BBS Topic's Hotness Forecast Based on Back-Propagation Neural Network

Abstract

Online hot topic detection is a significant research field in web data mining, which can help people make policy decision or benefit to people's daily life. Actually, in recent years more and more hot topics are arising from BBS, often referred as online forum. BBS provide a communication platform for people to discuss and express their views. It's obvious that forecasting the hotness topics on BBS is important and meaningful. In this paper we present an approach to predict the hotness of topics based on BPNN (Back-Propagation Neural Network). Text sentiment internet user's attention and hotspot relative have been considered to forecast the hotness of topics. At the last the experiment results over SINA reading forum show our approach is effective.

An on-line adaptive topic evolution model in web discussions

“…It is defined as the average value of KL (p||q) and KL (q||p), expressing the similarity between topic p and q. As || || || (6) Define distance vector DV in dimension , in which DV(k) refers to the similarity of topic k in time t-1 and time t. Introducing the measurement of New Topic Confidence Level (NTCL, the percentage that a topic reaches the confidence level at time t) to determine makes the percentage of the distance less than in all distances be equal to NTCL. The NTDetect algorithm is described in Algorithm1: Algorithm 1 NTDetect(DV,NTCL,β) Parameter: DV: distance vector of topics, NTCL: New Topic Confidence Level,β: Dirichlet distribution of the words 1.…”

“…In recent years, many researches on topic evolution targeted on news topics, while only a few of focused on the evolution of web forum topic [4] [5]. These researches mainly concentrated on the detection and early warning of hot topics [6][7] [8]. The information in web forum topic is different from that in news topics due to its short content, concept ambiguity and fast change.…”