CN113918701B - Billboard display method and device - Google Patents
Billboard display method and device Download PDFInfo
- Publication number
- CN113918701B CN113918701B CN202111220334.6A CN202111220334A CN113918701B CN 113918701 B CN113918701 B CN 113918701B CN 202111220334 A CN202111220334 A CN 202111220334A CN 113918701 B CN113918701 B CN 113918701B
- Authority
- CN
- China
- Prior art keywords
- analysis result
- result
- slot
- intention
- slot position
- 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
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
- G06F16/33—Querying
- G06F16/338—Presentation of query results
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Machine Translation (AREA)
Abstract
The application discloses a billboard display method and a billboard display device, which are used for solving the problem that the billboard cannot support complex display requirements. The billboard display method comprises the following steps: inputting a billboard display request expressed by a natural language; processing the billboard display request through a natural language processing model to obtain a first intention analysis result and a first slot position analysis result; eliminating the intention ambiguity of the first intention analysis result, and obtaining a second intention analysis result and a second slot position analysis result according to the disambiguation result and the first slot position analysis result; eliminating slot position text ambiguity of the second slot position analysis result according to a preset slot position text processing rule to obtain a third slot position analysis result; obtaining a billboard display parameter according to the second intention analysis result, the third slot analysis result and the global billboard parameter; and updating the display content of the billboard according to the billboard display parameters. Through multiple disambiguation processes, complex display requirements are effectively identified.
Description
Technical Field
The application relates to the technical field of business intelligence, in particular to a billboard display method and device.
Background
With the continuous development and popularization of business intelligent platforms, enterprise users put higher demands on the usability and intellectualization of the platforms. In a traditional business intelligent platform, a user carries out intelligent analysis and display of business data through complex interface operation, operation training needs to be carried out on the user, and new analysis requirements cannot be quickly responded. For this problem, the industry proposes a solution idea of using natural language interaction to perform intelligent data analysis, and represents technologies such as NL2SQL and visual QA.
In the process of realizing the prior art, the inventor finds that:
NL2SQL can automatically parse a single question sentence in natural language into executable SQL statements, which have an accuracy of about 90% on a simple sentence, but have an accuracy of less than 50% on a complex sentence. The visualized QA technology analyzes question input in real time, can recommend similar question in real time, or prompts a user to actively select a semantic entity, but still needs to perform complex interface interaction, and is difficult to process complex sentences. The existing natural language interaction technology improves the usability and the intellectualization of a business intelligent platform to a certain extent, but the complex sentence resolving capability and the interaction capability still need to be improved.
Therefore, it is desirable to provide a billboard display method and apparatus for solving the technical problem that the billboard cannot support complex display requirements.
Disclosure of Invention
The embodiment of the application provides a billboard display method and a billboard display device, which are used for solving the technical problem that the billboard cannot support complex display requirements.
Specifically, the billboard display method comprises the following specific steps:
inputting a billboard display request expressed by a natural language;
processing the billboard display request through a natural language processing model to obtain a first intention analysis result and a first slot position analysis result;
eliminating the intention ambiguity of the first intention analysis result, and obtaining a second intention analysis result and a second slot position analysis result according to the disambiguation result and the first slot position analysis result;
eliminating slot position text ambiguity of the second slot position analysis result according to a preset slot position text processing rule to obtain a third slot position analysis result;
obtaining a billboard display parameter according to the second intention analysis result, the third slot analysis result and the global billboard parameter;
and updating the display content of the billboard according to the billboard display parameters.
Further, the sample data set of the natural language processing model is generated by a sample generation algorithm, and the method comprises the following specific steps:
collecting samples according to preset intention category data, preset slot types and BI data, and adding the collected samples into an initial data set;
and according to the preset sampling number, circularly executing the sample acquisition operation to obtain a sample data set.
Further, the method for resolving the intent ambiguity of the first intent parsing result and obtaining a second intent parsing result and a second slot parsing result according to the disambiguation result and the first slot parsing result comprises the following specific steps:
analyzing the first intention analysis result according to a preset ambiguity judging method according to the probability distribution data of the first intention analysis result to obtain an ambiguity judging result;
when the ambiguity judgment result indicates that the first intention analysis result is ambiguous, processing the first intention analysis result through a preset intention ambiguity elimination method to obtain an intention analysis result to be verified;
according to a predefined rule, verifying the analysis result of the intention to be verified and the slot position type corresponding to the first slot position analysis result to obtain a verification result;
and when the verification result meets a preset condition, obtaining a second intention analysis result and a second slot position analysis result according to the verification result.
Further, the method for resolving the intent ambiguity of the first intent parsing result and obtaining a second intent parsing result and a second slot parsing result according to the disambiguation result and the first slot parsing result comprises the following specific steps:
analyzing the first intention analysis result according to a preset ambiguity judging method according to the probability distribution data of the first intention analysis result to obtain an ambiguity judging result;
when the ambiguity judgment result indicates that the first intention analysis result is ambiguous, verifying the first intention analysis result and the slot position type corresponding to the first slot position analysis result according to a predefined rule to obtain a verification result;
and when the verification result meets a preset condition, obtaining a second intention analysis result and a second slot position analysis result according to the verification result.
Further, according to a preset slot text processing rule, slot text ambiguity of the second slot parsing result is eliminated, and a third slot parsing result is obtained, wherein the method comprises the following specific steps:
determining the slot position type of the second slot position analysis result;
when the slot type is a non-BI core data type, selecting a local verification method adaptive to the type characteristics to verify the slot text according to the data type characteristics to obtain a verification result;
and updating the slot position text of the second slot position analysis result according to the verification result to obtain a third slot position analysis result.
