CN107515857B - Semantic understanding method and system based on customization technology - Google Patents

Semantic understanding method and system based on customization technology Download PDF

Info

Publication number
CN107515857B
CN107515857B CN201710774597.9A CN201710774597A CN107515857B CN 107515857 B CN107515857 B CN 107515857B CN 201710774597 A CN201710774597 A CN 201710774597A CN 107515857 B CN107515857 B CN 107515857B
Authority
CN
China
Prior art keywords
intention
template
customized
probability
skill
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
Application number
CN201710774597.9A
Other languages
Chinese (zh)
Other versions
CN107515857A (en
Inventor
黄鑫
陈志刚
王智国
胡国平
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
iFlytek Co Ltd
Original Assignee
iFlytek Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by iFlytek Co Ltd filed Critical iFlytek Co Ltd
Priority to CN201710774597.9A priority Critical patent/CN107515857B/en
Publication of CN107515857A publication Critical patent/CN107515857A/en
Application granted granted Critical
Publication of CN107515857B publication Critical patent/CN107515857B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/30Semantic analysis

Abstract

The invention discloses a semantic understanding method and a semantic understanding system based on customization technology, wherein the method comprises the following steps: receiving user request data; acquiring a path of a current customization skill, and acquiring semantic resources of the current customization skill according to the path; determining that the request data belongs to a statement template of the current customized skill, wherein the statement template comprises one or more semantic slots, and each semantic slot corresponds to an entity; and determining a semantic slot and a corresponding entity in the request data according to the saying template to obtain a semantic understanding result. By using the invention, the semantic understanding of the user request data can be realized quickly and accurately.

