and setting the content of one head of the multi-head attention mechanism as slot position information through a multi-head attention mechanism, obtaining an intention corresponding to the slot position information based on the historical reply information and second input information, and filling the slot position.
The Encoder module is responsible for coding the input information of the user. The specific implementation mode is to use the Encoder layer of the Transformer model, and eight-head self-attention can be used in multi-head attention. In a task-based dialog system, when a slot filling problem is encountered, multiple rounds of question and answer with a user are required for slot filling. To address this problem in the end-to-end model, the method sets one of the headers of the multi-header mechanism to slot bit information. In the multi-turn question answering, besides the information input by the user, the contextual information, namely the historical reply information, is also considered. The context storage module provides historical reply information to the Encoder module, the historical reply information including questions and answers referring to the first n rounds between the user and the robot. This input also serves as one of the headers of the Encoder module. And all the information is output to the Decoder module after being encoded.
The multi-head attention mechanism is an attention mechanism in deep learning, and after information attention is input, a relation vector between each word and other words can be obtained, so that the machine can better understand the input vector.
Slot filling refers to the process of completing information for the purpose of translating user intent into user-specific instructions. In the intelligent question answering, the robot may recognize the first input information as incomplete information, that is, the first input information is insufficient and clear, and the robot cannot give an accurate reply under the condition of only the first input information, so that the robot outputs the first reply information, obtains more useful information based on the second input information of the user, obtains an intention corresponding to the slot position information, and fills the slot position.
For example, user B: ' Hel me book an airline ticket, starting from Hangzhou. "two slots should be filled here, and" air ticket "and" state "are filled in as" vehicle "and" destination ". Then in the intelligent question-answering system, robot No. 2 understands that the user wants to buy a ticket to hangzhou, but because of the lack of a departure location, a round of inquiry dialogue is also performed. And finally, the robot No. 2 generates an instruction according to the information supplemented by the user B, and calls an interface to purchase an air ticket from a place to Hangzhou.
S103: and determining that the slot filling is completed, matching corresponding second reply information for the second input information, and determining that the second reply information is final reply information.
Further, matching the second input information with corresponding second reply information, and determining that the second reply information is the final reply information, specifically including: acquiring second reply information matched with the second input information from the preset reply data set prestored in the context storage module, and determining the second reply information as final reply information; or generating the second reply information based on the second input information, and determining the second reply information as final reply information.
In some embodiments of the present description, before matching the second input information with the corresponding second reply information, the method further includes: and judging the task type of the intention through the multi-head attention mechanism according to second input information and the additional information input into the Decoder module.
Further, second reply information matched with the second input information is acquired from the preset answer data set prestored in the context storage module, and the second reply information is determined to be final reply information; or generating the second reply information based on the second input information, and determining that the second reply information is the final reply information, which specifically includes: if the task type is a specific task, acquiring information matched with the specific task from the context storage module as second reply information, and determining the second reply information as final reply information; if the task type is a non-specific task, judging whether a candidate answer matched with the non-specific task exists; if so, taking the candidate answer as the second reply message, and determining the second reply message as a final reply message; if not, generating the second reply information based on the second input information, and determining the second reply information as final reply information.
The invention improves the end-to-end intelligent dialogue framework, which is suitable for both task type dialogue system and non-task type dialogue system. The task type in the task type dialog system is a specific task, the task type in the non-task type dialog system is a non-specific task, and the non-task type dialog system can also be understood as a chatting type dialog system.
In some embodiments of the present specification, the determining whether there is a candidate answer matching the non-specific task specifically includes: and acquiring a candidate answer set pre-stored in the context module, and judging whether the candidate answer set has the candidate answer with the matching degree with the non-specific task higher than a preset threshold value.
In some embodiments of the present description, before matching the second input information with the corresponding second reply information, the method further comprises: acquiring additional information in an additional information module, wherein the additional information at least comprises a conversation subject, a conversation scene and a user image.
