<p>This paper presents a new method in the field of TV video classification. The main idea is to identify each character in the video, then analyze their biography to infer the video’s category. For example, actors are generally associated with films and series, athletes with sports programs, and politicians with news broadcasts. Once the name of each character and their number of appearances are detected, the system performs online web searching to gather biographical information about each individual. This data is then processed using word embedding techniques combined with a dual weighting strategy—one that accounts for both the frequency of each individual’s appearance and the collective distribution of all identified characters across the video. This approach enables the system to extract semantic signals from the biographies and effectively predict the overall topic of the video. Unlike most existing systems that are limited to short video classification, our approach works with both short and long videos. The use of biographies makes the classification process faster, as text processing is less computationally intensive than full visual or audio analysis. Our system can be trained on any dataset of well-known individuals, such as CelebA. In our case, we constructed the Algerian Famous Faces Dataset (AFFD), which includes 477 individuals with 200 frames per person, specifically designed for Algerian TV video classification. The targeted categories include sports programs, cooking shows, movies, and political content.</p>