Abstract:

Video titles,which are used to describe video content in the text-based video retrieval and recommendation system,are essential sources of information. However,there are no in-depth researches on the description ability of video titles. In this paper,a method is proposed to evaluate the description ability of titles,in which the description ability is divided into three levels that are respectively describable,incomprehensible and comprehensible but undescribable,and the evaluation of the description ability is considered as a classification problem. Moreover,the support vector machine ( SVM) with high recognition accuracy for small samples is used as the classifier. Besides,in order to fill up the shortage of title information,the search results in the search engine are used as the supplements. Finally,the proposed method is applied to the classification of 5000 automotive video titles. that,with the help of the search results,the F scores increase by 3% for both describable and incomprehensible titles.

Key words: video title, description ability, text analysis, video retrieval, support vector machine