It can be seen that there are three sources for the input information to the Decoder module: the first is the information output by the Encoder module, i.e. the above-mentioned first coded information; the second is additional information such as conversation subjects, conversation scenes and the like output by the additional information module; and thirdly, optional answer information provided by the context storage module.
In some embodiments of the present description, the method further provides: after determining that the second reply message is a final reply message, the method further comprises: and responding to a data cleaning request, and deleting specified question and answer related information at regular time, wherein the question and answer related information comprises the historical reply information, the second input information, the second reply information and the additional information.
In some embodiments of the present specification, fig. 2 is a schematic structural diagram of an implementation method of an intelligent question-answering system based on an end-to-end framework according to an embodiment of the present application, and as shown in fig. 2, a process control module may implement control over an entire process.
And the flow control module controls all the other modules. For the Encoder module, the flow control module determines which information it needs to store to the context storage module, and controls the Encoder module to take into account the previous rounds of information in a multi-round dialog. For the Decoder module, the process control module controls whether the reply information output by the Decoder module should be stored in the context storage module. For the additional information module, the flow control module determines which additional information in the additional information module should be input to the Decoder module. For the context storage module, the process control module can control the context storage module to delete the historical reply information at regular time, and whether the information in the context storage module is added into the Encoder module, how many rounds of historical reply information are considered, and the like. It can be seen that the process control module is the user's access to the system controls.
In conclusion, the method for implementing the intelligent question-answering system based on the end-to-end framework can solve the problem that the intelligent question-answering system based on the end-to-end framework model is poor in flexibility when the situation that slot filling needs to be conducted and multiple interactions are conducted on a user is processed.
Fig. 3 is a schematic structural diagram of an end-to-end framework-based intelligent question answering system implementation device provided in the embodiment of the present application, and as shown in fig. 3, the end-to-end framework-based intelligent question answering system implementation device provided in the embodiment of the present application may include: at least one processor, and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to: according to first input information input into the Encoder module, the Decode module matches corresponding first reply information for the first input information, so that the flow control module stores the first input information and the first reply information into the context storage module to form historical reply information; setting the content of one head of the multi-head attention mechanism as slot position information through a multi-head attention mechanism, obtaining an intention corresponding to the slot position information based on the historical reply information and second input information, and filling the slot position; and determining that the slot filling is completed, matching corresponding second reply information for the second input information, and determining that the second reply information is final reply information.
The processor and the memory may communicate via a bus, and the device may further include an input/output interface for communicating with other devices.
Some embodiments of the present application provide a non-transitory computer storage medium corresponding to the end-to-end framework based intelligent question-answering system of fig. 1, having stored thereon computer-executable instructions configured to: according to first input information input into the Encoder module, the Decode module matches corresponding first reply information for the first input information, so that the flow control module stores the first input information and the first reply information into the context storage module to form historical reply information; setting the content of one head of the multi-head attention mechanism as slot position information through a multi-head attention mechanism, obtaining an intention corresponding to the slot position information based on the historical reply information and second input information, and filling the slot position; and determining that the slot filling is completed, matching corresponding second reply information for the second input information, and determining that the second reply information is final reply information.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the device, non-volatile computer storage medium embodiment, since it is substantially similar to the method embodiment, the description is relatively simple, and in relation to the description, reference may be made to some portions of the description of the method embodiment.
The above description is only an example of the present application and is not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modifications, equivalents, improvements, etc. that come within the spirit of the disclosure are intended to be included within the scope of the claims of this disclosure.
Claims (10)
1. The method for realizing the intelligent question-answering system based on the end-to-end framework is characterized by comprising the following steps:
according to first input information input into the Encoder module, the Decode module matches corresponding first reply information for the first input information, so that the flow control module stores the first input information and the first reply information into the context storage module to form historical reply information;
setting the content of one head of the multi-head attention mechanism as slot position information through a multi-head attention mechanism, obtaining an intention corresponding to the slot position information based on the historical reply information and second input information, and filling the slot position;
and determining that the slot filling is completed, matching corresponding second reply information for the second input information, and determining that the second reply information is final reply information.
