WO2024036694A1 - 控制指令的发送方法、控制装置、存储介质及电子装置 - Google Patents

控制指令的发送方法、控制装置、存储介质及电子装置 Download PDF

Info

Publication number
WO2024036694A1
WO2024036694A1 PCT/CN2022/120312 CN2022120312W WO2024036694A1 WO 2024036694 A1 WO2024036694 A1 WO 2024036694A1 CN 2022120312 W CN2022120312 W CN 2022120312W WO 2024036694 A1 WO2024036694 A1 WO 2024036694A1
Authority
WO
WIPO (PCT)
Prior art keywords
intention
intent
query vector
voice
target
Prior art date
Application number
PCT/CN2022/120312
Other languages
English (en)
French (fr)
Inventor
刘建国
孙凯
张旭
区波
刘朝振
李岩
张向磊
Original Assignee
青岛海尔科技有限公司
海尔智家股份有限公司
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 青岛海尔科技有限公司, 海尔智家股份有限公司 filed Critical 青岛海尔科技有限公司
Publication of WO2024036694A1 publication Critical patent/WO2024036694A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B15/00—Systems controlled by a computer
    • G05B15/02—Systems controlled by a computer electric
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B19/00—Programme-control systems
    • G05B19/02—Programme-control systems electric
    • G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/20—Pc systems
    • G05B2219/26—Pc applications
    • G05B2219/2642—Domotique, domestic, home control, automation, smart house