Description

Semantic understanding method and system based on customization technology
Technical Field
The invention relates to the field of natural language processing, in particular to a semantic understanding method and a semantic understanding system based on customization technology.
Background
With the development of mobile intelligent terminals and information network technologies, people use voice interaction applications in more and more scenes, for example, applications such as a smart speaker as a portal, and the voice interaction uses weather/stock/music/traffic conditions/alarm clock/reminders, each of which includes one or more skills. For example, the car-mounted device is used as an entrance, and voice interaction uses LBS/navigation/radio station/music and other applications or skills. Thus, there are more and more open platforms of conversational artificial intelligence, such as hundred DuerOS, amazon's Alexa, etc., that provide a system for application developers or skill developers to develop applications or skills. Meanwhile, for the purpose of creating a mutually-shared and win-win rich ecology by an open platform, a market in a similar form of a skill mall is also provided, so that the developed skills of developers can be released into the mall, can be directly selected and used by terminal users, and can also be selected and used by third-party application developers in applications and provided for the terminal users through the terminal applications.
In order to further meet the personalized requirements of different applications, an application developer can customize the skills according to the requirements of the application developer when using the skills, some skills further provide a customization function, for example, the customization authority is set for the skills, and the application developer can customize the skills within the range of the customization authority when using the skills to obtain the customization skills. Of course, the application developer can not only apply the customized skills to the newly developed application program, but also release the customized skills as shared skills to a skill mall for other developers or end users to choose to use.
For applications in natural language understanding, such as voice interactive applications, a user typically uses the application with a client that receives user request data, semantically understands the user request data with skills in the application, and returns a response to the user.
For the customization skills, how to quickly and accurately understand the semantics of the data requested by the user does not have an effective solution at present.
Disclosure of Invention
The embodiment of the invention provides a semantic understanding method and a semantic understanding system based on customization technology, which are used for realizing rapid and accurate semantic understanding of user request data.
Therefore, the invention provides the following technical scheme:
a semantic understanding method based on customization technology comprises the following steps:
receiving user request data;
acquiring a path of a current customization skill, and acquiring semantic resources of the current customization skill according to the path;
determining that the request data belongs to a statement template of the current customized skill, wherein the statement template comprises one or more semantic slots, and each semantic slot corresponds to an entity;
and determining a semantic slot and a corresponding entity in the request data according to the saying template to obtain a semantic understanding result.
Preferably, said determining that said requested data belongs to a template of a current customized skill comprises:
determining an intent that the requested data pertains to a current customization skill;
and matching the request data with the explanation templates corresponding to the intentions to obtain the explanation templates corresponding to the request data.
Preferably, the method further comprises:
training a skill intention determination model in advance according to a statement template of the current custom skill intention; an intention ranking model is constructed by using the intention probability output during the training of the skill intention determining model;
the determining the intent that the requested data belongs to the current custom skill comprises:
determining, using the skill intent determination model, a probability that the requested data belongs to each of the customized skill intents;
inputting the probability that the request data belongs to each intention in the customized skill intentions into the intention ranking model, and obtaining the ranked intention probability according to the output of the intention ranking model;
and selecting the intention corresponding to the intention probability with the highest ranking as the intention of the request data.
Preferably, the current intent to customize skills includes: a customization intent and a customized intent; the customization intention refers to an intention which is newly added when the skill is customized currently and an intention obtained by modifying the customized skill intention; the customized intent refers to an unmodified intent of the base skill intents used in the current customization of the skill;
the method further comprises the following steps:
training a customizing intention determining model and a customized intention determining model in advance according to the saying template of the customizing intention and the saying template of the customized intention; an intention ordering model is constructed by utilizing the customization intention determining model and the intention probability output when the customized intention determining model is trained;
the determining the intent that the requested data belongs to the current custom skill comprises:
determining a probability that the request data belongs to each of the customization intents and a probability that each of the customized intents using the customization intent determination model and customized intent determination model;
inputting the probability that the request data respectively belong to each intention in the customizing intentions and the probability of each intention in the customized intentions into the intention sequencing model, and obtaining the sequenced intention probability according to the output of the intention sequencing model;
and selecting the intention corresponding to the intention probability with the highest ranking as the intention of the request data.
Preferably, the method further comprises:
training a skill description template to determine a model according to a description template of the current skill customization intention, and constructing a description template ordering model by using the description template probability output during model training;
the determining that the requested data belongs to a statement template of a current customized skill comprises:
determining a model by using the skill utterance template, and determining the probability that the request data belongs to each utterance template in the intention of customizing skills;
inputting the probability of each statement template in the intention that the request data belongs to the customized skills into the statement template sequencing model, and obtaining the sequenced statement template probability according to the output of the statement template sequencing model;
and selecting the description template corresponding to the description template probability with the top ranking as the description template corresponding to the request data.
Preferably, the current intent to customize skills includes: a customization intent and a customized intent; the customization intention refers to an intention which is newly added when the skill is customized currently and an intention obtained by modifying the customized skill intention; the customized intent refers to an unmodified intent of the base skill intents used in the current customization of the skill;
the method further comprises the following steps:
training a customizing intention explanation template determining model and a customized intention explanation template determining model in advance according to an explanation template of a customizing intention and an explanation template of a customized intention; constructing a description template ordering model by using the customization intention description template determining model and the description template probability output during the training of the customized intention description template determining model;
the determining that the requested data belongs to a statement template of a current customized skill comprises:
determining a probability that the request data belongs to each utterance template in the customization intent and a probability of each utterance template in the customized intent using the customization intent utterance template determination model and the customized intent utterance template determination model;
inputting the probability that the request data respectively belongs to each description template in the customization intention and the probability of each description template in the customized intention into the description template sequencing model, and obtaining the sequenced description template probability according to the output of the description template sequencing model;
and selecting the description template corresponding to the description template probability with the top ranking as the corresponding description template.
A custom skill based semantic understanding system comprising:
the receiving module is used for receiving user request data;
the resource acquisition module is used for acquiring a path of the current customization skill and acquiring semantic resources of the current customization skill according to the path;
the explanation template determining module is used for determining that the request data belongs to an explanation template of the current customized skill, the explanation template comprises one or more semantic slots, and each semantic slot corresponds to one entity;
and the semantic understanding module is used for determining a semantic slot and a corresponding entity in the request data according to the saying template to obtain a semantic understanding result.
Preferably, the utterance template determination module includes:
an intent determination module to determine an intent that the requested data pertains to a current customization skill;
and the template matching module is used for matching the request data with the explanation templates corresponding to the intentions to obtain the explanation templates corresponding to the request data.
Preferably, the system further comprises:
the first model building module is used for training a skill intention determination model in advance according to a description template of the intention of the current customized skill; an intention ranking model is constructed by using the intention probability output during the training of the skill intention determining model;
the intent determination module includes:
a first intent probability determination unit for determining a probability that the request data belongs to each of the customized skill intentions using the skill intent determination model;
the first sequencing unit is used for inputting the probability that the request data belongs to each intention in the customized skill intentions into the intention sequencing model and obtaining the sequenced intention probability according to the output of the intention sequencing model;
and the first selection unit is used for selecting the intention corresponding to the intention probability ranked at the top as the intention of the request data.
Preferably, the current intent to customize skills includes: a customization intent and a customized intent; the customization intention refers to an intention which is newly added when the skill is customized currently and an intention obtained by modifying the customized skill intention; the customized intent refers to an unmodified intent of the base skill intents used in the current customization of the skill;
the system further comprises:
the second model building module is used for training the customizing intention determining model and the customized intention determining model in advance according to the saying template of the customizing intention and the saying template of the customized intention; an intention ordering model is constructed by utilizing the customization intention determining model and the intention probability output when the customized intention determining model is trained;
the intent determination module includes:
a second intention probability determination unit for determining a probability that the request data belongs to each of the customizing intentions and a probability of each of the customized intentions using the customizing intention determination model and the customized intention determination model;
the second sequencing unit is used for inputting the probability that the request data respectively belongs to each intention in the customizing intentions and the probability of each intention in the customized intentions into the intention sequencing model and obtaining the sequenced intention probability according to the output of the intention sequencing model;
and the second selection unit is used for selecting the intention corresponding to the intention probability ranked furthest ahead as the intention of the request data.
Preferably, the system further comprises:
the third model building module is used for training the skill description template to determine a model in advance according to the description template of the current skill customization intention, and building a description template ordering model by using the description template probability output during model training;
the utterance template determination module includes:
a third probability determination unit, configured to determine, using the skill utterance template determination model, a probability that the request data belongs to each utterance template in the intent to customize skills;
the third sequencing unit is used for inputting the probability of each statement template in the intention that the request data belongs to the customized skill into the statement template sequencing model and obtaining the probability of the sequenced statement templates according to the output of the statement template sequencing model;
and the third selecting unit is used for selecting the description template corresponding to the description template probability which is ranked most front as the description template corresponding to the request data.
Preferably, the current intent to customize skills includes: a customization intent and a customized intent; the customization intention refers to an intention which is newly added when the skill is customized currently and an intention obtained by modifying the customized skill intention; the customized intent refers to an unmodified intent of the base skill intents used in the current customization of the skill;
the system further comprises:
the fourth model building module is used for training the customizing intention explanation template determining model and the customized intention explanation template determining model in advance according to the explanation template of the customizing intention and the explanation template of the customized intention; constructing a description template ordering model by using the customization intention description template determining model and the description template probability output during the training of the customized intention description template determining model;
the utterance template determination module includes:
a fourth probability determination unit for determining a probability that the request data belongs to each of the utterance templates in the customization intention and a probability of each of the utterance templates in the customized intention by using the customization intention utterance template determination model and the customized intention utterance template determination model;
the fourth ordering unit is used for inputting the probability that the request data respectively belongs to each description template in the customizing intention and the probability of each description template in the customized intention into the description template ordering model and obtaining the ordered description template probability according to the output of the description template ordering model;
and the fourth selection unit is used for selecting the description template corresponding to the description template probability which is ranked most front as the corresponding description template.
According to the semantic understanding method and system based on the customization skills, provided by the embodiment of the invention, after user request data is received, a path of the current customization skills is obtained, all semantic resources of the customization skills are obtained according to the path, a statement template of the current customization skills of the request data is determined, a semantic slot and a corresponding entity in the request data are determined according to the statement template, and a semantic understanding result is obtained. By using the invention, the semantic understanding of the user request data can be realized quickly and accurately.
Drawings
In order to more clearly illustrate the embodiments of the present application or technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present invention, and other drawings can be obtained by those skilled in the art according to the drawings.
FIG. 1 is a flow chart of a semantic understanding method based on customization skills according to an embodiment of the present invention;
FIG. 2 is a flow chart of an embodiment of the present invention for determining a utterance template corresponding to requested data;
FIG. 3 is another flow chart of determining a grammar template corresponding to requested data in an embodiment of the invention;
FIG. 4 is another flow chart of determining a grammar template corresponding to requested data in an embodiment of the invention;
FIG. 5 is another flow chart of determining a grammar template corresponding to requested data in an embodiment of the invention;
FIG. 6 is a schematic structural diagram of a semantic understanding system based on customization skills according to an embodiment of the invention.
Detailed Description
In order to make the technical field of the invention better understand the scheme of the embodiment of the invention, the embodiment of the invention is further described in detail with reference to the drawings and the implementation mode.
As shown in fig. 1, the method is a flowchart of a semantic understanding method based on customization techniques, and includes the following steps:
step 101, receiving user request data.
The user request data is used for expressing a current request of a user, and specifically may be voice data, text data, or other multimedia request data, such as image data. If the user says that the voice data is requested, "I want to eat the palace chicken bouillon"; of course, if the user request data is voice data, the voice data also needs to be recognized as text data, and if the user request data is image data, the image data needs to be recognized as text data through an image recognition method.
And 102, acquiring a path of the current customization skill, and acquiring semantic resources of the current customization skill according to the path.
The customization skills refer to the skills developed by an application developer according to the application requirements of the application developer on the basis of other skills. Generally, a great number of classified sharing skills are contained in a skill mall, an application developer can select the sharing skills in corresponding categories according to application requirements and then customize the sharing skills according to the requirements of the application developer, for example, the application developer intends to develop an application for listening to music, and after browsing the skill mall, a music walkman skill is found, the skill includes a music listening service, and the application developer performs skill customization according to the requirements of the application developer on the basis of the skill, for example, social functions such as adding a song recommended to a friend on the music walkman skill, broadcasting a piece of music and guessing the name of the song by the friend are added on the music walkman skill.
Semantic resources of a skill, i.e. resources used in semantic understanding of the skill, such as semantic templates, semantic slots, entities, etc., are generally stored on a server. Because the skill number of an open platform is large, a distributed storage mode is usually adopted, namely semantic resources with different skills are possibly stored on different servers; thus, for multiple customized skills, i.e., customized skills, the semantic resources thereof are typically stored on multiple servers. Therefore, when semantic understanding is performed, all semantic resources of the current customization skills can be acquired from different servers at first, and the semantic resources are put on the same server, so that the semantic resources are conveniently loaded into a memory when a program is executed.
Since each skill registers the address of the server where the semantic resource of the skill is located with the resource management server at the time of development, customizing the skill is no exception. Therefore, when the semantic resources of the current customized skill are obtained, the server address of the semantic resource of each skill on the path can be sequentially found according to the path of the current customized skill, all the semantic resources of the customized skill on the path can be obtained according to the server address, and the semantic resources are put on the same server.