2. The method according to claim 1, wherein the matching, by the Decoder module, corresponding first reply information for the first input information according to the first input information input to the Encoder module specifically comprises:
according to first input information input into an Encoder module, the Encoder module encodes the first input information to obtain first encoded information;
and acquiring the first coding information, decoding the first coding information to obtain first decoding information, and acquiring first reply information matched with the first decoding information from a preset reply data set.
3. The method of claim 1, wherein before matching the second input information with the corresponding second reply information, the method further comprises:
acquiring additional information in an additional information module, wherein the additional information at least comprises a conversation subject, a conversation scene and a user image.
4. The method of claim 3, wherein after determining that the second reply message is a final reply message, the method further comprises:
and responding to a data cleaning request, and deleting specified question and answer related information at regular time, wherein the question and answer related information comprises the historical reply information, the second input information, the second reply information and the additional information.
5. The method according to claim 4, wherein the matching of the second input message with the corresponding second reply message and the determination of the second reply message as the final reply message specifically comprises:
acquiring second reply information matched with the second input information from the preset reply data set prestored in the context storage module, and determining the second reply information as final reply information; or generating the second reply information based on the second input information, and determining the second reply information as final reply information.
6. The method of claim 5, wherein after the obtaining of the additional information in the additional information module and before the matching of the corresponding second reply information for the second input information, the method further comprises:
and judging the task type of the intention through the multi-head attention mechanism according to second input information and the additional information input into the Decoder module.
7. The method according to claim 6, wherein the second reply message matching the second input message is obtained from the preset answer data set pre-stored in the context storage module, and the second reply message is determined as a final reply message; or generating the second reply information based on the second input information, and determining that the second reply information is the final reply information, which specifically includes:
if the task type is a specific task, acquiring information matched with the specific task from the context storage module as second reply information, and determining the second reply information as final reply information;
if the task type is a non-specific task, judging whether a candidate answer matched with the non-specific task exists; if so, taking the candidate answer as the second reply message, and determining the second reply message as a final reply message; if not, generating the second reply information based on the second input information, and determining the second reply information as final reply information.
8. The method of claim 7, wherein the determining whether there is a candidate answer matching the non-specific task specifically comprises:
and acquiring a candidate answer set pre-stored in the context module, and judging whether the candidate answer set has the candidate answer with the matching degree with the non-specific task higher than a preset threshold value.
9. Intelligent question-answering system realization equipment based on an end-to-end framework is characterized by comprising:
at least one processor, and,
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to enable the at least one processor to:
according to first input information input into the Encoder module, the Decode module matches corresponding first reply information for the first input information, so that the flow control module stores the first input information and the first reply information into the context storage module to form historical reply information;
setting the content of one head of the multi-head attention mechanism as slot position information through a multi-head attention mechanism, obtaining an intention corresponding to the slot position information based on the historical reply information and second input information, and filling the slot position;
and determining that the slot filling is completed, matching corresponding second reply information for the second input information, and determining that the second reply information is final reply information.