Definitions

  • the present disclosure relates to the field of smart home technology, and specifically to a method for sending control instructions, a control device, a storage medium and an electronic device.
  • the main purpose of the present disclosure is to provide a control instruction sending method, a control device, a storage medium and an electronic device to solve the technical problem in the prior art that the control intention of voice control cannot be effectively recognized.
  • a method for sending control instructions including: receiving a voice request sent by a target object, and determining the intent query vector corresponding to the voice request; in a pre-constructed intent knowledge graph Determine the voice intent corresponding to the intent query vector, where the intent knowledge graph is at least used to indicate the association between target control instructions of multiple working states and the voice intent; the intent query vector is not found in the intent knowledge graph.
  • the intention query vector is identified through the target neural network model to determine the target control instruction corresponding to the intention query vector; the target control instruction is sent to the target home appliance device used to execute the target control instruction.
  • a control device including: a voice recognition module configured to obtain voice information and convert the voice information into an intent query vector; an intent recognition module configured to store intent knowledge in the intent recognition module In the graph and target neural network model, the intent recognition module is configured to perform intent recognition on the intent query vector to obtain corresponding control instructions; the control instruction sending module is configured to send control instructions to the target home appliance.
  • a control device including: a voice recognition module configured to receive a voice request sent by a target object and determine an intent query vector corresponding to the voice request; a first intent recognition module, Set to determine the voice intention corresponding to the intention query vector in a pre-constructed intention knowledge graph, wherein the intention knowledge graph is at least used to indicate the association between the target control instructions of multiple working states and the voice intention;
  • the second intent recognition module is configured to perform intent recognition on the intent query vector through a target neural network model to determine the intent query vector when the voice intent corresponding to the intent query vector is not found in the intent knowledge graph.
  • the target control instruction corresponding to the intention query vector; the control instruction sending module is configured to send the target control instruction to the target home appliance for executing the target control instruction.
  • a computer-readable storage medium includes a stored program, wherein the method provided above is executed when the program is run.
  • an electronic device including a memory and a processor.
  • a computer program is stored in the memory, and the processor is configured to execute the method provided above through the computer program.
  • the intent query vector corresponding to the voice request after determining the intent query vector corresponding to the voice request, first determine the voice intent corresponding to the intent query vector in the pre-constructed intent knowledge graph.
  • the voice intention corresponding to the intention query vector can be found in the pre-built intention knowledge graph, the corresponding working status control instruction is determined according to the voice intention; when the voice intention corresponding to the intention query vector is not found in the intention knowledge graph.
  • the intention query vector is identified through the target neural network model, and the target control instruction corresponding to the intention query vector is determined through intention recognition. After the corresponding target control instruction is determined, the target control instruction is sent to the target home appliance that executes the target control instruction, so that the target home appliance executes the target control instruction.
  • Figure 1 is an overall control flow chart of a method for sending control instructions of an intelligent device according to an embodiment of the present disclosure
  • Figure 2 is a main control flow chart of a method for sending control instructions of an intelligent device according to an embodiment of the present disclosure
  • Figure 3 shows the flow of a voice recognition method in a method for sending control instructions for an intelligent device according to an embodiment of the present disclosure
  • Figure 4 is a structural block diagram of a control device for an intelligent device according to an embodiment of the present disclosure
  • Figure 5 is a schematic diagram of the hardware environment of an interaction method for smart devices according to an embodiment of the present disclosure.
  • Embodiment 1 of the present disclosure provides a method for sending control instructions.
  • the method of sending the control instruction includes: receiving the voice request sent by the target object, and determining the intent query vector corresponding to the voice request; determining the voice intent corresponding to the intent query vector in the pre-constructed intent knowledge graph, wherein the intent knowledge graph is at least The association between target control instructions used to indicate multiple working states and voice intentions; when the voice intention corresponding to the intention query vector is not found in the intention knowledge graph, the intention query vector is analyzed through the target neural network model Identify to determine the target control instruction corresponding to the intent query vector; send the target control instruction to the target home appliance used to execute the target control instruction.
  • the intention knowledge graph and the target neural network are combined to realize the intention recognition of the voice request sent by the target object (the target object can be the user) and obtain the corresponding target control instructions, which can facilitate and effectively identify the user's voice.
  • the control intention of the control is used to facilitate sending target control instructions to the target home appliance used to execute the target control command, thereby facilitating accurate control of the work of the corresponding target home appliance and avoiding device conflicts when the user controls through voice.
  • This method of sending control instructions is widely used in whole-house intelligent digital control application scenarios such as Smart Home, Smart Home, Smart Home Equipment Ecology, and Intelligence House Ecology.
  • the above method can be applied to a hardware environment composed of a terminal device 102 and a server 104 as shown in FIG. 5 .
  • the server 104 is connected to the terminal device 102 through the network and can be used to provide services (such as application services, etc.) for the terminal or the client installed on the terminal.
  • a database can be set up on the server or independently from the server to provide services for the terminal.
  • the server 104 provides data storage services, and cloud computing and/or edge computing services can be configured on the server or independently of the server to provide data computing services for the server 104 .
  • the above-mentioned network may include but is not limited to at least one of the following: wired network, wireless network.
  • the above-mentioned wired network may include but is not limited to at least one of the following: wide area network, metropolitan area network, and local area network.
  • the above-mentioned wireless network may include at least one of the following: WIFI (Wireless Fidelity, Wireless Fidelity), Bluetooth.
  • the terminal device 102 may be, but is not limited to, a PC, a mobile phone, a tablet, a smart air conditioner, a smart hood, a smart refrigerator, a smart oven, a smart stove, a smart washing machine, a smart water heater, a smart washing equipment, a smart dishwasher, or a smart projection device.
  • smart TV smart clothes drying rack, smart curtains, smart audio and video, smart sockets, smart audio, smart speakers, smart fresh air equipment, smart kitchen and bathroom equipment, smart bathroom equipment, smart sweeping robot, smart window cleaning robot, smart mopping robot, Smart air purification equipment, smart steamers, smart microwave ovens, smart kitchen appliances, smart purifiers, smart water dispensers, smart door locks, etc.
  • the voice intent corresponding to the intent query vector is first determined from the pre-constructed intent knowledge graph.
  • the voice intention corresponding to the intention query vector can be found in the pre-built intention knowledge graph, the corresponding working status control instruction is determined according to the voice intention; when the voice intention corresponding to the intention query vector is not found in the intention knowledge graph.
  • the intention query vector is identified through the target neural network model, and the target control instruction corresponding to the intention query vector is determined through intention recognition. After the corresponding target control instruction is determined, the target control instruction is sent to the target home appliance that executes the target control instruction, so that the target home appliance executes the target control instruction.
  • using the target neural network model to perform intent recognition on the intent query vector includes: using the target neural network model to generalize the intent query vector; and performing intent recognition on the intent query vector based on the results of the generalization processing. Adopting the above method can facilitate rapid intent recognition of the intent query vector and improve the response speed to the user's voice request.
  • generalization processing in using the target neural network model to generalize the intent query vector means that different language expressions can be used to control the same state control instructions, which can specifically include the extraction of keywords, etc. to achieve generalization processing.
  • the intent recognition of the intent query vector is performed based on the results of the generalization processing, including: comparing the processing results of the generalization processing with the speech intentions stored in the pre-constructed intent knowledge graph; When there is a speech intention that matches the processing result of the generalization process in the intent knowledge graph, the speech intent that matches the intent query vector is determined as the recognition result of the intent recognition of the intent query vector; in the pre-built intent knowledge graph When no speech intention matching the processing result of the generalization processing is found, the target neural network is used to perform intent inference on the intent query vector, and the intent query vector is identified based on the speech intent obtained by the intent inference.
