The Researchof Video Tags Generationand Classification Based On Text Mining

In the era of rapid development of information technology,the way people get information is gradually from traditional media,like newspapers,radio,and television,to transfer to multimedia data on the internet.Widely used of mobile phone,iPad and other mobile devices leads to network video,music,text and other data has become an important source of obtaining,storing information.Due to the rapid growth of the scale of cyber source,the amount of video data is enormous.If you don't sort out these video data effectively,it will become very difficult for the user to quickly get the data,or the resource managers to deal with this information.Therefore,through the classification of video resources,construction of the structure of the database,it will greatly improve the rate of Internet users to access to their desired video resources,and can help resource managers to efficiently analyze and process video data.From where,they can dig out more business value from the analysis of user behavior and preferences.Video classification method is mainly divided into two categories.One is based on image features,which is based on the image analysis of key frames of video classification.The study found that its time and space cost is high,and with the influence of the image quality,it has many limitations and is low efficiency.The other is based on text mining,through the analysis of the text description of video classification method for video.The corpus used for mining mainly comes from basic description,the user's contribution of video label and the comment text.Due to the relatively high efficiency of text data processing,the technology is relatively mature.Therefore,video classification method based on text mining has its feasibility and research value.This paper mainly studies the following aspects:1)Text mining based on the video description,tags,comments,to extract valuable information for video classification.Then train the classifier with text classification model.According to the text description of the unknown video,the classifier will classify it.According to the classification results,the performance of the classifier could be evaluated.2)Key focuses on Text Mining:feature space dimensionality reduction.Feature item evaluation function will determine the feature items space structure,and then affect the results of the final classification.In this paper,based on the existing text mining theory,the feature evaluation method,TCD(Term Category Discrimination),is proposed.By the experiments of video classification test and contrast test to IG(Information Gain),?~2Statistics and other methods,it is verified that the TCD feature evaluation function the classification accuracy was improved with better performance on different data sets.3)To redesign the calculation method of video similarity based on Naive Bayesian's posteriori probability and classification results.By combining the TCD feature value with the TFIDF value,video is labeled,then applies both to the video recommendation strategy.4)To design video recommendation system based on video classification.According to the user's viewing behavior,the video tags are converted into user tags.Based on computing of video similarity and user tags,a variety of recommendation strategies are set up.And the function of recommendation interface,backstage data,recommendation effect,system performance will be tested.To prove that applying video classification based on text mining to video recommendation can optimize the effect and improve the efficiency.