Acrost: A Topic Self-adaptive Academic Conference Retrieval System

According to the statistics available, the number of international academicconferences held all over the world each year has reached more than10thousand times,and there are also millions of people to attend. Moreover, academic conferences have thecharacteristics of variety and complexity, some of them are one-off, and others are serial.Facing the urgent demand of retrieval about academic conferences from huge number ofresearchers, current existing academic search engines or digital libraries which concentrateon document retrieval can’t give a right response.Acrost, a CFP-Oriented and Topic Self-Adaptive Academic Conference RetrievalSystem, is proposed. It has the way of topic-based retrieval, and it provides services ofboth academic conference retrieval and unique submission recommend. In order to obtainsufficient data source, it uses two methods:(1) the manner based on the general searchengine. This way can save amount of resource costs, and by using Support Vector Machineclassifier, noise information will be filtered;(2) the focused crawler based on Vector SpaceModal. The crawler will crawled academic conference webpages with a specified website.After downloading the original academic conference webpages, Regular Expression andConditional Random Fields will be used to complete the tasks of Information Extractionand Entity Recognition for dealing with semi-structured and non-structured webpagesrespectively, and then the metadata will be collected. Then, inverted index will be createdby using Lucene for the metadata; meanwhile, a topic discovery method based onincremental hierarchical clustering algorithm is proposed to parse out topics of useruploaded PDF document automatically. In addition, an academic conference evaluationmodel based on academic impact factor is established in Acrost, and the consideredindicators contain cited count per paper, acceptance rate, etc.The experimental results show that the Recall, Precision and F-measure for service ofacademic conference retrieval are respectively84.8%,90.5%,87.6%; and60.8%,68.7%,64.5%for service of submission recommend; Besides, Acrost can respond to user servicerequest quickly. This shows that Acrost has a good performance on relevancedetermination and run speed.