  • the method further includes: determining whether the voice intention derived from the intention inference can be executed; when it is determined that the voice intention derived from the intention inference can be executed, The speech intention obtained by intention inference is stored in the pre-built intention knowledge graph; when it is determined that the speech intention obtained by intention inference cannot be executed, the intention recognition of the intention query vector is exited.
  • the voice intention derived from intention inference into the pre-built intention knowledge graph, it is possible to quickly obtain the corresponding voice intention when the same voice request is obtained on the next side, and to improve the operation response speed.
  • the intent information stored in the intent knowledge graph includes intent nodes
  • determining the voice intent corresponding to the intent query vector in the pre-constructed intent knowledge graph includes: a pre-sequence action of obtaining the intent query vector, where, The pre-order action refers to the execution action before receiving the voice request; the fused pre-order action of the intent query vector is matched with the intent node stored in the intent knowledge graph to obtain the most matching intent node; the intent query is determined based on the best matching intent node
  • the vector corresponds to the speech intent.
  • determining the intent query vector corresponding to the voice request includes: generating the intent query vector according to the received voice request combined with environmental variables where the target home appliance is currently located. Adopting such a method can further improve the accuracy of intention query vector generation, so as to more accurately determine the corresponding target control instruction, and thereby facilitate accurate control of the target home appliance to execute the target control instruction.
  • the "current target environment variable” may include information such as temperature, humidity, and user ID of the current environment.
  • the pre-built intention knowledge graph includes multiple personal intention knowledge graphs; among them, multiple personal intention knowledge graphs correspond to different user voice intentions; the voice corresponding to the intention query vector is determined in the pre-built intention knowledge graph Intentions include: obtaining the speech characteristics of the target object from the voice request sent by the target object; determining the target personal intention knowledge graph corresponding to the target object from multiple personal intention knowledge graphs based on the speech characteristics; determining the target personal intention knowledge graph corresponding to the target object based on the speech characteristics.
  • the speech intent corresponding to the intent query vector can facilitate the setting of different personal intention knowledge maps according to different target objects (users). Each target object corresponds to a different personal intention knowledge map.
  • the target personal intention knowledge corresponding to the target object is determined based on the voice characteristics.
  • the graph determines the corresponding speech intention.
  • the GCN graph neural network is used, which is theoretically faster in handling inference problems than models based on deep networks such as ordinary deep learning networks CNN or LSTM. Because traditional deep learning networks such as CNN/LSTM are good at processing data in Euclidean space.
  • the inference data uses a graph data structure. If the traditional deep learning network is used, the calculation and training time will be long, which will affect the user experience.
  • This solution uses a solution that combines the knowledge graph database with the model. In this way, if the inferred edge exists in the knowledge graph database, it will directly trigger the return result. If not, the model will be used for reasoning, and the generated edge will be stored in the knowledge graph database. , so that the next request can be responded to quickly.
  • the model in this embodiment includes a query interface, a knowledge graph database, a trained graph neural learning model (GCN), an intent execution module, and a new intent embedding module.
  • GCN graph neural learning model
  • an intention query vector is generated based on the environment variables at that time and enters the intention recognition module.
  • the knowledge graph database is queried. There are various intention nodes in the knowledge graph database, and the query is performed by integrating the pre-order actions. If there is no pre-order action, the most matching node of the current vector is directly queried. If the intention is hit, the command will be issued directly. If there is no hit, it will be transferred to the already trained GCN deep network model for judgment.
  • the GCN deep learning network can generalize the query vector and match the existing intent nodes in the knowledge graph database, then the knowledge graph database is queried and the corresponding action is performed. If it cannot hit, but it can be determined by the model that it is a certain intention, then the intention is stored in the knowledge graph database (this action does not need to be executed online, but is executed regularly). If the intention is executable and the action is risk-free, directly connect to the action execution module for action execution. Otherwise the action will not be executed.
  • the training principle of GCN deep learning model is as follows:
  • Each paragraph (that is, a certain paragraph said by a user, which is finally implemented as a certain query) can be expressed as u1, u2,... uN.
  • the ultimate goal is to convert every word spoken by the user into a specific executable intention.
  • Each query can be turned into a directed graph according to the order:
  • Each query needs to determine the intention u i , which contains environmental attributes [w1, w2, w3,...], w is the node attribute, which can be any obtainable user attribute value, such as gender, age, and the room in which the query is made. , weather, temperature, humidity, etc. at that time.
  • V is the attribute of the node
  • E is the relationship between intentions
  • R is the attribute of the intention relationship
  • W is the weight between the relationships.
  • A is the connection matrix, used to represent the connection between the intention node and the intention node.
  • the number of nodes is n
  • A is an n ⁇ n matrix
  • H 0 X
  • X is the characteristic matrix of the node.
  • the feature dimension is k
  • X is an nxk matrix.
  • the modeling process in this embodiment is: drawing the real state transition diagram of the device according to the specific model and design base plate.
  • Each model of machine has a unique state transition diagram, and the states are drawn according to the properties of the device itself.
  • each node has a relationship with other nodes, and the relationship also has attributes such as spring, summer, an action sequence, or a scene.
  • the node itself stores the attributes corresponding to when the intention occurs, such as the time, space, user gender, and age when the intention occurs.
  • the relationships between nodes indicate the sequence of events occurring between intentions in the graph.
  • the solution using GCN+ knowledge graph database can add new intentions online and identify previously unavailable intentions. There is currently no similar manufacturer that has done this solution. This saves computing resources, eliminates the need to re-debug the code, and makes iteration faster. Fast iteration means that user needs can be met faster.
  • the edge computing and cloud solution implements this patent: the actions and attribute relationships generated by each person are stored locally, that is, each person's personal knowledge graph is stored locally. This is used as edge computing to directly recommend to individual users.
  • the knowledge graph is regularly pushed to the cloud. After the push, everyone's habits are summarized into a super knowledge in the cloud, and cold start recommendations are made to other new users.
  • GCN graph deep convolution network can better model graph data.
  • the sequence calculated using the correlation algorithm can be directly put into the GCN for training without additional data processing. save time.
  • What GCN calculates is directly the prediction, unification, and regression of the graph data structure.
  • the generated nodes can either become nodes in the knowledge graph database, or can be combined with two components (model component + knowledge graph database component) for generalized intent recognition, that is, the words spoken in the user query are not stored in the knowledge graph database.
  • GCN is better than traditional deep learning networks such as CNN for tasks such as vector matching.
  • Embodiment 2 of the present disclosure provides a control device, including: a speech recognition module, an intention recognition module and a control instruction sending module, which acquires voice information and converts the voice information into an intention query vector.
  • the intent knowledge graph and the target neural network model are stored in the intent recognition module, and the intent recognition module is configured to perform intent recognition on the intent query vector to obtain the corresponding control instructions.
  • the control instruction sending module is configured to send control instructions to the target home appliance equipment.
  • Embodiment 3 of the present disclosure provides a control device, including: a voice recognition module configured to receive a voice request sent by a target object and determine an intent query vector corresponding to the voice request; first Intention identification module, configured to determine the voice intention corresponding to the intention query vector in a pre-constructed intention knowledge graph, wherein the intention knowledge graph is at least used to indicate the relationship between target control instructions of multiple working states and voice intentions.
  • the second intention recognition module is configured to perform intention recognition on the intention query vector through the target neural network model when the voice intention corresponding to the intention query vector is not found in the intention knowledge graph,
  • the control instruction sending module is configured to send the target control instruction to the target home appliance for executing the target control instruction.
  • Embodiment 4 of the present disclosure provides a computer-readable storage medium.
  • the computer-readable storage medium includes a stored program. When the program is run, the method provided above is executed.
  • Embodiment 5 of the present disclosure provides an electronic device.
  • the electronic device includes a memory and a processor.
  • a computer program is stored in the memory, and the processor is configured to execute the method provided above through the computer program.

