Research And Implementation Of Chat Robot Algorithm Based On Reinforcement Learning

With the development and application of deep learning algorithms in the field of vision,the application of textual problems related to natural language processing has gradually begun to be applied.Based on the background of big data analysis and the increasingly mature deep learning algorithm model,intelligent chat bots have gradually been widely used.Among them,the customer service chat robot is widely used,which saves a lot of manpower and time.And with the development of algorithms,intelligent chat robots are constantly optimized,the answers are more diverse,and the accuracy rate is constantly improving.Many research institutions and large domestic companies are developing chat bots,but relying on traditional relatively stable machine learning algorithms.The application of deep learning models in natural language processing is only the initial stage.After continuous research and development,I believe that there will be better development in this field in the future.At present,the research on intelligent chat robot mainly develops open field dialogue.The applied deep learning algorithm is mainly based on the seq2 seq framework in the cyclic neural network.This paper mainly studies deep reinforcement learning and modeling of the seq2 seq model on chat robots.The main application of language processing.Then the core algorithm of seq2 seq model is described,including mathematical modeling of RNN and improved LSTM and CNN model.Then add the attention mechanism to the seq2 seq model to improve the model effect.Finally,the model effect is verified by self-built corpus.This paper proposes a model(LDA+RL)fusion method combining topic model and deep reinforcement learning to improve the accuracy of response.The learning method of intensive learning is trial-and-error learning.It is difficult to train in the production of chat robots,and it requires a lot of predictions in the process of training the model to achieve better results.In order to speed up the training is the speed and efficiency of calculation,we Using the topic model as a pre-processing model to discover potential topic information in the text,according to the potential topic information,we can grasp the key points of the reply when replying to the information,and the accuracy of the generated response will increase.In this paper,we use the web crawler technology to capture the video and dialogue corpus,and after data cleaning,it becomes the trainable corpus input into the LDA+RL model.In addition,in order to increase the contrast,we also use the corpus we built to train RL.Finally,based on the experimental results,we use the combination of objective evaluation and subjective evaluation to evaluate the effect of the model.The results show that the hybrid model after adding the topic model has better training effect than the reinforcement learning.