Research On Internet Hot Topic Recommendation Based On Deep Learning

With the rapid development of Internet technology and the popularity of globalization,the scale of Internet users continues to expand,Internet become an important platform for Internet users to access information followed.A large number of corpuses are exploding on the Internet,therefore extracting valuable information from redundant text data has become an important research content in the current computer field.If we rely solely on these disorganized public opinion data,we will not only increase the workload,but also reduce accuracy due to subjective consciousness of people.Based on the above research background,the starting point of the research on hot topic recommendation research is how to quickly find and recommend the topics that most hot and most discussions currently on the net platform from the vast and diverse public information.In recent years,deep learning,as a frontier field,blends of multi-disciplines have made many breakthroughs in the frontier are,the application of Deep Learning in Natural Language Processing field has brought new solutions to the problems.The main work of this paper is as follows:The traditional language model can not avoid ignoring the main information in the process of word vectorization and can not combine with the whole content text analysis,in this paper,we design a text representation of word feature method based on convolutional neural network,which extracts features from texts by convolutional core sliding up and down.Using Word2 Vec pre-training language model to realize the transformation of word vector.Introducing TF-IDF algorithm to enhance text features,and the text feature is extracted by convolutional neural network.Through the experiment,exploring the parameters that influence the result of model,and adjusting the parameters to make the model reach the optimal effect.Compared with the neural network language model,the text representation method proposed in this paper has higher F values on two different news data sets,proves the reliability of this text representation method.In this paper,optimizing the initial point selection problem of K-means algorithm after understanding the traditional K-MEANS and DBSCAN clustering algorithm,optimizing the initial point selection problem of K-means Algorithm,and proposes a clustering algorithmbased on the maximum density in the moving range,the algorithm is used to design and implement the recommendation part of hot topics,the algorithm not only solves the problem of selecting the number of points,but also solves the problem of selecting the location of points,and also remedy time-consuming shortcomings of DBSCAN algorithm.Using the method of ring domain scope is to speed up the computing speed.The effectiveness of the proposed clustering algorithm and its superiority in generalization are verified by experiments.The experimental results show that the proposed algorithm has a good effect on the accuracy of clustering and the evaluation of operation time.Combining the word feature text representation based on the convolutional neural network with a clustering algorithm based on the maximum density of moving range,search for the optimal text based on the Dataframe storage to achieve the recommendation purpose of the text,proposing a research framework based on deep learning for hot topics recommendation of Internet,and through the experiment,the target is realized and the feasibility is proved.