Landscapes

  • Engineering & Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Manufacturing & Machinery (AREA)
  • Quality & Reliability (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Selective Calling Equipment (AREA)

Abstract

本公开提供了一种控制指令的发送方法、控制装置、存储介质及电子装置，涉及智能家居/智慧家庭技术领域，该方法包括：接收目标对象发送的语音请求，并确定与语音请求对应的意图查询向量；在预先构建的意图知识图谱中确定与意图查询向量对应的语音意图，意图知识图谱至少用于指示多个工作状态的目标控制指令与语音意图之间的关联关系；在意图知识图谱中未查找到意图查询向量对应的语音意图的情况下，通过目标神经网络模型对意图查询向量进行意图识别，以确定意图查询向量对应的目标控制指令；向用于执行目标控制指令的目标家电设备发送目标控制指令。

Description

控制指令的发送方法、控制装置、存储介质及电子装置
本公开要求于2022年08月18日提交中国专利局、申请号为202210993981.9、发明名称“控制指令的发送方法、控制装置、存储介质及电子装置”的中国专利申请的优先权，其全部内容通过引用结合在本公开中。
技术领域
本公开涉及智能家居技术领域，具体而言，涉及一种控制指令的发送方法、控制装置、存储介质及电子装置。
背景技术
目前，随着智能家居普及率的提高，各大厂商都在智能家居的研发上投入了大量的力量。而智能家居一项基本的任务就是意图识别，识别了用户的意图才能正确服务好用户，提高用户体验，现有技术中使用语音控制智能家居的用户占到用户总数的92％。
然而，现在随着智能家电语音控制系统的普及，越来越多的智能家电支持语音控制，而用户使用家电时经常出现设备冲突的情况，如太冷了，太热了，由于家电有多个设备与用户绑定，故语音控制系统面临说同样一句话，具体设备无法有效识别对应控制意图的问题。
发明内容
本公开的主要目的在于提供一种控制指令的发送方法、控制装置、存储介质及电子装置，以解决现有技术中无法有效识别语音控制的控制意图的技术问题。
为了实现上述目的，根据本公开的一个方面，提供了一种控制指令的发送方法，包括：接收目标对象发送的语音请求，并确定与语音请求对应的意图查询向量；在预先构建的意图知识图谱中确定与意图查询向量对应的语音意图，其中，意图知识图谱至少用于指示多个工作状态的目标控制指令与语音意图之间的关 联关系；在意图知识图谱中未查找到意图查询向量对应的语音意图的情况下，通过目标神经网络模型对意图查询向量进行意图识别，以确定意图查询向量对应的目标控制指令；向用于执行目标控制指令的目标家电设备发送目标控制指令。
根据本公开的另一方面，提供了一种控制装置，包括：语音识别模块，设置为获取语音信息并将语音信息转化为意图查询向量；意图识别模块，设置为意图识别模块内存储有意图知识图谱和目标神经网络模型，意图识别模块设置为对意图查询向量进行意图识别以得到对应的控制指令；控制指令发送模块，设置为控制指令发送模块设置为目标家电设备发送控制指令。
根据本公开的又一方面，提供了一种控制装置，包括：语音识别模块，设置为接收目标对象发送的语音请求，并确定与所述语音请求对应的意图查询向量；第一意图识别模块，设置为在预先构建的意图知识图谱中确定与所述意图查询向量对应的语音意图，其中，所述意图知识图谱至少用于指示多个工作状态的目标控制指令与语音意图之间的关联关系；第二意图识别模块，设置为在所述意图知识图谱中未查找到所述意图查询向量对应的语音意图的情况下，通过目标神经网络模型对所述意图查询向量进行意图识别，以确定所述意图查询向量对应的目标控制指令；控制指令发送模块，设置为向用于执行所述目标控制指令的目标家电设备发送所述目标控制指令。
根据本公开的又一方面，提供了一种计算机可读的存储介质，计算机可读的存储介质包括存储的程序，其中，程序运行时执行上述提供的方法。
根据本公开的又一方面，提供了一种电子装置，包括存储器和处理器，存储器中存储有计算机程序，处理器被设置为通过计算机程序执行上述提供的方法。
应用本公开的技术方案，在确定与语音请求对应的意图查询向量后，先通过预先构建的意图知识图谱中确定与意图查询向量对应的语音意图。当能够在预先构建的意图知识图谱中查找到与意图查询向量对应的语音意图时，根据语音意图确定与之对应的工作状态控制指令；当在意图知识图谱中没有查找到与意图查询向量对应的语音意图时，通过目标神经网络模型对意图查询向量进行意图识别，并通过意图识别确定与该意图查询向量对应的目标控制指令。在确定对应的目标 控制指令后，向执行该目标控制指令的目标家电发送该目标控制指令，以使该目标家电执行该目标控制指令。
附图说明
此处所说明的附图用来提供对本公开的进一步理解，构成本公开的一部分，本公开的示意性实施例及其说明用于解释本公开，并不构成对本公开的不当限定。在附图中：
图1据本公开实施例的一种智能设备的控制指令的发送方法的整体控制流程图；
图2根据本公开实施例的一种智能设备的控制指令的发送方法的主要控制流程图；
图3根据本公开实施例的一种智能设备的控制指令的发送方法中的语音识别方法的流程；
图4根据本公开实施例的一种智能设备的控制装置的结构框图；
图5根据本公开实施例的一种智能设备的交互方法的硬件环境示意图。
具体实施方式
为了使本技术领域的人员更好地理解本公开方案，下面将结合本公开实施例中的附图，对本公开实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本公开一部分的实施例，而不是全部的实施例。基于本公开中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本公开保护的范围。
需要说明的是，本公开的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本公开的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或 单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
如图1和图2示，本公开的实施例一提供了一种控制指令的发送方法。控制指令的发送方法包括：接收目标对象发送的语音请求，并确定与语音请求对应的意图查询向量；在预先构建的意图知识图谱中确定与意图查询向量对应的语音意图，其中，意图知识图谱至少用于指示多个工作状态的目标控制指令与语音意图之间的关联关系；在意图知识图谱中未查找到意图查询向量对应的语音意图的情况下，通过目标神经网络模型对意图查询向量进行意图识别，以确定意图查询向量对应的目标控制指令；向用于执行目标控制指令的目标家电设备发送目标控制指令。
