Topic detectionmodel in a single-domain corpus inspired by thernhuman memory cognitive process

摘要

A corpus (eg, patents or news texts) is an important knowledge resource that contains variousrntopics, such as specific technologies or social events. Topic detectionmodels of corpus, eg, LatentrnDirichlet Allocation and KeyGraph, provide an important basis for exploring the status quo andrntrends in science, technology, or social events. However, these models suffer from low retrievalrnperformance as they only consider text own explicit semantics in a single-domain corpus. In addition,rnmany incremental models, such as online-LDA, are based on time slices. In this paper, a newrntopic detectionmodel is proposed to improve the topic detection performance of a single-domainrncorpus,which is inspired by a human memory cognitive process (THC). First, to improve the accuracy,rndistributions over words and inter-word relations across a corpus are utilized as backgroundrnknowledge, which is a type of implicit semantics, and we can find a more semantic-sensitive partrnof texts. Second, to realize online topic detection without time slices, we introduce a probabilityrngain-based dynamic probabilistic model to detect latent topics by learning a model based onrnthe dynamic human memory cognitive process. These two steps constitute the framework ofrnour model. The experimental results for four public datasets (Reuters-R8, Reuters-R52,WebKB,rnand Cade12) reveal that our model is approximately ten percent higher than other baselines (eg,rnKeyGraph and LDA) on the Adjusted Rand Index (ARI).