摘要： Topic Detection and Tracking is a popular topic clustering method in the big data age, which aims at automatic recognition of new topics and continuous tracking of known topics in news information flow. Traditional Topic Detection and Tracking mainly studies short text. With the rapid development of digital devices and communication techniques, the news is going to be longer and richer. So nowadays traditional Topic Detection and Tracking is faced with three problems, first, long news text usually contains multiple traditional clustering algorithm cannot accurately identify them. Second, traditional clustering mostly uses multi-dimensional computation based on word bag, but the time-consuming of this multi-dimensional computation increases exponentially with the increase of the length and number of articles. Third, long-text news contains more information. How to show the continuity and relevance of long-text news in a better way is very important and meaningful. Therefore, an improved clustering algorithm based on single-pass is presented in this paper, which can solve the above problems primly. Experiments show that, compared with K-means clustering algorithm, agglomerative hierarchical clustering algorithm, Density-Based Spatial Clustering of Applications with Noise and hierarchical clustering on the constructed concept graph, the accuracy of this algorithm is improved by about 20% to 30%, the recall rate is increased by 10% to 20%, and the algorithm time is reduced by more than 40%. With the increase of the number of articles, the time-consuming curve of the improved single-pass clustering algorithm approximates a linear function. For each additional article, the time required for the algorithm is only 0.1-0.5 times that of other algorithms. Besides, by adding timelines and extracting topics in the theme during presentation, the algorithm can effectively mine the continuity and relevance information of news topics and track the changes of news topics.