采用这样的方法，采用意图知识图谱和目标神经网络结合的方式实现对目标对象(目标对象可以为用户)发送的语音请求进行意图识别并获取对应的目标控制指令，能够便于有效地识别用户的语音控制的控制意图，从而便于向用于执行目标控制指令的目标家电设备发送目标控制指令，从而便于准确控制对应的目标家电的工作，避免用户通过语音控制时出现设备冲突的情况。
该控制指令的发送方法广泛应用于智慧家庭(Smart Home)、智能家居、智能家用设备生态、智慧住宅(Intelligence House)生态等全屋智能数字化控制应用场景。可选地，在本实施例中，上述方法可以应用于如图5示的由终端设备102和服务器104所构成的硬件环境中。如图5示，服务器104通过网络与终端设备102进行连接，可用于为终端或终端上安装的客户端提供服务(如应用服务等)，可在服务器上或独立于服务器设置数据库，用于为服务器104提供数据存储服务，可在服务器上或独立于服务器配置云计算和/或边缘计算服务，用于为服务器104提供数据运算服务。
上述网络可以包括但不限于以下至少之一：有线网络，无线网络。上述有线网络可以包括但不限于以下至少之一：广域网，城域网，局域网，上述无线网络可以包括但不限于以下至少之一：WIFI(Wireless Fidelity，无线保真)，蓝牙。 终端设备102可以并不限定于为PC、手机、平板电脑、智能空调、智能烟机、智能冰箱、智能烤箱、智能炉灶、智能洗衣机、智能热水器、智能洗涤设备、智能洗碗机、智能投影设备、智能电视、智能晾衣架、智能窗帘、智能影音、智能插座、智能音响、智能音箱、智能新风设备、智能厨卫设备、智能卫浴设备、智能扫地机器人、智能擦窗机器人、智能拖地机器人、智能空气净化设备、智能蒸箱、智能微波炉、智能厨宝、智能净化器、智能饮水机、智能门锁等。
具体地，采用本实施例提供的控制指令的发送方法，在确定与语音请求对应的意图查询向量后，先通过预先构建的意图知识图谱中确定与意图查询向量对应的语音意图。当能够在预先构建的意图知识图谱中查找到与意图查询向量对应的语音意图时，根据语音意图确定与之对应的工作状态控制指令；当在意图知识图谱中没有查找到与意图查询向量对应的语音意图时，通过目标神经网络模型对意图查询向量进行意图识别，并通过意图识别确定与该意图查询向量对应的目标控制指令。在确定对应的目标控制指令后，向执行该目标控制指令的目标家电发送该目标控制指令，以使该目标家电执行该目标控制指令。
在本实施例中，利用目标神经网络模型对意图查询向量进行意图识别，包括：利用目标神经网络模型对意图查询向量进行泛化处理；根据泛化处理的结果对意图查询向量进行意图识别。采用上述方法，能够便于快速对意图查询向量进行意图识别，提高对用户的语音请求的反应速度。
需要说明的是，利用目标神经网络模型对意图查询向量进行泛化处理中的“泛化处理”是指可以利用不同的语言表达方式来控制同一状态控制指令，具体可以包括通过关键词的提取等来实现泛化处理。
具体地，本实施例中根据泛化处理的结果对意图查询向量进行意图识别，包括：将泛化处理的处理结果与预先构建的意图知识图谱中存储的语音意图进行比对；在预先构建的意图知识图谱中存在与泛化处理的处理结果匹配的语音意图的情况下，将与意图查询向量匹配的语音意图确定为对意图查询向量进行意图识别的识别结果；在预先构建的意图知识图谱中未查找到与泛化处理的处理结果匹配的语音意图的情况下，利用目标神经网络对意图查询向量进行意图推断，并根据 意图推断得到的语音意图对意图查询向量进行意图识别。采用这样的方法，能够便于先根据泛化处理的结果快速获取意图知识图谱中是否有匹配的语音意图，提高操作反应速度。并在没有匹配结果的情况下，通过目标神经网络进行意图推断，提高意图识别的准确性，以便于准确判断用户的目标控制指令，并控制对应的目标设备执行该目标控制指令。
在本实施例中，利用目标神经网络对意图查询向量进行意图推断之后，方法还包括：确定意图推断得到的语音意图是否能够执行；当确定出意图推断得到的语音意图能够执行的情况下，将意图推断得到的语音意图存储至预先构建的意图知识图谱；在确定出意图推断得到的语音意图无法执行的情况下，退出对意图查询向量的意图识别。采用这样的方法，通过将意图推断得到的语音意图存储至预先构建的意图知识图谱中，能够便于在下一侧获取相同的语音请求时快速获取对应的语音意图，便于提高操作反应速度。
具体地，本实施例中意图知识图谱中存储的意图信息包括意图节点，在预先构建的意图知识图谱中确定与意图查询向量对应的语音意图，包括：获取意图查询向量的前序动作，其中，前序动作是指接收语音请求前的执行动作；将意图查询向量的融合前序动作与意图知识图谱中存储的意图节点进行匹配，以获取最匹配意图节点；根据最匹配意图节点确定与意图查询向量对应的语音意图。采用这样的方法，通过查询意图查询向量的前序动作，并将意图查询向量与前序动作进行融合，能够便于更准确地确定匹配出对应的语音意图，从而便于更准确地获取对应的目标控制指令，以便于控制对应的目标家电设备执行该目标控制指令。
在本实施例中，确定与语音请求对应的意图查询向量，包括：根据接收的语音请求结合目标家电当下所处的环境变量生成意图查询向量。采用这样的方法，能够便于进一步提高意图查询向量生成的准确性，以便于更准确地确定对应的目标控制指令，进而便于准确控制目标家电执行该目标控制指令。
需要说明的是，“当下所处目标环境变量”可以包括当下环境的温度、湿度以及用户ID等信息。
如图3所示，预先构建的意图知识图谱包括多个个人意图知识图谱；其中， 多个个人意图知识图谱对应不同的用户语音意图；预先构建的意图知识图谱中确定与意图查询向量对应的语音意图，包括：从目标对象发送的语音请求获取目标对象的语音特征；根据语音特征从多个个人意图知识图谱中确定与目标对象对应的目标个人意图知识图谱；根据目标个人意图知识图谱中确定与意图查询向量对应的语音意图。采用这样的方法，能够便于根据不同的目标对象(用户)设置不同的个人意图知识图谱，各个目标对象对应有不同的个人意图知识图谱，这样，根据语音特征确定与目标对象对应的目标个人意图知识图谱确定对应的语音意图，通过上述个性化的个人意图知识图谱的设置，能够便于根据不同的目标对象进行个性化和差异化的个人意图知识图谱的设置，从而便于快速获取对应的语音意图，提高反应速度。
在本实施例中，通过使用图神经网络技术建模，使用已有的数据训练生成用户意图推断路径，并将用户意图推断关系存为知识图谱数据，故当用户产生新的请求时，若能在知识图谱中找到相应的意图则直接变成动作触发行为。若不能找到，则将数据存储使用已经训练好的图卷积神经网络(GCN)进行意图推断，如果获得正反馈则将此条边存入知识图谱数据库，以备下次快速查询。