Furthermore, in order to conveniently use the current customization skill subsequently, the number of the current customization skill and the address of the server where the semantic resource corresponding to the number of the current customization skill is located can be stored, and when the customization skill is reused next time, the number can be directly obtained from the corresponding server according to the stored server address.
Step 103, determining that the request data belongs to a statement template of the current customized skill, wherein the statement template comprises one or more semantic slots, and each semantic slot corresponds to one entity.
Since each skill may include multiple intents, each intent includes multiple grammar templates. Therefore, when determining that the request data belongs to the utterance template of the current customization skill, the intention of the request data belonging to the current customization skill can be determined, and then the request data is matched with the utterance templates corresponding to the intention to obtain the utterance template corresponding to the request data; it is also possible to directly determine to which grammar template the requested data belongs for the current customization skill.
The above-described several different ways will be described in detail later.
And 104, determining a semantic slot and a corresponding entity in the request data according to the saying template to obtain a semantic understanding result.
Specifically, the semantic slots and the corresponding entities in the request data may be determined by an exact matching method or a fuzzy matching method. For example, the user request data is 'my main point exploded diced chicken', the intention of the request data is an ordering intention, according to the explanation templates of the ordering intention, the explanation template of the current user request data belonging to the current customization skill can be determined to be 'my main point dis', wherein the dis is a semantic slot, the 'my main point exploded diced chicken' of the user request data is matched with the explanation template, and the entity corresponding to the semantic slot dis is 'uterus exploded diced'.
It should be noted that, if the user request data is matched with the utterance template when determining the utterance template to which the user request data belongs, the semantic slot and the corresponding entity in the user request data can be directly obtained according to the matching result, and the matching does not need to be performed again.
Fig. 2 is a flowchart of determining a saying template corresponding to requested data in the embodiment of the present invention, which includes the following steps:
step 201, training a customizing intention determining model and a customized intention determining model in advance according to a saying template of a customizing intention and a saying template of a customized intention; and constructing an intention ordering model by using the customization intention determining model and the intention probability output by the customization intention determining model during training.
The current intent to customize skills is divided into two categories, namely: a customization intent and a customized intent; the customization intention refers to an intention which is newly added when the skill is customized currently and an intention obtained by modifying the customized skill intention; the customized intent refers to an unmodified intent of the base skill intents used in the current customization of the skill. It can be seen that the customized intent is a subset of the intent in the customized skill, as the customized intent is likely to be modified at the time of customization.
When the customized intention determining model is trained, the explanation template and the entity of the customized intention are used as the positive example training data, the entity is the entity data corresponding to the semantic slot in the explanation template, the entity data is added into the corresponding semantic slot to obtain a piece of positive example training data, a large number of explanation templates and entities with non-customized intentions are collected to be used as the negative example training data, for example, part of explanation templates and entities can be selected from the general intentions of an open platform to be used as the negative example training data, and the customized intention determining model is trained. The input of the customization intention determining model is word vectors of positive example training data and negative example training data, and the output is the probability that each training data belongs to each customization intention. If the custom skill contains 5 intents, the output of the custom intent determination model is the probability that each training data belongs to these 5 intents; the training data are the positive and negative examples data.
Similarly, the training process of the customized intent determination model is similar to the training process of the customized intent determination model, namely, the utterance template and the entity of the customized intent are used as the normal training data, a large number of utterance templates and entities without the customized intent are collected as the reverse training data, for example, a part of the utterance template and the entity can be selected from the general intent of an open platform to be used as the reverse training data, and the customized intent determination model is trained. The input of the customized intent determination model is word vectors of positive training data and negative training data; outputting a probability that each training data belongs to each customized intent; if the customized skill contains 10 intents, the output of the customized intent determination model is the probability that each training data belongs to these 10 intents.
The customization intention determining model and the customized intention determining model can adopt a classification model commonly used in pattern recognition, such as a deep neural network model, a convolutional neural network model, a support vector machine model and the like, and a specific training method is the same as that of the prior art and is not described in detail herein.
When an intention sequencing model is constructed, a customization intention determining model and a customized intention determining model need to be marked for training, training data output by the model belongs to intention probability with the highest priority in the probability of each intention, and the priorities of other probabilities are not limited; and taking the probability that each training data belongs to each intention as the input of the ranking model, outputting the probability as the ranked probability, ranking the highest intention probability with the highest priority in the top after ranking, and not limiting the sequence of other intention probabilities. If there are 5 custom intent probabilities and 10 custom intent probabilities, the input to the intent ranking model is the 15 intent probabilities, the output is the 15 ranked intent probabilities, and the top is the highest priority intent probability.
The intention ranking model can adopt a neural network model, such as a convolutional neural network model, and the specific training method is the same as that of the prior art and is not described in detail herein.
Step 202, utilizing the customizing intention determining model and the customized intention determining model, determining the probability that the request data belongs to each intention in the customizing intentions and the probability of each intention in the customized intentions.
Specifically, when the probability that the request data belongs to each intention in the customizing intentions and the probability of each intention in the customized intentions are determined, the word vectors of the request data are directly used as the input of the customizing intention determining model and the customized intention determining model respectively, and the probability that the request data belongs to each intention in the customizing intentions and the probability of each intention in the customized intentions respectively are obtained according to the output of the models.
Step 203, inputting the probability that the request data respectively belongs to each intention in the customizing intentions and the probability of each intention in the customized intentions into the intention ranking model, and obtaining the ranked intention probability according to the output of the intention ranking model.
And step 204, selecting the intention corresponding to the intention probability with the highest ranking as the intention of the request data.
Step 205, matching the request data with each statement template corresponding to the intention to obtain the statement template corresponding to the request data.
Specifically, a method of exact matching or fuzzy matching may be employed.
Correspondingly, the semantic understanding is carried out on the request data by using the description template to obtain a semantic understanding result. If the user request data is 'my main point diced chicken with explosion in palace', the intention of the request data is the dish ordering intention, and according to the saying template of the dish ordering intention, the semantic slot of the current user request data can be determined to be 'my main point dish', wherein dish is the semantic slot, and the corresponding entity is 'diced chicken with explosion in palace'.
Fig. 3 is another flowchart of determining a saying template corresponding to the requested data in the embodiment of the present invention, which includes the following steps:
step 301, training a skill intention determining model in advance according to a description template of the intention of the current customized skill, and constructing an intention ranking model by using the intention probability output during training of the skill intention determining model.
Specifically, the utterance templates and corresponding entities of the customized intention and the customized intention are used as regular training data together, a large number of utterance templates and entities of the non-customized intention and the non-customized intention are collected as counter training data, for example, a part of utterance templates and entities can be selected from general intentions of an open platform to be used as counter training data, and the skill intention determination model is trained. The input of the skill intention determination model is word vectors of positive training data and negative training data, and the output is the probability that each training data belongs to each intention in the current customized skill.
When an intention ranking model is constructed, the intention probability output by the model during model training can be directly determined by using the skill intention, and the specific training method is similar to the process, namely when the skill intention determining model is labeled for training, the output of the model is the intention probability with the highest priority in the probability that the training data belongs to each intention, and the priorities of other probabilities are not limited; and training an intention ranking model by using the labeled intention probability.