10. A non-transitory computer storage medium storing computer-executable instructions configured to:
according to first input information input into the Encoder module, the Decode module matches corresponding first reply information for the first input information, so that the flow control module stores the first input information and the first reply information into the context storage module to form historical reply information;
setting the content of one head of the multi-head attention mechanism as slot position information through a multi-head attention mechanism, obtaining an intention corresponding to the slot position information based on the historical reply information and second input information, and filling the slot position;
and determining that the slot filling is completed, matching corresponding second reply information for the second input information, and determining that the second reply information is final reply information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110725000.8A CN113420137A (en) | 2021-06-29 | 2021-06-29 | Method, device and medium for implementing intelligent question-answering system based on end-to-end framework |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110725000.8A CN113420137A (en) | 2021-06-29 | 2021-06-29 | Method, device and medium for implementing intelligent question-answering system based on end-to-end framework |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113420137A true CN113420137A (en) | 2021-09-21 |
Family
ID=77717788
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110725000.8A Pending CN113420137A (en) | 2021-06-29 | 2021-06-29 | Method, device and medium for implementing intelligent question-answering system based on end-to-end framework |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113420137A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110069607A (en) * | 2017-12-14 | 2019-07-30 | 株式会社日立制作所 | For the method, apparatus of customer service, electronic equipment, computer readable storage medium |
| CN111046132A (en) * | 2019-10-25 | 2020-04-21 | 众安信息技术服务有限公司 | Customer service question and answer processing method and system for retrieving multiple rounds of conversations |
| CN111858854A (en) * | 2020-07-20 | 2020-10-30 | 上海汽车集团股份有限公司 | Question-answer matching method based on historical dialogue information and related device |
| CN112100354A (en) * | 2020-09-16 | 2020-12-18 | 北京奇艺世纪科技有限公司 | Man-machine conversation method, device, equipment and storage medium |
| WO2021051404A1 (en) * | 2019-09-20 | 2021-03-25 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for auxiliary reply |
-
2021
- 2021-06-29 CN CN202110725000.8A patent/CN113420137A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110069607A (en) * | 2017-12-14 | 2019-07-30 | 株式会社日立制作所 | For the method, apparatus of customer service, electronic equipment, computer readable storage medium |
| WO2021051404A1 (en) * | 2019-09-20 | 2021-03-25 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for auxiliary reply |
| CN111046132A (en) * | 2019-10-25 | 2020-04-21 | 众安信息技术服务有限公司 | Customer service question and answer processing method and system for retrieving multiple rounds of conversations |
| CN111858854A (en) * | 2020-07-20 | 2020-10-30 | 上海汽车集团股份有限公司 | Question-answer matching method based on historical dialogue information and related device |
| CN112100354A (en) * | 2020-09-16 | 2020-12-18 | 北京奇艺世纪科技有限公司 | Man-machine conversation method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111309889B (en) | Method and device for text processing | |
| US11449678B2 (en) | Deep learning based dialog method, apparatus, and device | |
| KR102178738B1 (en) | Automated assistant calls from appropriate agents | |
| CN111710336A (en) | Speech intention recognition method and device, computer equipment and storage medium | |
| CN107977236B (en) | Question-answering system generation method, terminal device, storage medium and question-answering system | |
| CN108922564B (en) | Emotion recognition method and device, computer equipment and storage medium | |
| JP6677419B2 (en) | Voice interaction method and apparatus | |
| CN108959388B (en) | Information generation method and device | |
| CN112632961A (en) | Natural language understanding processing method, device and equipment based on context reasoning | |
| CN116884391B (en) | Multimode fusion audio generation method and device based on diffusion model | |
| CN112084317A (en) | Method and apparatus for pre-training a language model | |
| CN111027291A (en) | Method and device for adding punctuation marks in text and training model and electronic equipment | |
| CN111462726B (en) | Method, device, equipment and medium for answering out call | |
| CN105632495A (en) | Voice recognition method and apparatus | |
| CN115129878A (en) | Conversation service execution method, device, storage medium and electronic equipment | |
| US12026544B2 (en) | Self-play to improve task-oriented dialog systems and methods | |
| KR102464121B1 (en) | Apparatus for providing query answering based on relation between query and response and method there of | |
| CN111506717B (en) | Question answering method, device, equipment and storage medium | |
| CN113420137A (en) | Method, device and medium for implementing intelligent question-answering system based on end-to-end framework | |
| US20230177263A1 (en) | Identifying chat correction pairs for trainig model to automatically correct chat inputs | |
| CN115346520A (en) | Method, apparatus, electronic device and medium for speech recognition | |
| CN112965593A (en) | AI algorithm-based method and device for realizing multi-mode control digital human interaction | |
| CN110399615B (en) | Transaction risk monitoring method and device | |
| CN113704424A (en) | Natural language task generation method based on pointer network | |
| CN111222322A (en) | Information processing method and electronic 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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210921 |