Research And Implementation Of Smart Home Interactive Question Answering Method Based On Slot Filling

In recent years,more and more smart devices have entered ordinary households,bringing great convenience to users’ daily life and quietly changing people’s traditional lifestyle.In the face of more and more smart devices,the traditional way of using multiple physical control terminals,such as infrared remote control or data panels and other forms of control,can no longer meet the user’s control needs of a wide range of smart devices.And voice control,a non-contact control form to free the user’s hands,is considered the most natural and convenient means of control.However,most of the existing voice control systems respond only to the user’s explicit control commands,and cannot recognize and respond to the control commands with unclear semantics or implicit control requirements included in the user’s daily conversation.To address the above problems,this paper proposes a control model that recognizes user control intentions and understands dialogue semantics through human-machine multi-round question-and-answer interactions,mainly based on slot-filling methods,combined with interactive Q&A,and uses it for intelligent control of smart home devices.In this paper,the interactive Q&A approach for smart home based on slot filling is investigated from the following three aspects.(1)To address the problem that the implicit control demand device subjects in users’ daily dialogues are difficult to identify,an intention recognition algorithm incorporating BILSTM and TextCNN neural network is proposed based on the BERT pre-training model to realize the device control subject recognition in interactive Q&A.The BERT pre-training model is first used to obtain a textual word vector representation,and then BILSTM and TextCNN neural networks are used for feature extraction,respectively,and the fused textual features are classified via a fully connected layer.Experiments were conducted on the home control conversation dataset and microblog sentiment analysis dataset.The average recognition rate of explicit and implicit commands reached 93.5% in the home control conversation dataset,and better results were also achieved in several other classification datasets.(2)To address the lack of device control command elements in users’ daily conversations,this paper proposes a slot-filling algorithm combined with syntactic analysis which is used for text semantic understanding in interactive Q&A and completes semantic element completion.The lexical features of the dialog text are first obtained by a pre-trained component syntactic parser,and then fused with the text word vector and fed into a BILSTM+CRF-based sequence annotation network to obtain the slot annotation results of the text.Experiments are conducted on the self-built home control dialogue slot dataset and the classical slot-filling dataset ATIS based on the BIO annotation method.An F1 value of 91.7% is achieved on the self-built home dialogue slot dataset,which is 3.5% higher than the F value of the method without fused syntactic analysis,while an F1 value of 90.6% is achieved on the ATIS dataset.(3)To address the problem that existing voice control systems for smart home devices cannot effectively recognize implicit control commands,this paper proposes an interactive Q&A control method for smart homes is proposed by combining the above intention recognition algorithm and slot filling algorithm,and contrasts a prototype home control system on this basis.Firstly,the standard control command slots,filling information and interactive feedback statements are preset;then the human-computer interaction Q&A logic is designed.The corresponding slot-based standard control command presets are obtained by combining the recognition results of the intention recognition algorithm,and the slot filling algorithm is relied on to extract the key slot information of the dialogue text for filling the preset labeled control commands.The transformation from user dialogue to standardized home control commands is completed by multiple rounds of interactive Q&A between the user and the system.Then combine the intention recognition algorithm and slot filling algorithm to realize the transformation of the standardized home control commands.Finally,the prototype system of question-and-answer control based on this method is built.The system test shows that the recognition rate of the model built in this paper reaches 96.5% for explicit commands after two rounds of interaction and 94.0% for implicit commands after three rounds of interaction.The interactive question-and-answer method studied in this paper is oriented to the practical needs in the smart home environment,and effectively improves the drawbacks of the home speech control system,such as engendering suboptimal solutions easily and difficult to recognize implicit commands.The test results show that the recognition rate of home control commands in smart home environment has been improved to a certain extent and has certain usability,which provides some support for the subsequent research in the field of smart home control.