Step 302, using the skill intent determination model, determining a probability that the requested data belongs to each of the customized skill intents.
Specifically, when the probability that the request data belongs to each intention in the skill intentions is determined, the word vector of the request data is directly used as the input of a skill intention determination model, and the probability that the request data belongs to each intention in the skill intentions is obtained according to the output of the model.
And 303, inputting the probability that the request data belongs to each intention in the customized skill intentions into the intention ranking model, and obtaining the ranked intention probability according to the output of the intention ranking model.
And step 304, selecting the intention corresponding to the highest-ranked intention probability as the intention of the request data.
And 305, matching the request data with each statement template corresponding to the intention to obtain the statement template corresponding to the request data.
Fig. 4 is another flowchart of determining a saying template corresponding to the requested data in the embodiment of the present invention, which includes the following steps:
step 401, training a skill description template determination model according to a description template of the current skill customization intention, and establishing a description template ordering model by using the description template probability output during model training.
Specifically, a description template of the current skill customization intention and a corresponding entity are taken as the formal training data together, the entity is entity data corresponding to a semantic slot in the description template, and the entity data is added into the corresponding semantic slot to obtain a piece of formal training data; collecting a large number of utterance templates and entities of the intent of the non-customized skill as counterexample training data, correspondingly adding the entity data into the corresponding utterance templates to obtain a piece of counterexample training data, for example, selecting a part of the utterance templates and the entities from the general intent of the open platform as counterexamples, and training the skill utterance templates to determine the model. And determining word vectors of the model with positive training data and negative training data as input of the skill utterance template, and outputting the probability that each training data belongs to each utterance template in the current customized skill.
When constructing a saying template sequencing model, a skill saying template can be directly used to determine the saying template probability output by the model during model training, the specific training method is similar to the process, namely when the skill saying template is marked to determine the model training, the training data output by the model belongs to the saying template probability with the highest priority in the probability of each saying template, and the priorities of other probabilities are not limited; and training a saying template ordering model by using the annotated saying template probability.
Step 402, determining a model by using the skill utterance template, and determining the probability that the request data belongs to each utterance template in the intention of customizing skills.
And 403, inputting the probability of each statement template in the intention that the request data belongs to the customized skills into the statement template sequencing model, and obtaining the probability of the sequenced statement templates according to the output of the statement template sequencing model.
And 404, selecting the description template corresponding to the description template probability with the top ranking as the description template corresponding to the request data.
Fig. 5 is another flowchart of determining a saying template corresponding to the requested data in the embodiment of the present invention, which includes the following steps:
step 501, training a customizing intention explanation template determining model and a customized intention explanation template determining model in advance according to an explanation template of a customizing intention and an explanation template of a customized intention; and constructing a description template ordering model by using the customization intention description template determining model and the description template probability output when the customized intention description template determining model is trained.
The current intent to customize skills is divided into two categories, namely: a customization intent and a customized intent; the customization intention refers to an intention which is newly added when the skill is customized currently and an intention obtained by modifying the customized skill intention; the customized intent refers to an unmodified intent of the base skill intents used in the current customization of the skill.
When the customized intention explanation template determining model is trained, an explanation template and an entity of a customized intention are used as positive example training data, the entity is entity data corresponding to a semantic slot in the explanation template, the entity data is added into the corresponding semantic slot to obtain a piece of positive example training data, a large number of explanation templates and entities with non-customized intentions are collected to be used as negative example training data, correspondingly, the entity data is added into the corresponding explanation template to obtain a piece of negative example training data, for example, a part of explanation templates and the entity can be selected from general intentions of an open platform to be used as negative example training data, and the customized intention explanation template determining model is trained. The custom intention explanation template determines word vectors of positive example training data and negative example training data which are input into the model, and outputs the probability of each training data belonging to each explanation template in the custom intention.
Similarly, the training process of the customized intention template determination model is similar to the training process of the customized intention template determination model, namely, the customized intention template and the customized intention template determination model are used as the positive example training data, a large number of non-customized intention templates and non-customized intention templates are collected as the negative example training data, for example, a part of the customized intention template and non-customized intention templates can be selected from the general intention of an open platform to be used as the negative example training data, and the customized intention template determination model can be trained. The input of the customized intention explanation template determining model is word vectors of positive example training data and negative example training data; the probability of each training data belonging to each grammar template in the customized intent is output.
The customized intention template determination model and the customized intention template determination model can adopt a classification model commonly used in pattern recognition, such as a deep neural network model, a convolutional neural network model, a support vector machine model and the like, and a specific training method is the same as that of the prior art and is not described in detail herein.
When an explanation template sequencing model is constructed, a customized intention explanation template determining model and a customized intention explanation template determining model need to be marked for training, training data output by the model belongs to the explanation template probability with the highest priority in the probability of each explanation template in the skill intention, and the priorities of other probabilities are not limited; and taking the probability of each description template in the current customized skill intention of each training data as the input of the sequencing model, and outputting the probability as the sequenced probability, wherein the probability of the description template with the highest priority is sequenced at the top, and the sequence of the probabilities of other description templates is not limited.
The utterance template ordering model can adopt a neural network model, such as a convolutional neural network model, and the specific training method is the same as that of the prior art and is not described in detail herein.
Step 502, using the customization intention explanation template determination model and the customized intention explanation template determination model, determining the probability that the request data belongs to each explanation template in the customization intention and the probability of each explanation template in the customized intention.
Specifically, the word vectors of the request data are directly used as the input of the customizing intention template determining model and the customized intention template determining model respectively, and the probability of each meaning template of the request data belonging to each intention in the customizing intentions and the probability of each meaning template of each intention in the customized intentions are obtained according to the model output respectively.
Step 503, inputting the probability that the request data respectively belongs to each saying template in the customizing intention and the probability of each saying template in the customized intention into the saying template sequencing model, and obtaining the sequenced saying template probability according to the output of the saying template sequencing model.
Step 504, selecting the utterance template corresponding to the utterance template probability ranked most forward as the corresponding utterance template.
According to the semantic understanding method based on the customization skills, provided by the embodiment of the invention, after user request data is received, a path of the current customization skills is obtained, all semantic resources of the customization skills are obtained according to the path, a statement template of the request data belonging to the current customization skills is determined, a semantic slot and a corresponding entity in the request data are determined according to the statement template, and a semantic understanding result is obtained. By using the invention, the semantic understanding of the user request data can be realized quickly and accurately.
Correspondingly, the embodiment of the invention also provides a semantic understanding system based on the customized skills, which is a structural schematic diagram of the system as shown in fig. 6.
In this embodiment, the system includes:
a receiving module 601, configured to receive user request data;