Further, according to a preset slot text processing rule, slot text ambiguity of the second slot parsing result is eliminated, and a third slot parsing result is obtained, wherein the method comprises the following specific steps:
determining the slot position type of the second slot position analysis result;
when the slot type is a BI core data type, calling a business intelligent data interface to obtain interface data;
matching the slot position text of the second slot position analysis result with the interface data to obtain a matching result;
and updating the slot position text of the second slot position analysis result according to the matching result to obtain a third slot position analysis result.
Further, obtaining a billboard display parameter according to the second intention analysis result, the third slot analysis result and the global billboard parameter, comprising the following specific steps:
performing semantic verification according to the second intention analysis result and the third slot analysis result and in combination with a preset global billboard parameter to obtain a verification result;
and when the verification result is that the verification is passed, updating the second intention analysis result and the third slot analysis result to the global billboard parameter to obtain a billboard display parameter.
The present application further provides a billboard display device, comprising:
the input module is used for inputting a billboard display request expressed by a natural language;
the natural language processing module is used for processing the billboard display request through a natural language processing model to obtain a first intention analysis result and a first slot position analysis result;
the first processing module is used for eliminating the intention ambiguity of the first intention analysis result and obtaining a second intention analysis result and a second slot position analysis result according to the disambiguation result and the first slot position analysis result;
the second processing module is used for eliminating slot position text ambiguity of the second slot position analysis result according to a preset slot position text processing rule to obtain a third slot position analysis result;
the display preprocessing module is used for obtaining a billboard display parameter according to the second intention analysis result, the third slot analysis result and the global billboard parameter;
and the updating module is used for updating the display content of the billboard according to the billboard display parameter.
Further, the sample data set of the natural language processing model is generated by a sample generation algorithm, and the method comprises the following specific steps:
collecting samples according to preset intention category data, preset slot types and BI data, and adding the collected samples into an initial data set;
and according to the preset sampling number, circularly executing the sample acquisition operation to obtain a sample data set.
Further, the first processing module is configured to eliminate an intention ambiguity of the first intention parsing result, and obtain a second intention parsing result and a second slot parsing result according to the disambiguation result and the first slot parsing result, and specifically configured to:
analyzing the first intention analysis result according to a preset ambiguity judging method according to the probability distribution data of the first intention analysis result to obtain an ambiguity judging result;
when the ambiguity judgment result indicates that the first intention analysis result is ambiguous, processing the first intention analysis result through a preset intention ambiguity elimination method to obtain an intention analysis result to be verified;
according to a predefined rule, verifying the analysis result of the intention to be verified and the slot position type corresponding to the first slot position analysis result to obtain a verification result;
and when the verification result meets a preset condition, obtaining a second intention analysis result and a second slot position analysis result according to the verification result.
The technical scheme provided by the embodiment of the application at least has the following beneficial effects:
through multiple disambiguation processes, complex display requirements are effectively identified.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the application and together with the description serve to explain the application and not to limit the application. In the drawings:
fig. 1 is a flowchart of a billboard display method according to an embodiment of the present disclosure;
fig. 2 is a schematic structural diagram of a billboard display device according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the technical solutions of the present application will be described in detail and completely with reference to the following specific embodiments of the present application and the accompanying drawings. It should be apparent that the described embodiments are only some of the embodiments of the present application, 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 application.
Referring to fig. 1, a billboard display method includes the following steps:
s100: a kanban display request expressed by a natural language is input.
It should be noted that, in the practical application of the present application, the natural language may be in various forms. Preferably, in order to fit the actual application scenario and facilitate the operation, the billboard presentation request may be request information expressed in a voice format or request information directly input in a text format. The voice or text information can be switched to different languages according to actual conditions. It should be noted that the billboard display request here may be the content or the form of the display that the user wishes to have the billboard displayed. A kanban is here understood to be a component or control or the like in software for controlling the graphical or the like display of relevant data. Obviously, the display data obtained after the treatment through the billboard can be finally displayed to the user through the display terminal. Obviously, the billboard display request is input in a voice or text mode, so that the billboard display request is more suitable for application scenes and convenient to use.
S200: and processing the billboard display request through a natural language processing model to obtain a first intention analysis result and a first slot position analysis result.
It is understood that the natural language processing model can be understood as a natural language understanding model, which is a processing model in the field of artificial intelligence. The kanban display request expressed in the natural language is processed by the natural language processing model, and then corresponding data which can be processed by a computer can be generated. The processing herein may be considered as part of a human-machine interaction process. The first intention parsing result may be understood as an intention in the natural language understanding field, and the first slot position parsing result may be understood as a slot position corresponding to the intention. It is apparent that the slot herein also includes a slot type and a slot text. The generation process of the first intention parsing result may be understood as a process of intention recognition. The generation process of the first slot parsing result may be understood as a slot filling process. Intent recognition may be understood as determining what the user is going to do. The process of intent recognition is a process of text classification. Slot filling may be understood as determining what to do after determining what to do by the user. Obviously, the billboard display request is analyzed in a natural language processing model mode, so that the working efficiency can be effectively improved.
Further, the sample data set of the natural language processing model is generated by a sample generation algorithm, and the method comprises the following specific steps:
collecting samples according to preset intention category data, preset slot types and BI data, and adding the collected samples into an initial data set;
and according to the preset sampling number, circularly executing the sample acquisition operation to obtain a sample data set.
