Intelligent TV Control Instruction Recognition Based On BiLSTM

Named entity recognition(NER)is an important branch of natural language processing technology.By identifying the entities in the text and marking them out,we can highlight the key points from the natural language,extract the key needs of users,facilitate the further processing of the system,and realize the functions expected by users.Intelligent voice TV is the representative of the development direction of the combination of traditional electronic products and AI.In the human-computer interaction,the traditional way of operating panel control machine will be replaced by more intelligent ways such as voice control,gesture recognition and face recognition to a certain extent.It is more convenient for people’s life to liberate their hands.With the development of deep learning,the research of named entity recognition technology has a new idea,and the named entity recognition technology based on neural network has a greater improvement in performance.The data structure of intelligent voice TV instruction recognition is simple and the amount of data is large,so the application of this combination in intelligent voice TV will have a very considerable effect.Aiming at the task of named entity recognition in the field of intelligent voice TV,this paper studies the design and implementation of named entity recognition model under deep learning.Based on the description of the concept,structure and principle of distributed representation of word vector,seq2 seq,LSTM,CRF and attention mechanism,this paper introduces deep learning,tensorfl The related technologies of ow and crawler are analyzed,and the requirements of intelligent voice TV system are analyzed.The data and process of the system are modeled by activity diagram,entity contact diagram and data flow diagram respectively.The data set of intelligent voice TV in specific field is collected,constructed,classified and preprocessed.The model structure of intelligent voice TV instruction recognition is designed The number of layers,neurons and other related parameters of the whole network.Under the method of deep learning,the algorithm design and code implementation of intelligent speech instruction recognition based on bilstm are realized,the results of visual recognition are visualized,the performance indexes of the method are tested,the ablation experiment is designed for attention mechanism,and the other variants of LSTM and cyclic neural network,such as Gru,are compared and the relevant experimental results are counted.The application of named entity recognition based on bilstm in the field of intelligent voice TV shows the feasibility and efficiency of the structure extraction entity of bilsm,attention and conditional random field.The experiment of function test shows that the recognition and annotation function of instruction can be realized with very high accuracy,which lays a foundation for the next step to realize the complete function of intelligent voice TV system with this result.The experiment of performance test shows that the effectiveness of attention mechanism is better than that of the model without attention mechanism.However,the effect is not obvious in this experiment.The sequence of data set is short,and the advantage of attention mechanism can only be shown when the sequence is too long.Another comparative experiment is the difference of RNN units The purpose is LSTM,compared with Gru,the performance of which is not much different.However,both of them have advantages and disadvantages.The number of GRU parameters is less than that of LSTM,and the convergence speed of the model is fast.However,if the data set scale is very large,the expression ability of LSTM will be better.In this paper,the successful application of named entity recognition task of intelligent voice TV proves that deep learning can extract feature effect and attention mechanism can select subsets structurally,avoid high-dimensional operation,and contribute to the performance of the whole model.It is expected that more technologies will be combined with deep learning technology in the future to create a convenient AI + life for human beings.