a resource obtaining module 602, configured to obtain a path of a current customization skill, and obtain a semantic resource of the current customization skill according to the path;
a description template determining module 603, configured to determine that the request data belongs to a description template of a current customized skill, where the description template includes one or more semantic slots, and each semantic slot corresponds to an entity;
and a semantic understanding module 604, configured to determine a semantic slot and a corresponding entity in the request data according to the utterance template, and obtain a semantic understanding result.
Since each skill may include multiple intents, each intent includes multiple grammar templates. Therefore, the above-mentioned utterance template determination module 603 may have various implementations when determining that the requested data belongs to an utterance template of a current customized skill. This will be described in detail below.
In one embodiment of the utterance template determination module 603, it may include an intent determination module and a template matching module. Wherein the intent determination module is to determine an intent that the requested data pertains to a current customization skill; and the template matching module is used for matching the request data with the explanation templates corresponding to the intentions to obtain the explanation templates corresponding to the request data. In this embodiment, it is necessary to determine, by the intent determination module, an intent of the request data belonging to the current customized skill, and then match, by the template matching module, the request data with each utterance template corresponding to the intent to obtain an utterance template corresponding to the request data.
It should be noted that, in practical applications, the intention determining module may determine the intention of the requested data belonging to the current customized skill according to a pre-constructed intention determining model. Since the customization skills are skills developed by the application developer on the basis of the customized skills, the current customization skills include not only the customization skills but also the customized skills. Accordingly, current intent to customize skills includes: a customization intent and a customized intent; the customization intention refers to an intention which is newly added when the skill is customized currently and an intention obtained by modifying the customized skill intention; the customized intent refers to an unmodified intent of the base skill intents used in the current customization of the skill.
Accordingly, the intention determination model may be one model trained based on the utterance template of the customized intention and the utterance template of the customized intention, or may be two models trained based on the utterance template of the customized intention and the utterance template of the customized intention, respectively.
For example, one embodiment of the system further comprises: a first model building module (not shown) for training the skill intention determination model in advance according to the utterance template of the intention of the current customized skill (i.e. the utterance template of the customized intention and the utterance template of the customized intention); and constructing an intention ordering model by using the intention probability output during the training of the skill intention determining model. The training mode of the model has been described in detail above, and is not described in detail herein.
Accordingly, one specific structure of the intent determination model includes the following units:
a first intent probability determination unit for determining a probability that the request data belongs to each of the customized skill intentions using the skill intent determination model;
the first sequencing unit is used for inputting the probability that the request data belongs to each intention in the customized skill intentions into the intention sequencing model and obtaining the sequenced intention probability according to the output of the intention sequencing model;
and the first selection unit is used for selecting the intention corresponding to the intention probability ranked at the top as the intention of the request data.
For another example, an embodiment of the system further comprises: a second model building module (not shown) for training the customizing intention determining model and the customized intention determining model in advance based on the explanation template of the customizing intention and the explanation template of the customized intention; and constructing an intention ordering model by using the customization intention determining model and the intention probability output by the customization intention determining model during training. The training mode of the model has been described in detail above, and is not described in detail herein.
Accordingly, one specific structure of the intent determination model includes the following units:
a second intention probability determination unit for determining a probability that the request data belongs to each of the customizing intentions and a probability of each of the customized intentions using the customizing intention determination model and the customized intention determination model;
the second sequencing unit is used for inputting the probability that the request data respectively belongs to each intention in the customizing intentions and the probability of each intention in the customized intentions into the intention sequencing model and obtaining the sequenced intention probability according to the output of the intention sequencing model;
and the second selection unit is used for selecting the intention corresponding to the intention probability ranked furthest ahead as the intention of the request data.
In another embodiment of the utterance template determination module 603, an utterance template determination model may be constructed in advance, from which it is directly determined which utterance template the requested data belongs to in the current customization skills.
Also, since the customization skills are skills that the application developer develops on the basis of the customized skills, the current customization skills include not only the customization skills but also the customized skills. Accordingly, the utterance template determination model may be one model trained based on the utterance template for the customized intention and the utterance template for the customized intention, or may be two models trained based on the utterance template for the customized intention and the utterance template for the customized intention, respectively.
For example, one embodiment of the system further comprises: and a third model building module (not shown) for training the skill description template determination model in advance according to the description template of the current skill customization intention, and building a description template ordering model by using the description template probability output during model training.
Accordingly, a specific structure of the utterance template determination module 603 includes the following units:
a third probability determination unit, configured to determine, using the skill utterance template determination model, a probability that the request data belongs to each utterance template in the intent to customize skills;
the third sequencing unit is used for inputting the probability of each statement template in the intention that the request data belongs to the customized skill into the statement template sequencing model and obtaining the probability of the sequenced statement templates according to the output of the statement template sequencing model;
and the third selecting unit is used for selecting the description template corresponding to the description template probability which is ranked most front as the description template corresponding to the request data.
For another example, an embodiment of the system further comprises: a fourth model building module (not shown) for training the customizing intention utterance template determination model and the customized intention utterance template determination model in advance based on the customizing intention utterance template and the customized intention utterance template; and constructing a description template ordering model by using the customization intention description template determining model and the description template probability output when the customized intention description template determining model is trained.
Accordingly, a specific structure of the utterance template determination module 603 includes the following units:
a fourth probability determination unit for determining a probability that the request data belongs to each of the utterance templates in the customization intention and a probability of each of the utterance templates in the customized intention by using the customization intention utterance template determination model and the customized intention utterance template determination model;
the fourth ordering unit is used for inputting the probability that the request data respectively belongs to each description template in the customizing intention and the probability of each description template in the customized intention into the description template ordering model and obtaining the ordered description template probability according to the output of the description template ordering model;
and the fourth selection unit is used for selecting the description template corresponding to the description template probability which is ranked most front as the corresponding description template.
According to the semantic understanding system based on the customization skills, provided by the embodiment of the invention, after user request data is received, a path of the current customization skills is obtained, all semantic resources of the customization skills are obtained according to the path, a statement template of the request data belonging to the current customization skills is determined, a semantic slot and a corresponding entity in the request data are determined according to the statement template, and a semantic understanding result is obtained. By using the invention, the semantic understanding of the user request data can be realized quickly and accurately.
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. Furthermore, the above-described system embodiments are merely illustrative, wherein the modules illustrated as separate components may or may not be physically separate, may be located in one place, or may be distributed over multiple network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
The above embodiments of the present invention have been described in detail, and the present invention has been described herein with reference to particular embodiments, but the above embodiments are merely intended to facilitate an understanding of the methods and apparatuses of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.