It can be understood that in the initial implementation stage of the technical solution of the present application, due to the lack of labeled samples for natural language processing model training, the rapid deployment of the system will be affected. In order to solve the cold start problem, the present application provides a sample generation algorithm, which can generate a labeled sample for training a natural language processing model by using preset intention category data, a preset slot type and BI data, where the labeled sample can be understood as a sample data set. The BI data mainly refers to indexes, dimensions and dimension data of customer original data after the customer original data are subjected to BI data governance specifications. For example, in the sentence "display total amount", the format may be converted to "display [ total amount ] (index)", and the "total amount" herein corresponds to index data. In a sentence "show [2020 ] (time) [ wuhan ] (dimension) [ profit ] (index) of [ jianghan ] (dimension) [ shorea ] (dimension)," wuhan "corresponds to dimension data," jiangshorea "corresponds to dimension data, and" profit "corresponds to index data. The BI data corresponds to an entity type of the slot, and the entity type of the slot herein may be understood as a preset slot type in the present application. And when the sample is collected, randomly selecting data from the BI data by adopting a non-replacement random sampling method to fill the data to the slot text position of the slot in the sample to form the slot text, and circulating from the beginning again after the slot text is exhausted. According to the actual display requirement, the preset intention category data can comprise a display billboard, a sequencing billboard, a trend billboard, a comparison billboard, a drill-down operation, a drill-up operation, a switching operation, a newly-added data slot position, a newly-added calculation operation, a newly-added comparison operation and the like. The preset slot type can comprise a basic type, a data-related type, a billboard-related type and other auxiliary types. Obviously, the preset intention category data and the preset slot type are only used for examples, and the actually set names and number can be adjusted accordingly according to the requirements. It should be noted that the basic types herein mainly include common entity types such as time, place, and number; the data correlation types mainly comprise three types of dimension modeling, namely indexes, dimensions and dimension items; the related types of the kanban comprise chart types, sorting modes, comparison operations, calculation operations and the like; other auxiliary types herein include types for disambiguating various slots. Obviously, the samples collected by the sample generation algorithm accord with the actual use environment, and the natural language processing model obtained by training can effectively improve the recognition rate.
S300: and eliminating the intention ambiguity of the first intention analysis result, and obtaining a second intention analysis result and a second slot position analysis result according to the disambiguation result and the first slot position analysis result.
It can be understood that, the first intention parsing result is an intention recognition result automatically parsed by the natural language processing model in a pipeline processing manner, and the possibility of ambiguity existing in the intention parsing of some complex sentences becomes high. In order to improve the recognition rate of the complex sentence, the first intention parsing result may be subjected to intention disambiguation processing. The intention disambiguation process here may be understood as processing the first intention parsing result into intention information with clear intention. The intention information with clear intention can be understood as the second intention resolving result here. After the intent ambiguity is resolved, slot position information corresponding to the second intent parsing result can be obtained by further processing according to the second intent parsing result and combining the first slot position parsing result, and the slot position information can be understood as the second slot position parsing result. Obviously, by eliminating the intention ambiguity, the recognition rate of the complex sentence can be effectively improved.
Specifically, the method for eliminating the intent ambiguity of the first intent parsing result and obtaining a second intent parsing result and a second slot parsing result according to the disambiguation result and the first slot parsing result includes the following steps:
analyzing the first intention analysis result according to a preset ambiguity judging method according to the probability distribution data of the first intention analysis result to obtain an ambiguity judging result;
when the ambiguity judgment result indicates that the first intention analysis result is ambiguous, processing the first intention analysis result through a preset intention ambiguity elimination method to obtain an intention analysis result to be verified;
according to a predefined rule, verifying the analysis result of the intention to be verified and the slot position type corresponding to the first slot position analysis result to obtain a verification result;
and when the verification result meets a preset condition, obtaining a second intention analysis result and a second slot position analysis result according to the verification result.
Note that, the probability distribution data here may be understood as a probability that the intention represented by the first intention analysis result is each intention in the preset intention category data. The preset ambiguity decision method can be set according to actual needs. In a specific implementation process, the following ambiguity determination methods may be preset: when the probability value corresponding to the intention with the maximum probability in the probability distribution data is smaller than a threshold value, judging that ambiguity exists; when the difference between the probability values corresponding to the two intentions with the highest probability in the probability distribution data is smaller than a threshold value, judging that ambiguity exists; when the first intent parsing result does not have corresponding intent data, but the first slot parsing result contains entity text, such first intent parsing result is determined to be ambiguous. The preset intent disambiguation method can be set according to actual needs. In a specific implementation process, the preset intent disambiguation method may be intent disambiguation of the first intent parsing result, which is completed in a human-computer interaction manner. The preset intention disambiguation method can also be that the intention disambiguation is completed through an intention sorting algorithm according to a machine learning model. For example, when the first intention parsing result does not have corresponding intention data, but a slot type exists in the first slot parsing result, and entity text data exists in a corresponding slot text, it is determined that the first intention parsing result has ambiguity; and prompting a user to select related preset intention category data according to the slot position type and the slot position text in the first slot position analysis result, thereby realizing intention ambiguity elimination. The predefined rule can be understood as the format, parameter number requirement, and the like of the preset slot bit data corresponding to the preset intention type data. For example, in the preset intention category data, the intention categories of "bulletin board" and "add calculation operator" are assumed to be preset. Under the intention of 'display kanban', the preset condition is that the number of indexes or dimensions cannot exceed 3; with the intention of "adding a calculation operator", the preset condition is that there must be at least one calculation operator (only type is verified, no value is verified). Obviously, when verification is performed, if the corresponding data does not meet the preset conditions, the corresponding verification fails, and finally the method for removing the intention ambiguity exits; and if the corresponding data meet the preset conditions, the verification is passed, and the next step of processing is carried out. Specifically, a set of formal verification rule sets with clear boundaries and suitable functions needs to be formulated by combining the functions of the existing billboard, so that an effective predefined rule is formed. It should be noted that when the verification result meets the preset condition after being judged, it may be determined that the verification result includes semantic clear intention information with ambiguity eliminated, that is, the second intention parsing result. According to the intention information with clear semantics, the operations of slot inheritance, slot replacement and the like can be further performed on the first slot analysis result. Thus, the slot data after the preliminary processing, i.e., the second slot parsing result, is obtained. Obviously, when the intention ambiguity elimination is carried out by combining the ambiguity judgment method, the intention ambiguity elimination method and the predefined rule, the processing efficiency can be effectively improved.
