Automatically Generating the Description of Chinese Video from the Web Using Topic Model
Quan Qi · ASME Press eBooks · 2011
At present, videos from the Web are always lack of effective description for retrieval and recommendation. Automatically generating the description of the videos from the Web is a hard task. In this paper, we propose a method which utilizes the titles of videos and Latent Dirichlet Allocation (LDA), a hidden topic model, to generate pseudo-descriptions for videos. In our method, we first evaluate video titles' expression ability. The title which has enough information to outline the video it belongs to will be send to the trained LDA to get its topic distributions. Using the topic distributions and word-topic distributions gotten from the LDA, we get the most likely words that belong to the video as its pseudo-description. In experiments, 62305 Chinese automotive video titles are crawled from web for testing our method. The generated descriptions of videos are evaluated manually. The result shows that the pseudo-descriptions can describe the videos to some extent, and this method provides a promising way for automatically generating descriptions of the Web videos.