Claims (12)

1. A semantic understanding method based on customization technology is characterized by comprising the following steps:
receiving user request data;
acquiring a path of a current customization skill, and acquiring semantic resources of the current customization skill according to the path;
determining that the requested data belongs to a template of a current customized skill, comprising: determining the intention of the request data to the current customization skill, and determining the description template of the request data according to the description templates contained in the intention; the statement template comprises one or more semantic slots, and each semantic slot corresponds to one entity;
and determining a semantic slot and a corresponding entity in the request data according to the saying template to obtain a semantic understanding result.
2. The method of claim 1, wherein said determining that the requested data belongs to a current custom skill profile specifically comprises:
and matching the request data with the explanation templates corresponding to the intentions to obtain the explanation templates corresponding to the request data.
3. The method of claim 2, further comprising:
training a skill intention determination model in advance according to a statement template of the current custom skill intention; an intention ranking model is constructed by using the intention probability output during the training of the skill intention determining model;
the determining the intent that the requested data belongs to the current custom skill comprises:
determining, using the skill intent determination model, a probability that the requested data belongs to each of the customized skill intents;
inputting the probability that the request data belongs to each intention in the customized skill intentions into the intention ranking model, and obtaining the ranked intention probability according to the output of the intention ranking model;
and selecting the intention corresponding to the intention probability with the highest ranking as the intention of the request data.
4. The method of claim 2, wherein the intent of the current custom skill comprises: a customization intent and a customized intent; the customization intention refers to an intention which is newly added when the skill is customized currently and an intention obtained by modifying the customized skill intention; the customized intent refers to an unmodified intent of the base skill intents used in the current customization of the skill;
the method further comprises the following steps:
training a customizing intention determining model and a customized intention determining model in advance according to the saying template of the customizing intention and the saying template of the customized intention; an intention ordering model is constructed by utilizing the customization intention determining model and the intention probability output when the customized intention determining model is trained;
the determining the intent that the requested data belongs to the current custom skill comprises:
determining a probability that the request data belongs to each of the customization intents and a probability that each of the customized intents using the customization intent determination model and customized intent determination model;
inputting the probability that the request data respectively belong to each intention in the customizing intentions and the probability of each intention in the customized intentions into the intention sequencing model, and obtaining the sequenced intention probability according to the output of the intention sequencing model;
and selecting the intention corresponding to the intention probability with the highest ranking as the intention of the request data.
5. The method of claim 1, further comprising:
training a skill description template to determine a model according to a description template of the current skill customization intention, and constructing a description template ordering model by using the description template probability output during model training;
the determining that the requested data belongs to a statement template of a current customized skill comprises:
determining a model by using the skill utterance template, and determining the probability that the request data belongs to each utterance template in the intention of customizing skills;
inputting the probability of each statement template in the intention that the request data belongs to the customized skills into the statement template sequencing model, and obtaining the sequenced statement template probability according to the output of the statement template sequencing model;
and selecting the description template corresponding to the description template probability with the top ranking as the description template corresponding to the request data.
6. The method of claim 1, wherein the intent to currently customize skills comprises: a customization intent and a customized intent; the customization intention refers to an intention which is newly added when the skill is customized currently and an intention obtained by modifying the customized skill intention; the customized intent refers to an unmodified intent of the base skill intents used in the current customization of the skill;
the method further comprises the following steps:
training a customizing intention explanation template determining model and a customized intention explanation template determining model in advance according to an explanation template of a customizing intention and an explanation template of a customized intention; constructing a description template ordering model by using the customization intention description template determining model and the description template probability output during the training of the customized intention description template determining model;
the determining that the requested data belongs to a statement template of a current customized skill comprises:
determining a probability that the request data belongs to each utterance template in the customization intent and a probability of each utterance template in the customized intent using the customization intent utterance template determination model and the customized intent utterance template determination model;
inputting the probability that the request data respectively belongs to each description template in the customization intention and the probability of each description template in the customized intention into the description template sequencing model, and obtaining the sequenced description template probability according to the output of the description template sequencing model;
and selecting the description template corresponding to the description template probability with the top ranking as the corresponding description template.
7. A semantic understanding system based on customized skills, comprising:
the receiving module is used for receiving user request data;
the resource acquisition module is used for acquiring a path of the current customization skill and acquiring semantic resources of the current customization skill according to the path;
a statement template determining module, configured to determine that the requested data belongs to a statement template of a current customized skill, including: determining the intention of the request data to the current customization skill, and determining the description template of the request data according to the description templates contained in the intention; the statement template comprises one or more semantic slots, and each semantic slot corresponds to one entity;
and the semantic understanding module is used for determining a semantic slot and a corresponding entity in the request data according to the saying template to obtain a semantic understanding result.
8. The system of claim 7, wherein the utterance template determination module comprises:
an intent determination module to determine an intent that the requested data pertains to a current customization skill;
and the template matching module is used for matching the request data with the explanation templates corresponding to the intentions to obtain the explanation templates corresponding to the request data.
9. The system of claim 8, further comprising:
the first model building module is used for training a skill intention determination model in advance according to a description template of the intention of the current customized skill; an intention ranking model is constructed by using the intention probability output during the training of the skill intention determining model;
the intent determination module includes:
a first intent probability determination unit for determining a probability that the request data belongs to each of the customized skill intentions using the skill intent determination model;
the first sequencing unit is used for inputting the probability that the request data belongs to each intention in the customized skill intentions into the intention sequencing model and obtaining the sequenced intention probability according to the output of the intention sequencing model;
and the first selection unit is used for selecting the intention corresponding to the intention probability ranked at the top as the intention of the request data.
10. The system of claim 8, wherein the current intent to customize skills comprises: a customization intent and a customized intent; the customization intention refers to an intention which is newly added when the skill is customized currently and an intention obtained by modifying the customized skill intention; the customized intent refers to an unmodified intent of the base skill intents used in the current customization of the skill;
the system further comprises:
the second model building module is used for training the customizing intention determining model and the customized intention determining model in advance according to the saying template of the customizing intention and the saying template of the customized intention; an intention ordering model is constructed by utilizing the customization intention determining model and the intention probability output when the customized intention determining model is trained;
the intent determination module includes:
a second intention probability determination unit for determining a probability that the request data belongs to each of the customizing intentions and a probability of each of the customized intentions using the customizing intention determination model and the customized intention determination model;
the second sequencing unit is used for inputting the probability that the request data respectively belongs to each intention in the customizing intentions and the probability of each intention in the customized intentions into the intention sequencing model and obtaining the sequenced intention probability according to the output of the intention sequencing model;
and the second selection unit is used for selecting the intention corresponding to the intention probability ranked furthest ahead as the intention of the request data.
11. The system of claim 7, further comprising:
the third model building module is used for training the skill description template to determine a model in advance according to the description template of the current skill customization intention, and building a description template ordering model by using the description template probability output during model training;
the utterance template determination module includes:
a third probability determination unit, configured to determine, using the skill utterance template determination model, a probability that the request data belongs to each utterance template in the intent to customize skills;
the third sequencing unit is used for inputting the probability of each statement template in the intention that the request data belongs to the customized skill into the statement template sequencing model and obtaining the probability of the sequenced statement templates according to the output of the statement template sequencing model;
and the third selecting unit is used for selecting the description template corresponding to the description template probability which is ranked most front as the description template corresponding to the request data.
12. The system of claim 7, wherein the current intent to customize skills comprises: a customization intent and a customized intent; the customization intention refers to an intention which is newly added when the skill is customized currently and an intention obtained by modifying the customized skill intention; the customized intent refers to an unmodified intent of the base skill intents used in the current customization of the skill;
the system further comprises:
the fourth model building module is used for training the customizing intention explanation template determining model and the customized intention explanation template determining model in advance according to the explanation template of the customizing intention and the explanation template of the customized intention; constructing a description template ordering model by using the customization intention description template determining model and the description template probability output during the training of the customized intention description template determining model;
the utterance template determination module includes:
a fourth probability determination unit for determining a probability that the request data belongs to each of the utterance templates in the customization intention and a probability of each of the utterance templates in the customized intention by using the customization intention utterance template determination model and the customized intention utterance template determination model;
the fourth ordering unit is used for inputting the probability that the request data respectively belongs to each description template in the customizing intention and the probability of each description template in the customized intention into the description template ordering model and obtaining the ordered description template probability according to the output of the description template ordering model;
and the fourth selection unit is used for selecting the description template corresponding to the description template probability which is ranked most front as the corresponding description template.
CN201710774597.9A 2017-08-31 2017-08-31 Semantic understanding method and system based on customization technology Active CN107515857B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710774597.9A CN107515857B (en) 2017-08-31 2017-08-31 Semantic understanding method and system based on customization technology