Further, the method for resolving the intent ambiguity of the first intent parsing result and obtaining a second intent parsing result and a second slot parsing result according to the disambiguation result and the first slot parsing result comprises the following specific steps:
analyzing the first intention analysis result according to a preset ambiguity judging method according to the probability distribution data of the first intention analysis result to obtain an ambiguity judging result;
when the ambiguity judgment result indicates that the first intention analysis result is ambiguous, verifying the first intention analysis result and the slot position type corresponding to the first slot position analysis result according to a predefined rule to obtain a verification result;
and when the verification result meets a preset condition, obtaining a second intention analysis result and a second slot position analysis result according to the verification result.
It can be appreciated that not all of the content is ambiguous when the first intent parsing result is ambiguous in intent. When it is determined that the intention represented by the first intention parsing result is not ambiguous, it may be determined that intention information in the first intention parsing result is semantically clear. The first intention analysis result after the judgment can be understood as the second intention analysis result here. After the intent information with clear semantics and the slot position type corresponding to the first slot position analysis result are verified, operations such as slot position inheritance and replacement can be further performed on the first slot position analysis result. Thus, the slot data after the preliminary processing, i.e., the second slot parsing result, is obtained. Obviously, the data accuracy can be effectively improved by verifying the first intention analysis result with clear semantics and the slot type corresponding to the first slot analysis result.
S400: and eliminating slot position text ambiguity of the second slot position analysis result according to a preset slot position text processing rule to obtain a third slot position analysis result.
Note that the second slot parsing result may be understood as newly identified slot information. According to different slot types in the slot information, the verification and ambiguity elimination modes of the slot text are different.
Specifically, the method for eliminating slot position text ambiguity of the second slot position analysis result according to a preset slot position text processing rule to obtain a third slot position analysis result includes the following specific steps:
determining the slot position type of the second slot position analysis result;
when the slot type is a non-BI core data type, selecting a local verification method adaptive to the type characteristics to verify the slot text according to the data type characteristics to obtain a verification result;
and updating the slot position text of the second slot position analysis result according to the verification result to obtain a third slot position analysis result.
It can be understood that the slot type of the second slot parsing result is directly related to the checksum disambiguation mode of the slot text in the second slot parsing result. Before disambiguation of the slot text is performed, the slot type of the second slot parsing result needs to be confirmed. The non-BI core data types include types of format specifications such as time and number, and small-scale enumerable types such as location, calculation operation, and chart types. The local verification method comprises a format specification verification method of data and a set matching verification method. When the slot type is a format specification type, the slot text of the second display slot parsing result can be directly checked according to the format specification of the corresponding data. When the slot type is a small-scale enumeratable type, verification can be performed by matching the data set corresponding to the slot type with the slot text of the second display slot parsing result. For example, a slot type of a time type, the corresponding slot text needs to satisfy a prescribed time format. The small-scale slot position type capable of enumerating the type can be matched with the slot position text and the text set corresponding to the slot position type, so that the verification operation is completed. If the slot position text is verified to have ambiguity, the corresponding slot position text ambiguity can be eliminated through a human-computer interaction mode or an automatic selection mode based on machine learning, the slot position text of the second slot position analysis result is updated, and a third slot position analysis result is finally obtained. Obviously, the slot text is verified by a local verification method with the adaptive type characteristics, so that the slot text ambiguity elimination step can be effectively simplified, and the slot text ambiguity elimination efficiency is improved.
Further, according to a preset slot text processing rule, slot text ambiguity of the second slot parsing result is eliminated, and a third slot parsing result is obtained, wherein the method comprises the following specific steps:
determining the slot position type of the second slot position analysis result;
when the slot type is a BI core data type, calling a business intelligent data interface to obtain interface data;
matching the slot position text of the second slot position analysis result with the interface data to obtain a matching result;
and updating the slot position text of the second slot position analysis result according to the matching result to obtain a third slot position analysis result.
It should be noted that the BI core data type herein includes indexes, dimensions, dimension items, and the like. When the slot type is the BI core data type, the business intelligence data interface needs to be called in real time to complete the verification. The BI core data type may be understood herein as a business intelligence core data type, and the business intelligence data interface may be understood as a BI data interface or a BI interface. Specifically, before specifically performing ambiguity elimination, a commercial intelligent data interface needs to be called in real time to obtain interface data, and the interface data is matched with the slot position text of the second slot position analysis result. And when the two are completely and uniquely matched, judging that the slot position text of the second slot position analysis result has no ambiguity. So far, the slot position text of the second slot position analysis result can be updated by using interface data, and a third slot position analysis result is finally obtained. And when the two can not be completely and uniquely matched, judging that the slot position text of the second slot position analytic result has ambiguity, and triggering slot position text ambiguity elimination operation. The interface data here may be understood as BI data. The slot text disambiguation operation can be completed in a human-computer interaction mode, and can also be completed by an automatic disambiguation method based on machine learning. When the slot position text ambiguity elimination operation is carried out in a man-machine interaction mode, the corresponding relation between the slot position text and the BI data needs to be interactively confirmed with a user, after the corresponding relation is confirmed successfully, the slot position text of the second slot position analysis result is replaced by the definite BI data, and finally a third slot position analysis result is obtained. Obviously, the slot text is verified through the commercial intelligent data interface, so that the slot text ambiguity elimination step can be effectively simplified, and the slot text ambiguity elimination efficiency is improved.
S500: and obtaining a billboard display parameter according to the second intention analysis result, the third slot analysis result and the global billboard parameter.