本实施例中采用GCN图神经网络，在处理推理问题上比普通深度学习网络CNN或者LSTM等深度网络为基础的模型理论上速度要快。因为CNN/LSTM等传统的深度学习网络擅长处理欧几里得空间中的数据。而推理数据使用的是图数据结构，如果使用传统的深度学习网络计算训练时间都长，影响用户体验。本方案采用的是知识图谱数据库与模型结合的方案，这样若推理的边存在于知识图谱数据库中，则直接触发返回结果，若没有则使用模型进行推理，并将生成的边存入知识图谱数据库中，从而下次请求来时能快速相应。
如图1示，本实施例中的模型包含查询接口、知识图谱数据库、已训练好的图神经学习模型(GCN)、意图执行模块、新意图嵌入模块。意图来临时，根据当时的环境变量生成一个意图查询向量，进入意图识别模块。首先查询知识图谱数据库，知识图谱数据库中存有各类意图节点，融合前序动作进行查询，若无前序动作则直接查询当前向量最匹配节点。意图命中则直接下发指令，若没有命中 则转到已经训练好的GCN深度网络模型进行判断。若GCN深度学习网络能够对查询向量进行泛化并且能够匹配知识图谱数据库中已有意图节点，则查询知识图谱数据库，并执行相应动作。若不能命中，但可以由模型判断出是某种意图，则将该意图存入知识图谱数据库(该动作可以不用在线执行，定期执行)。若该意图是可执行的，并且该动作是无风险的，则直接连接动作执行模块，进行动作执行。否则动作将不会执行。
对于GCN深度学习模型训练原理如下：
M个用户某次查询短文本，表示为p[1]、p[2]、……p[M]。每段话(即某个用户说的某一段话，最终落地为某个查询)可以表示为u1、u2、……uN。最终目标是将用户说的每句话转成某个具体的可执行的意图。各个查询根据顺序可以变成有向图：
G(U,V,E,R,W)；
每个查询即需要判断的意图u i包含环境属性[w1,w2,w3,……]，w为节点属性可以为任意的可以获得的用户属性值如，性别、年龄、查询时处在的房间、当时的天气、温度、湿度等。V是节点的属性、E是意图之间关系、R是意图关系的属性、W是关系之间的权值。从图中可以注意到，每句话语都有一条与其自身相连的边，这代表了话语与其自身的关系。用更实际的话来说，这就是说出的话如何影响说话者自己的思想。
以下是GCN的公式
Figure PCTCN2022120312-appb-000001
其中A是连接矩阵，用于表示意图节点与意图节点之间的连接情况。节点数为n，A是n×n矩阵,其中H 0＝X,X为节点的特征矩阵。特征维度为k,X是nxk矩阵。
本实施例中的建模过程为：根据具体的型号及设计底板，将设备真实的状态转移图画出。每个型号的机器都有独一无二的状态转移图，根据设备本身的属性将状态图画出来。
对于每一个合法的语音请求，在音箱或者APP接收到以后，会变成在控制家电设备实际工作时所有的环境变量如温度湿度用户ID等属性。在部署整个系统时，首先要使用传统的深度学习方法，或者人工指定的方式在知识图谱数据库中部署一套完整的关系。每个节点为一种明确的意图(或者就是一个明确的可以执行的电器操作)。使用数据挖掘的方法如关联算法等方法，将这些动作产生的先后关系及频繁程度计算出来，然后使用一些工具如pyspark将这些关联关系转变为节点关系图。并将这些关联关系图存入一个容量很大性能很好的知识图谱数据库如(nebula)。
在该数据库中，每个节点存有与其他节点的关系，关系也有属性如春天发生、夏天发生、一个动作序列发生、还是某个场景发生等属性。节点本身存储有意图发生时本身对应的属性如发生时候的时间、空间、用户性别、年龄。节点之间的关系表明，意图之间发生的顺序图。当某个意图发生时，在图数据库上进行查询，就会有唯一一个节点数据与其匹配。即可执行动作。若无匹配动作则调用模型进行判断。
使用图神经网络方案进行embedding。同样数据集进行训练，图神经网络模型准确度更高，可解释性更强。使用图数据库存储意图，当意图直接命中时查询速度比普通关系数据库都要节省存储资源，查询速度也更快。
使用GCN+知识图谱数据库的方案能够达到在线添加新的意图，能够识别以前没有的意图。这个方案目前也没有类似厂家做过。这样节省计算资源，不用重新调试代码，迭代快。迭代快意味着能更快的满足用户需求。边缘计算与云的方案实施该专利：将每个人产生的动作及属性的关系存在本地，即将每个人个人的知识图谱存在本地，此作为边缘计算，直接对个人用户进行推荐。定期将知识图谱推送到云端，推送后在云端将所有人的习惯汇总成一个超级知识，对其他新用户做冷启动推荐。
物联网家电控制意图识别，成为用户体验的关键步骤，各大互联网厂家都在积极布局本方案的优点在于，使用知识图谱数据库(如用nebula)能够快速上线，如用关联算法算出各个意图之间的关系图，即可理解生成节点。若是对用户个人 习惯进行意图识别使用nebula这样的能够容量海量数据的知识图谱数据库，查询性能理论上要比使用flink这样的计算平台节省资源。如使用知识图谱数据库可能10台机器就能搞定使用flink集群需要30台机器达到的效果。
使用了GCN图深度卷积网络能够更好的对图数据进行建模，如使用关联算法计算出的序列变成三元组可以直接放入GCN中进行训练，不需要再进行额外的数据处理，节省时间。GCN算出的直接就是图数据结构的预测，同一化，回归。生成的节点既可以变成知识图谱数据库中的节点，也可以结合两个部件(模型部件+知识图谱数据库部件)进行泛化意图识别，即用户查询中讲的话不在知识图谱数据库中存储。而且GCN对应类似向量匹配这样的任务效果好于CNN等传统的深度学习网络。
本公开的实施例二提供了一种控制装置，包括：语音识别模块、意图识别模块和控制指令发送模块，获取语音信息并将语音信息转化为意图查询向量。意图识别模块内存储有意图知识图谱和目标神经网络模型，意图识别模块设置为对意图查询向量进行意图识别以得到对应的控制指令。控制指令发送模块设置为目标家电设备发送控制指令。
如图4所示，本公开的实施例三提供了一种控制装置，包括：语音识别模块，设置为接收目标对象发送的语音请求，并确定与所述语音请求对应的意图查询向量；第一意图识别模块，设置为在预先构建的意图知识图谱中确定与所述意图查询向量对应的语音意图，其中，所述意图知识图谱至少用于指示多个工作状态的目标控制指令与语音意图之间的关联关系；第二意图识别模块，设置为在所述意图知识图谱中未查找到所述意图查询向量对应的语音意图的情况下，通过目标神经网络模型对所述意图查询向量进行意图识别，以确定所述意图查询向量对应的目标控制指令；控制指令发送模块，设置为向用于执行所述目标控制指令的目标家电设备发送所述目标控制指令。
本公开的实施例四提供了一种计算机可读的存储介质，计算机可读的存储介质包括存储的程序，其中，程序运行时执行上述提供的方法。
本公开的实施例五提供了一种电子装置，该电子装置包括存储器和处理器， 存储器中存储有计算机程序，处理器被设置为通过计算机程序执行上述提供的方法。
以上所述仅是本公开的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本公开原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本公开的保护范围。