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710774597.9A CN107515857B (en) 2017-08-31 2017-08-31 Semantic understanding method and system based on customization technology

Publications (2)

Publication Number Publication Date
CN107515857A CN107515857A (en) 2017-12-26
CN107515857B true CN107515857B (en) 2020-08-18

Family

ID=60724962

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710774597.9A Active CN107515857B (en) 2017-08-31 2017-08-31 Semantic understanding method and system based on customization technology

Country Status (1)

Country Link
CN (1) CN107515857B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108563631A (en) * 2018-03-23 2018-09-21 江苏速度信息科技股份有限公司 A kind of automatic identifying method of natural language address descriptor
CN108984157A (en) * 2018-07-27 2018-12-11 苏州思必驰信息科技有限公司 Technical ability configuration and call method and system for voice dialogue platform
CN110334201B (en) * 2019-07-18 2021-09-21 中国工商银行股份有限公司 Intention identification method, device and system
US20210082400A1 (en) * 2019-09-16 2021-03-18 Oracle International Corporation Stop word data augmentation for natural language processing

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102622417A (en) * 2012-02-20 2012-08-01 北京搜狗信息服务有限公司 Method and device for ordering information records
CN102880649A (en) * 2012-08-27 2013-01-16 北京搜狗信息服务有限公司 Individualized information processing method and system
CN104077407A (en) * 2014-07-10 2014-10-01 中国工商银行股份有限公司 System and method for intelligent data searching
CN104199810A (en) * 2014-08-29 2014-12-10 科大讯飞股份有限公司 Intelligent service method and system based on natural language interaction
US20160004766A1 (en) * 2006-10-10 2016-01-07 Abbyy Infopoisk Llc Search technology using synonims and paraphrasing
CN105487663A (en) * 2015-11-30 2016-04-13 北京光年无限科技有限公司 Intelligent robot oriented intention identification method and system
CN106021339A (en) * 2016-05-09 2016-10-12 中国联合网络通信集团有限公司 A semantic query method and system for a resource tree
CN107015962A (en) * 2017-03-16 2017-08-04 北京光年无限科技有限公司 Towards the implementation method and device of the self-defined intention assessment of intelligent robot

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102262622A (en) * 2010-05-31 2011-11-30 国际商业机器公司 Document processing, template generating and conceptbase generating methods and devices
CN104360994A (en) * 2014-12-04 2015-02-18 科大讯飞股份有限公司 Natural language understanding method and natural language understanding system

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160004766A1 (en) * 2006-10-10 2016-01-07 Abbyy Infopoisk Llc Search technology using synonims and paraphrasing
CN102622417A (en) * 2012-02-20 2012-08-01 北京搜狗信息服务有限公司 Method and device for ordering information records
CN102880649A (en) * 2012-08-27 2013-01-16 北京搜狗信息服务有限公司 Individualized information processing method and system
CN104077407A (en) * 2014-07-10 2014-10-01 中国工商银行股份有限公司 System and method for intelligent data searching
CN104199810A (en) * 2014-08-29 2014-12-10 科大讯飞股份有限公司 Intelligent service method and system based on natural language interaction
CN105487663A (en) * 2015-11-30 2016-04-13 北京光年无限科技有限公司 Intelligent robot oriented intention identification method and system
CN106021339A (en) * 2016-05-09 2016-10-12 中国联合网络通信集团有限公司 A semantic query method and system for a resource tree
CN107015962A (en) * 2017-03-16 2017-08-04 北京光年无限科技有限公司 Towards the implementation method and device of the self-defined intention assessment of intelligent robot

Also Published As

Publication number Publication date
CN107515857A (en) 2017-12-26

Similar Documents

Publication Publication Date Title
CN107515857B (en) Semantic understanding method and system based on customization technology
CN105654950B (en) Adaptive voice feedback method and device
WO2016004763A1 (en) Service recommendation method and device having intelligent assistant
US20160259767A1 (en) Annotations in software applications for invoking dialog system functions
CN106098063B (en) Voice control method, terminal device and server
US20190081914A1 (en) Method and apparatus for generating candidate reply message
US20170368683A1 (en) User portrait based skill package recommendation device and method
CN109840089A (en) The system and method for carrying out visual analysis and programming for the session proxy to electronic equipment
US10558655B2 (en) Data query method supporting natural language, open platform, and user terminal
CN107977238A (en) Using startup method and device
CN106649825B (en) Voice interaction system and creation method and device thereof
CN107656996B (en) Man-machine interaction method and device based on artificial intelligence
US20170249934A1 (en) Electronic device and method for operating the same
JP2019008771A (en) Information input method
US20140101144A1 (en) Methods and Systems for Displaying Microblog Topics
CN111695345A (en) Method and device for recognizing entity in text
WO2014117490A1 (en) Method and device for recommending video from video library
US10115391B2 (en) Method and apparatus for providing voice feedback information to user in call
CN107450725B (en) Man-machine interaction application platform, method and storage medium
CN108334353B (en) Skill development system and method
JP2020004382A (en) Method and device for voice interaction
CN111090728A (en) Conversation state tracking method and device and computing equipment
CN107608660B (en) Shared skill application method and system
CN111261151A (en) Voice processing method and device, electronic equipment and storage medium
CN108831473B (en) Audio processing method and device

Legal Events

Date Code Title Description
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