Specifically, obtaining a billboard display parameter according to the second intention analysis result, the third slot analysis result and the global billboard parameter, includes the following specific steps:
performing semantic verification according to the second intention analysis result and the third slot analysis result and in combination with a preset global billboard parameter to obtain a verification result;
and when the verification result is that the verification is passed, updating the second intention analysis result and the third slot analysis result to the global billboard parameter to obtain a billboard display parameter.
It is to be understood that the second intent parsing result and the third slot parsing result are semantically unambiguous information that has been ambiguities determined and/or processed. Obviously, in order to ensure the accuracy of the final kanban display parameter, the second intention analysis result, the third slot analysis result and the existing global kanban parameter need to be verified. The verification is the comprehensive verification of the intention represented by the second intention analysis result, the slot position information represented by the third slot position analysis result and the existing global kanban parameter, and the comprehensive verification can adopt a semantic verification method during specific operation, can be used for verifying the corresponding relation between the intention and the slot position information, and can also be used for verifying the logical relation between the intention, the slot position information and the global kanban parameter. Obviously, after the verification is passed, the second intention parsing result and the third slot parsing result can be confirmed to be information without ambiguity, and the global billboard parameter is updated according to the information, so that the latest billboard display parameter capable of representing the user display requirement is obtained. And performing semantic verification on the final disambiguation second intention analysis result, the final disambiguation third slot analysis result and the global billboard parameter, so that the recognition rate of the complex display requirement can be further improved.
S600: and updating the display content of the billboard according to the billboard display parameters.
It will be appreciated that the billboard presentation parameters may be used to drive the billboard to be presented in a particular mode. The kanban display parameters can effectively represent the real intention of the user, and the updated kanban display contents can finally complete the complex kanban display request.
Referring to fig. 2, the present application further provides a billboard display apparatus 100, comprising:
an input module 11, configured to input a billboard presentation request expressed in a natural language;
the natural language processing module 12 is configured to process the billboard display request through a natural language processing model to obtain a first intention analysis result and a first slot analysis result;
the first processing module 13 is configured to eliminate the intent ambiguity of the first intent parsing result, and obtain a second intent parsing result and a second slot parsing result according to the disambiguation result and the first slot parsing result;
the second processing module 14 is configured to eliminate slot position text ambiguity of the second slot position analysis result according to a preset slot position text processing rule, and obtain a third slot position analysis result;
the display preprocessing module 15 is configured to obtain a billboard display parameter according to the second intention analysis result, the third slot analysis result, and the global billboard parameter;
and the updating module 16 is used for updating the billboard display content according to the billboard display parameters.
It should be noted that, in the practical application of the present application, the natural language may be in various forms. Preferably, in order to fit the actual application scenario and facilitate the operation, the billboard presentation request may be request information expressed in a voice format or request information directly input in a text format. The voice or text information can be switched to different languages according to actual conditions. It should be noted that the billboard display request here may be the content or the form of the display that the user wishes to have the billboard displayed. A kanban is here understood to be a component or control or the like in software for controlling the graphical or the like display of relevant data. Obviously, the display data obtained after the treatment through the billboard can be finally displayed to the user through the display terminal. Obviously, the billboard display request is input in a voice or text mode, so that the billboard display request is more suitable for application scenes and convenient to use. The natural language processing model can be understood as a natural language understanding model, and is a processing model in the field of artificial intelligence. The kanban display request expressed in the natural language is processed by the natural language processing model, and then corresponding data which can be processed by a computer can be generated. The processing herein may be considered as part of the process of human-computer interaction. The first intent parsing result may be understood as an intent in natural language understanding, and the first slot parsing result may be understood as a slot corresponding to the intent. It is apparent that the slot herein also includes a slot type and a slot text. The generation process of the first intention parsing result may be understood as a process of intention recognition. The generation process of the first slot parsing result may be understood as a slot filling process. Obviously, the billboard display request is analyzed in a natural language processing model mode, so that the working efficiency can be effectively improved. The first intention analysis result is an intention recognition result automatically analyzed by the natural language processing model in a pipeline processing mode, and the possibility of ambiguity existing in the intention analysis of some complex sentences is increased. In order to improve the recognition rate of the complex sentence, the first intention parsing result may be subjected to intention disambiguation processing. The intention disambiguation process here may be understood as processing the first intention parsing result into intention information with clear intention. The intention information with clear intention may be understood as a second intention analysis result. After the intent ambiguity is resolved, slot position information corresponding to the second intent parsing result can be obtained by further processing according to the second intent parsing result and combining the first slot position parsing result, and the slot position information can be understood as the second slot position parsing result. Obviously, by eliminating the intention ambiguity, the recognition rate of the complex sentence can be effectively improved. The second slot parsing result may be understood as newly identified slot information. According to different slot types in the slot information, the verification and ambiguity elimination modes of the slot text are different. The billboard presentation parameters may be used to drive the billboard to be presented in a particular mode. The kanban display parameters can effectively represent the real intention of the user, and the updated kanban display contents can finally complete the complex kanban display request.
Further, the sample data set of the natural language processing model is generated by a sample generation algorithm, and the method comprises the following specific steps:
collecting samples according to preset intention category data, preset slot types and BI data, and adding the collected samples into an initial data set;
and according to the preset sampling number, circularly executing the sample acquisition operation to obtain a sample data set.