Claims (17)

  1. 一种控制指令的发送方法，包括：
    接收目标对象发送的语音请求，并确定与所述语音请求对应的意图查询向量；
    在预先构建的意图知识图谱中确定与所述意图查询向量对应的语音意图，其中，所述意图知识图谱至少用于指示多个工作状态的目标控制指令与语音意图之间的关联关系；
    在所述意图知识图谱中未查找到所述意图查询向量对应的语音意图的情况下，通过目标神经网络模型对所述意图查询向量进行意图识别，以确定所述意图查询向量对应的目标控制指令；
    向用于执行所述目标控制指令的目标家电设备发送所述目标控制指令。
  2. 根据权利要求1所述的方法，其中，利用目标神经网络模型对所述意图查询向量进行意图识别，包括：
    利用目标神经网络模型对所述意图查询向量进行泛化处理；
    根据所述泛化处理的结果对所述意图查询向量进行意图识别。
  3. 根据权利要求2所述的方法，其中，根据所述泛化处理的结果对所述意图查询向量进行意图识别，包括：
    将所述泛化处理的处理结果与所述预先构建的意图知识图谱中存储的语音意图进行比对；
    在所述预先构建的意图知识图谱中存在与所述泛化处理的处理结果匹配的语音意图的情况下，将与所述意图查询向量匹配的语音意图确定为对所述意图查询向量进行意图识别的识别结果；
    在所述预先构建的意图知识图谱中未查找到与所述泛化处理的处理结果匹配的语音意图的情况下，利用所述目标神经网络对所述意图查询向量进行意 图推断，并根据所述意图推断得到的语音意图对所述意图查询向量进行意图识别。
  4. 根据权利要求3所述的方法，其中，利用所述目标神经网络对所述意图查询向量进行意图推断之后，所述方法还包括：
    确定所述意图推断得到的语音意图是否能够执行；
    当确定出所述意图推断得到的语音意图能够执行的情况下，将所述意图推断得到的语音意图存储至所述预先构建的意图知识图谱；
    在确定出所述意图推断得到的语音意图无法执行的情况下，退出对所述意图查询向量的意图识别。
  5. 根据权利要求1所述的方法，其中，所述意图知识图谱中存储的意图信息包括意图节点，在预先构建的意图知识图谱中确定与所述意图查询向量对应的语音意图，包括：
    获取所述意图查询向量的前序动作，其中，所述前序动作是指接收所述语音请求前的执行动作；
    将所述意图查询向量与前序动作融合后与所述意图知识图谱中存储的意图节点进行匹配，以获取最匹配意图节点；
    根据所述最匹配意图节点确定与所述意图查询向量对应的语音意图。
  6. 根据权利要求1所述的方法，其中，确定与所述语音请求对应的意图查询向量，包括：
    根据接收的所述语音请求结合所述目标家电当下所处的环境变量生成意图查询向量。
  7. 根据权利要求1所述的方法，其中，所述预先构建的意图知识图谱包括多个个人意图知识图谱；其中，所述多个个人意图知识图谱对应不同的用户语音意图；预先构建的意图知识图谱中确定与所述意图查询向量对应的语音意图，包括：
    从目标对象发送的语音请求获取所述目标对象的语音特征；
    根据所述语音特征从所述多个个人意图知识图谱中确定与所述目标对象对应的目标个人意图知识图谱；
    根据所述目标个人意图知识图谱中确定与所述意图查询向量对应的语音意图。
  8. 一种控制装置，包括：
    语音识别模块，设置为获取语音信息并将所述语音信息转化为意图查询向量；
    意图识别模块，所述意图识别模块内存储有意图知识图谱和目标神经网络模型，所述意图识别模块设置为对所述意图查询向量进行意图识别以得到对应的控制指令；
    控制指令发送模块，所述控制指令发送模块设置为目标家电设备发送所述控制指令。
  9. 一种控制装置，包括：
    语音识别模块，设置为接收目标对象发送的语音请求，并确定与所述语音请求对应的意图查询向量；
    第一意图识别模块，设置为在预先构建的意图知识图谱中确定与所述意图查询向量对应的语音意图，其中，所述意图知识图谱至少用于指示多个工作状态的目标控制指令与语音意图之间的关联关系；
    第二意图识别模块，设置为在所述意图知识图谱中未查找到所述意图查询向量对应的语音意图的情况下，通过目标神经网络模型对所述意图查询向量进行意图识别，以确定所述意图查询向量对应的目标控制指令；
    控制指令发送模块，设置为向用于执行所述目标控制指令的目标家电设备发送所述目标控制指令。
  10. 根据权利要求9所述的控制装置，其中，所述控制装置还包括：泛化处理模块，设置为利用目标神经网络模型对所述意图查询向量进行泛化处理；根据所述泛化处理的结果对所述意图查询向量进行意图识别。
  11. 根据权利要求10所述的控制装置，其中，所述控制装置还包括：比对模块，设置为将所述泛化处理的处理结果与所述预先构建的意图知识图谱中存储的语音意图进行比对；在所述预先构建的意图知识图谱中存在与所述泛化处理的处理结果匹配的语音意图的情况下，将与所述意图查询向量匹配的语音意图确定为对所述意图查询向量进行意图识别的识别结果；在所述预先构建的意图知识图谱中未查找到与所述泛化处理的处理结果匹配的语音意图的情况下，利用所述目标神经网络对所述意图查询向量进行意图推断，并根据所述意图推断得到的语音意图对所述意图查询向量进行意图识别。
  12. 根据权利要求11所述的控制装置，其中，所述控制装置还包括：判定模块，设置为确定所述意图推断得到的语音意图是否能够执行；当确定出所述意图推断得到的语音意图能够执行的情况下，将所述意图推断得到的语音意图存储至所述预先构建的意图知识图谱；在确定出所述意图推断得到的语音意图无法执行的情况下，退出对所述意图查询向量的意图识别。
  13. 根据权利要求9所述的控制装置，其中，所述控制装置还包括：获取模块，设置为获取所述意图查询向量的前序动作，其中，所述前序动作是指接收所述语音请求前的执行动作；将所述意图查询向量与前序动作融合后与所述意图知识图谱中存储的意图节点进行匹配，以获取最匹配意图节点；根据所述最匹配意图节点确定与所述意图查询向量对应的语音意图。
  14. 根据权利要求9所述的控制装置，其中，所述控制装置还包括：环境变量获取模块，设置为获取所述目标家电当下所处的环境变量；所述语音识别模块设置为根据接收的所述语音请求结合所述目标家电当下所处的环境变量生成意图查询向量。
  15. 根据权利要求9所述的控制装置，其中，所述预先构建的意图知识图谱包括多个个人意图知识图谱；其中，所述多个个人意图知识图谱对应不同的用户语 音意图；所述控制装置还包括：第三意图识别模块，设置为从目标对象发送的语音请求获取所述目标对象的语音特征；根据所述语音特征从所述多个个人意图知识图谱中确定与所述目标对象对应的目标个人意图知识图谱；根据所述目标个人意图知识图谱中确定与所述意图查询向量对应的语音意图。
  16. 一种计算机可读的存储介质，其中，所述计算机可读的存储介质包括存储的程序，其中，所述程序运行时执行权利要求1至7中任一项所述的方法。
  17. 一种电子装置，包括存储器和处理器，其中，所述存储器中存储有计算机程序，所述处理器被设置为通过所述计算机程序执行权利要求1至7中任一项所述的方法。
PCT/CN2022/120312 2022-08-18 2022-09-21 控制指令的发送方法、控制装置、存储介质及电子装置 WO2024036694A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210993981.9 2022-08-18
CN202210993981.9A CN115356939A (zh) 2022-08-18 2022-08-18 控制指令的发送方法、控制装置、存储介质及电子装置

