Dissertation
Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Text Processing

Research into Topic Detection, Tracking and Trend Prediction

Author QiLei
Tutor JiangMing
School Hangzhou University of Electronic Science and Technology
Course Applied Computer Technology
Keywords Topic Detection and Tracking synonym forest multi-vector model adaptive feedback grey prediction
CLC TP391.1
Type Master's thesis
Year 2014
Downloads 7
Quotes 0
Download Dissertation

We have completely got rid of the shackles of poor information with the emergence of Internet,and human beings enter an society rich of E-Byte information. Up to now, the key issue is no longerthe source of the information but how to access information fastly and accurately. Under thiscondition, Topic Detection and Tracking (Topic Detection and Tracking, TDT) technology wasproposed. By topic detection and tracking, people can be effectively aggregate and organizedispersed information to help users find the relationship between various events, to learn all thedetails of an event and to learn the relationship between events. In addition, the trend prediction canmakes it possible to know the topic development early or to respond to emergencies timely.In this paper, with regard to topic detection technique, a method proposed is a mix modelbased on expansion technology using a synonym forest combined with multi-vector model. Theaccuracy of term similarity can be improved through expansion technology using a synonym forestbecause synonym forest has somewhat semantic degree. Considering a topic has somecharacteristics:"Who","When","Where" and "What", this paper proposed a multi-vector model toimproved the accuracy of detection to some extent. Regarding technical difficulties of tracking-topic drift, this paper presents an adaptive feedback learning strategies. In the feedback learning,using an incremental approach to improve the topic model, and in each tracking process, constitutesa weak feedback tracker. The overall topic tracking model is the combinations of all weak trackers.Thus, this method can reduce the impact of erroneous feedback. Experiment results show that thetopic detection and tracking model proposed in this paper is better than the classic topic detectionand tracking model.Based on gray theory, this paper presented prediction model for topic trend field where are notnumber researchers interest in. The difficulties of prediction of topic trend are the short cycle, thesmall number of samples and uncertainty. Several related classical statistical forecasting model. Aregiven in this paper. After the analysis of the characteristics of the topic and the Grey System Theorywhich has advantage in studying poor information and uncertainty, the gray prediction algorithm isproposed. The strict comparison to two classical statistical forecasting models (exponentialsmoothing model and ARIMA model) shows that the prediction model proposed in this paper issuperior to the classical statistical forecasting models not only in terms of the complexity but alsothe prediction accuracy.

Related Dissertations
More Dissertations