It can be understood that in the initial implementation stage of the technical solution of the present application, due to the lack of labeled samples for natural language processing model training, the rapid deployment of the system will be affected. In order to solve the cold start problem, the present application provides a sample generation algorithm, which can generate a labeled sample for training a natural language processing model by using preset intention category data, a preset slot type and BI data, where the labeled sample can be understood as a sample data set. The BI data mainly refers to indexes, dimensions and dimension data of customer original data after the customer original data are subjected to BI data governance specifications. For example, in the sentence "display total amount", the format may be converted to "display [ total amount ] (index)", and the "total amount" herein corresponds to index data. In a sentence "show [2020 ] (time) [ wuhan ] (dimension) [ profit ] (index) of [ jianghan ] (dimension) [ shorea ] (dimension)," wuhan "corresponds to dimension data," jiangshorea "corresponds to dimension data, and" profit "corresponds to index data. The BI data corresponds to an entity type of the slot, and the entity type of the slot herein may be understood as a preset slot type in the present application. And when the sample is collected, randomly selecting data from the BI data by adopting a non-replacement random sampling method to fill the data to the slot text position of the slot in the sample to form the slot text, and circulating from the beginning again after the slot text is exhausted. According to the actual display requirement, the preset intention category data can comprise a display billboard, a sequencing billboard, a trend billboard, a comparison billboard, a drill-down operation, a drill-up operation, a switching operation, a newly-added data slot position, a newly-added calculation operation, a newly-added comparison operation and the like. The preset slot type can comprise a basic type, a data-related type, a billboard-related type and other auxiliary types. Obviously, the preset intention category data and the preset slot type are only used for examples, and the actually set names and number can be adjusted accordingly according to the requirements. It should be noted that the basic types herein mainly include common entity types such as time, place, and number; the data correlation types mainly comprise three types of dimension modeling, namely indexes, dimensions and dimension items; the related types of the kanban comprise chart types, sorting modes, comparison operations, calculation operations and the like; other auxiliary types herein include types for disambiguating various slots. Obviously, the samples collected by the sample generation algorithm accord with the actual use environment, and the natural language processing model obtained by training can effectively improve the recognition rate.
Further, the first processing module 13 is configured to eliminate an intention ambiguity of the first intention parsing result, and obtain a second intention parsing result and a second slot parsing result according to the disambiguation result and the first slot parsing result, and specifically configured to:
analyzing the first intention analysis result according to a preset ambiguity judging method according to the probability distribution data of the first intention analysis result to obtain an ambiguity judging result;
when the ambiguity judgment result indicates that the first intention analysis result is ambiguous, processing the first intention analysis result through a preset intention ambiguity elimination method to obtain an intention analysis result to be verified;
according to a predefined rule, verifying the analysis result of the intention to be verified and the slot position type corresponding to the first slot position analysis result to obtain a verification result;
and when the verification result meets a preset condition, obtaining a second intention analysis result and a second slot position analysis result according to the verification result.
Note that, the probability distribution data here may be understood as a probability that the intention represented by the first intention analysis result is each intention in the preset intention category data. The preset ambiguity decision method can be set according to actual needs. In a specific implementation process, the following ambiguity determination methods may be preset: when the probability value corresponding to the intention with the maximum probability in the probability distribution data is smaller than a threshold value, judging that ambiguity exists; when the difference between the probability values corresponding to the two intentions with the highest probability in the probability distribution data is smaller than a threshold value, judging that ambiguity exists; when the first intent parsing result does not have corresponding intent data, but the first slot parsing result contains entity text, such first intent parsing result is determined to be ambiguous. The preset intent disambiguation method can be set according to actual needs. In a specific implementation process, the preset intent disambiguation method may be intent disambiguation of the first intent parsing result, which is completed in a human-computer interaction manner. The preset intention disambiguation method can also be that the intention disambiguation is completed through an intention sorting algorithm according to a machine learning model. For example, when the first intention parsing result does not have corresponding intention data, but a slot type exists in the first slot parsing result, and entity text data exists in a corresponding slot text, it is determined that the first intention parsing result has ambiguity; and prompting a user to select related preset intention category data according to the slot position type and the slot position text in the first slot position analysis result, thereby realizing intention ambiguity elimination. The predefined rule can be understood as the format, parameter number requirement, and the like of the preset slot bit data corresponding to the preset intention type data. For example, in the preset intention category data, the intention categories of "bulletin board" and "add calculation operator" are assumed to be preset. Under the intention of 'display kanban', the preset condition is that the number of indexes or dimensions cannot exceed 3; with the intention of "adding a calculation operator", the preset condition is that there must be at least one calculation operator (only type is verified, no value is verified). Obviously, when verification is performed, if the corresponding data does not meet the preset conditions, the corresponding verification fails, and finally the method for removing the intention ambiguity exits; and if the corresponding data meet the preset conditions, the verification is passed, and the next step of processing is carried out. Specifically, a set of formal verification rule sets with clear boundaries and suitable functions needs to be formulated by combining the functions of the existing billboard, so that an effective predefined rule is formed. It should be noted that when the verification result meets the preset condition after being judged, it may be determined that the verification result includes semantic clear intention information with ambiguity eliminated, that is, the second intention parsing result. According to the intention information with clear semantics, the operations of slot inheritance, slot replacement and the like can be further performed on the first slot analysis result. Thus, the slot data after the preliminary processing, i.e., the second slot parsing result, is obtained. Obviously, when the intention ambiguity elimination is carried out by combining the ambiguity judgment method, the intention ambiguity elimination method and the predefined rule, the processing efficiency can be effectively improved.
The technical scheme provided by the embodiment of the application at least has the following beneficial effects:
through multiple disambiguation processes, complex display requirements can be effectively identified.
It is to be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, the statement that there is an element defined as "comprising" … … does not exclude the presence of other like elements in the process, method, article, or apparatus that comprises the element.
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 modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.