Publications (1)

Publication Number Publication Date
WO2024036694A1 true WO2024036694A1 (zh) 2024-02-22

Family

ID=84002587

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/120312 WO2024036694A1 (zh) 2022-08-18 2022-09-21 控制指令的发送方法、控制装置、存储介质及电子装置

Country Status (2)

Country Link
CN (1) CN115356939A (zh)
WO (1) WO2024036694A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115826428B (zh) * 2023-02-23 2023-10-27 海尔优家智能科技（北京）有限公司 家居设备的控制方法和装置、存储介质及电子装置
CN116992862B (zh) * 2023-09-26 2024-02-20 中科航迈数控软件(深圳)有限公司 数控机床异常辅助处理方法、装置、设备及介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110399609A (zh) * 2019-06-25 2019-11-01 众安信息技术服务有限公司 意图识别方法、装置、设备及计算机可读存储介质
CN112435660A (zh) * 2019-08-08 2021-03-02 上海博泰悦臻电子设备制造有限公司 车辆控制方法、系统及车辆
CN114694644A (zh) * 2022-02-23 2022-07-01 青岛海尔科技有限公司 语音意图识别方法、装置及电子设备
CN114863928A (zh) * 2022-07-07 2022-08-05 南昌华飞物联技术有限公司 智能家居语音控制方法、系统、电子设备及存储介质

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110784768B (zh) * 2019-10-17 2021-06-15 珠海格力电器股份有限公司 一种多媒体资源播放方法、存储介质及电子设备
CN110795532A (zh) * 2019-10-18 2020-02-14 珠海格力电器股份有限公司 一种语音信息的处理方法、装置、智能终端以及存储介质
CN113486233A (zh) * 2020-08-21 2021-10-08 青岛海信电子产业控股股份有限公司 一种内容推荐方法、设备及介质
CN112151035B (zh) * 2020-10-14 2023-08-11 珠海格力电器股份有限公司 一种语音控制方法、装置、电子设备及可读存储介质
CN114500143A (zh) * 2022-02-14 2022-05-13 珠海格力电器股份有限公司 一种基于知识图谱控制设备的方法及装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110399609A (zh) * 2019-06-25 2019-11-01 众安信息技术服务有限公司 意图识别方法、装置、设备及计算机可读存储介质
CN112435660A (zh) * 2019-08-08 2021-03-02 上海博泰悦臻电子设备制造有限公司 车辆控制方法、系统及车辆
CN114694644A (zh) * 2022-02-23 2022-07-01 青岛海尔科技有限公司 语音意图识别方法、装置及电子设备
CN114863928A (zh) * 2022-07-07 2022-08-05 南昌华飞物联技术有限公司 智能家居语音控制方法、系统、电子设备及存储介质

Also Published As

Publication number Publication date
CN115356939A (zh) 2022-11-18

Similar Documents

Publication Publication Date Title
WO2024036694A1 (zh) 控制指令的发送方法、控制装置、存储介质及电子装置
US11616845B2 (en) Electrical meter for determining a power main of a smart plug
US10818293B1 (en) Selecting a response in a multi-turn interaction between a user and a conversational bot
EP3857860B1 (en) System and method for disambiguation of internet-of-things devices
CN109547308A (zh) 一种智能家居的控制方法、装置、存储介质及服务器
WO2024001101A1 (zh) 文本意图的识别方法和装置、存储介质和电子装置
WO2023168838A1 (zh) 语句文本的识别方法和装置、存储介质及电子装置
WO2024021407A1 (zh) 知识图谱的更新方法和装置、存储介质及电子装置
CN115047778A (zh) 智能设备的控制方法和装置、存储介质和电子装置
WO2023207170A1 (zh) 洗涤程序的推荐方法及装置、存储介质及电子装置
Zheng et al. Mutual benefit aware task assignment in a bipartite labor market
CN114694644A (zh) 语音意图识别方法、装置及电子设备
CN103793536A (zh) 一种智能平台实现方法及装置
WO2024001189A1 (zh) 食物存储信息的确定方法及装置、存储介质及电子装置
CN114911535B (zh) 应用程序组件配置方法、存储介质及电子装置
WO2023168853A1 (zh) 使用意图的预测方法和装置、存储介质及电子装置
CN114915514B (zh) 意图的处理方法和装置、存储介质及电子装置
WO2024036804A1 (zh) 意图指令的确定方法及装置、存储介质及电子装置
CN114925158A (zh) 语句文本的意图识别方法和装置、存储介质及电子装置
Karimi et al. Comparing Prediction Models for Active Learning in Recommender Systems.
Ventura et al. Smart EDIFICE—Smart EveryDay interoperating future devICEs
CN112954051B (zh) 一种用于食材加工的远程控制方法及系统
TWI837187B (zh) 用於物聯網設備消歧的系統和方法
Lu et al. Mining and Analyzing Behavior Patterns of Smart Home Users Based on Cloud Computing
CN116432658A (zh) 语音数据的处理方法和装置、存储介质及电子装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22955470

Country of ref document: EP

Kind code of ref document: A1