Claims (10)
1. A billboard display method is characterized by comprising the following specific steps:
inputting a billboard display request expressed by a natural language;
processing the billboard display request through a natural language processing model to obtain a first intention analysis result and a first slot position analysis result;
eliminating the intention ambiguity of the first intention analysis result, and obtaining a second intention analysis result and a second slot position analysis result according to the disambiguation result and the first slot position analysis result;
eliminating slot position text ambiguity of the second slot position analysis result according to a preset slot position text processing rule to obtain a third slot position analysis result;
obtaining a billboard display parameter according to the second intention analysis result, the third slot analysis result and the global billboard parameter;
and updating the display content of the billboard according to the billboard display parameters.
2. The method according to claim 1, wherein the sample data set of the natural language processing model is generated by a sample generation algorithm, and comprises the following specific steps:
collecting samples according to preset intention category data, preset slot types and BI data, and adding the collected samples into an initial data set;
and according to the preset sampling number, circularly executing the sample acquisition operation to obtain a sample data set.
3. The display method according to claim 1, wherein the intent ambiguity of the first intent parsing result is resolved, and a second intent parsing result and a second slot parsing result are obtained according to the disambiguation result and the first slot parsing result, comprising the following steps:
analyzing the first intention analysis result according to a preset ambiguity judging method according to the probability distribution data of the first intention analysis result to obtain an ambiguity judging result;
when the ambiguity judgment result indicates that the first intention analysis result is ambiguous, processing the first intention analysis result through a preset intention ambiguity elimination method to obtain an intention analysis result to be verified;
according to a predefined rule, verifying the analysis result of the intention to be verified and the slot position type corresponding to the first slot position analysis result to obtain a verification result;
and when the verification result meets a preset condition, obtaining a second intention analysis result and a second slot position analysis result according to the verification result.
4. The display method according to claim 1, wherein the intent ambiguity of the first intent parsing result is resolved, and a second intent parsing result and a second slot parsing result are obtained according to the disambiguation result and the first slot parsing result, comprising the following steps:
analyzing the first intention analysis result according to a preset ambiguity judging method according to the probability distribution data of the first intention analysis result to obtain an ambiguity judging result;
when the ambiguity judgment result indicates that the first intention analysis result is ambiguous, verifying the first intention analysis result and the slot position type corresponding to the first slot position analysis result according to a predefined rule to obtain a verification result;
and when the verification result meets a preset condition, obtaining a second intention analysis result and a second slot position analysis result according to the verification result.
5. The display method according to claim 1, wherein the slot text ambiguity of the second slot parsing result is eliminated according to a preset slot text processing rule to obtain a third slot parsing result, and the method comprises the following specific steps:
determining the slot position type of the second slot position analysis result;
when the slot type is a non-BI core data type, selecting a local verification method adaptive to the type characteristics to verify the slot text according to the data type characteristics to obtain a verification result;
and updating the slot position text of the second slot position analysis result according to the verification result to obtain a third slot position analysis result.
6. The display method according to claim 1, wherein the slot text ambiguity of the second slot parsing result is eliminated according to a preset slot text processing rule to obtain a third slot parsing result, and the method comprises the following specific steps:
determining the slot position type of the second slot position analysis result;
when the slot type is a BI core data type, calling a business intelligent data interface to obtain interface data;
matching the slot position text of the second slot position analysis result with the interface data to obtain a matching result;
and updating the slot position text of the second slot position analysis result according to the matching result to obtain a third slot position analysis result.
7. The display method according to claim 1, wherein the kanban display parameters are obtained according to the second intention analysis result, the third slot analysis result and the global kanban parameters, and the method comprises the following specific steps:
performing semantic verification according to the second intention analysis result and the third slot analysis result and in combination with a preset global billboard parameter to obtain a verification result;
and when the verification result is that the verification is passed, updating the second intention analysis result and the third slot analysis result to the global billboard parameter to obtain a billboard display parameter.
8. A billboard display apparatus comprising:
the input module is used for inputting a billboard display request expressed by a natural language;
the natural language processing module is used for processing the billboard display request through a natural language processing model to obtain a first intention analysis result and a first slot position analysis result;
the first processing module is used for eliminating the intention ambiguity of the first intention analysis result and obtaining a second intention analysis result and a second slot position analysis result according to the disambiguation result and the first slot position analysis result;
the second processing module is used for eliminating slot position text ambiguity of the second slot position analysis result according to a preset slot position text processing rule to obtain a third slot position analysis result;
the display preprocessing module is used for obtaining a billboard display parameter according to the second intention analysis result, the third slot analysis result and the global billboard parameter;
and the updating module is used for updating the display content of the billboard according to the billboard display parameter.
9. The presentation device according to claim 8, wherein the sample data set of the natural language processing model is generated by a sample generation algorithm, comprising the following specific steps:
collecting samples according to preset intention category data, preset slot types and BI data, and adding the collected samples into an initial data set;
and according to the preset sampling number, circularly executing the sample acquisition operation to obtain a sample data set.
10. The display device according to claim 8, wherein the first processing module is configured to eliminate an intention ambiguity of the first intention parsing result, and obtain a second intention parsing result and a second slot parsing result according to the disambiguation result and the first slot parsing result, and specifically configured to:
analyzing the first intention analysis result according to a preset ambiguity judging method according to the probability distribution data of the first intention analysis result to obtain an ambiguity judging result;
when the ambiguity judgment result indicates that the first intention analysis result is ambiguous, processing the first intention analysis result through a preset intention ambiguity elimination method to obtain an intention analysis result to be verified;
according to a predefined rule, verifying the analysis result of the intention to be verified and the slot position type corresponding to the first slot position analysis result to obtain a verification result;
and when the verification result meets a preset condition, obtaining a second intention analysis result and a second slot position analysis result according to the verification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111220334.6A CN113918701B (en) | 2021-10-20 | 2021-10-20 | Billboard display method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111220334.6A CN113918701B (en) | 2021-10-20 | 2021-10-20 | Billboard display method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113918701A CN113918701A (en) | 2022-01-11 |
| CN113918701B true CN113918701B (en) | 2022-04-15 |
Family
ID=79241769
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111220334.6A Active CN113918701B (en) | 2021-10-20 | 2021-10-20 | Billboard display method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113918701B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108962233A (en) * | 2018-07-26 | 2018-12-07 | 苏州思必驰信息科技有限公司 | Voice dialogue processing method and system for voice dialogue platform |
| CN111026842A (en) * | 2019-11-29 | 2020-04-17 | 微民保险代理有限公司 | Natural language processing method, natural language processing device and intelligent question-answering system |
| CN111368548A (en) * | 2018-12-07 | 2020-07-03 | 北京京东尚科信息技术有限公司 | Semantic recognition method and device, electronic equipment and computer-readable storage medium |
| CN111949787A (en) * | 2020-08-21 | 2020-11-17 | 平安国际智慧城市科技股份有限公司 | Automatic question-answering method, device, equipment and storage medium based on knowledge graph |
| CN111985249A (en) * | 2020-09-03 | 2020-11-24 | 贝壳技术有限公司 | Semantic analysis method and device, computer-readable storage medium and electronic equipment |
| CN112035506A (en) * | 2019-10-28 | 2020-12-04 | 竹间智能科技（上海）有限公司 | Semantic recognition method and equipment |
| CN113435205A (en) * | 2021-04-29 | 2021-09-24 | 北京小米移动软件有限公司 | Semantic parsing method and device |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10217462B2 (en) * | 2016-08-31 | 2019-02-26 | Microsoft Technology Licensing, Llc | Automating natural language task/dialog authoring by leveraging existing content |
| US10496382B2 (en) * | 2018-02-22 | 2019-12-03 | Midea Group Co., Ltd. | Machine generation of context-free grammar for intent deduction |
| US10902220B2 (en) * | 2019-04-12 | 2021-01-26 | The Toronto-Dominion Bank | Systems and methods of generating responses associated with natural language input |
-
2021
- 2021-10-20 CN CN202111220334.6A patent/CN113918701B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108962233A (en) * | 2018-07-26 | 2018-12-07 | 苏州思必驰信息科技有限公司 | Voice dialogue processing method and system for voice dialogue platform |
| CN111368548A (en) * | 2018-12-07 | 2020-07-03 | 北京京东尚科信息技术有限公司 | Semantic recognition method and device, electronic equipment and computer-readable storage medium |
| CN112035506A (en) * | 2019-10-28 | 2020-12-04 | 竹间智能科技（上海）有限公司 | Semantic recognition method and equipment |
| CN111026842A (en) * | 2019-11-29 | 2020-04-17 | 微民保险代理有限公司 | Natural language processing method, natural language processing device and intelligent question-answering system |
| CN111949787A (en) * | 2020-08-21 | 2020-11-17 | 平安国际智慧城市科技股份有限公司 | Automatic question-answering method, device, equipment and storage medium based on knowledge graph |
| CN111985249A (en) * | 2020-09-03 | 2020-11-24 | 贝壳技术有限公司 | Semantic analysis method and device, computer-readable storage medium and electronic equipment |
| CN113435205A (en) * | 2021-04-29 | 2021-09-24 | 北京小米移动软件有限公司 | Semantic parsing method and device |
Non-Patent Citations (3)
| Title |
|---|
| Sentence simplification for spoken language understanding;Gokhan Tur等;《2011 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)》;20110712;第5628-5631页 * |
| 基于领域知识的增强约束词向量;王恒升等;《中文信息学报》;20190430;第37-47页 * |
| 对话*** intent和slot识别;腾讯知文实验室;《https://cloud.tencent.com/developer/article/1147476?fromSource=waituihttps://cloud.tencent.com/developer/article/1147476?fromSource=waitui》;20180613;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113918701A (en) | 2022-01-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9524291B2 (en) | Visual display of semantic information | |
| CN112100354A (en) | Man-machine conversation method, device, equipment and storage medium | |
| CN110765342A (en) | Information query method and device, storage medium and intelligent terminal | |
| CN116737908A (en) | Knowledge question-answering method, device, equipment and storage medium | |
| CN115328756A (en) | Test case generation method, device and equipment | |
| CN111858888A (en) | Multi-round dialogue system of check-in scene | |
| CN111738018A (en) | Intention understanding method, device, equipment and storage medium | |
| CN112749081A (en) | User interface testing method and related device | |
| CN115048505A (en) | Corpus screening method and device, electronic equipment and computer readable medium | |
| CN115509485A (en) | Filling-in method and device of business form, electronic equipment and storage medium | |
| CN110890090B (en) | Context-based auxiliary interaction control method and system | |
| CN114662470A (en) | Product comment information processing method and system combining big data | |
| CN108984777B (en) | Customer service method, apparatus and computer-readable storage medium | |
| CN113918701B (en) | Billboard display method and device | |
| CN112988704A (en) | AI consultation database cluster building method and system | |
| CN112288584A (en) | Insurance application processing method and device, computer readable medium and electronic equipment | |
| CN117130593A (en) | Code processing method, system and electronic equipment | |
| CN112287077A (en) | Statement extraction method and device for combining RPA and AI for document, storage medium and electronic equipment | |
| CN114490969B (en) | Question and answer method and device based on table and electronic equipment | |
| CN111125368A (en) | Legal opinion book generation method based on target object | |
| CN116186219A (en) | Man-machine dialogue interaction method, system and storage medium | |
| CN112860863A (en) | Machine reading understanding method and device | |
| KR102118322B1 (en) | Document translation server and translation method for generating original and translation files individually | |
| CN112036188A (en) | Method and device for recommending quality test example sentences | |
| CN110209831A (en) | Model generation, the method for semantics recognition, system, 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |