History-enhanced focused website segment crawler

摘要

The primary challenge in focused crawling research is how to efficiently utilize computing resources, e.g., bandwidth, disk space, and time, to find as many web pages related to a specific topic as possible. To meet this challenge, we previously introduced a machine-learning-based focused crawler that aims to crawl a group of relevant web pages located in the same directory path, called a website segment, and has achieved high efficiency so far. One of the limitations of our previous approach is that it may repeatedly visit a website that does not serve any relevant website segments, in the scenario where the website segments share the same linkage characteristics as the relevant ones in the training dataset. In this paper, we propose a “history-enhanced focused website segment crawler” to solve the problem. The idea behind it is that the priority score of an unvisited website segment should be reduced if the crawler has consecutively downloaded many irrelevant web pages from the website. To implement this idea, we propose a new prediction feature, called the “history feature”, that is extracted from the recent crawling results, i.e., relevant and irrelevant web pages gathered from the target website. Our experiment shows that our newly proposed feature could improve the crawling efficiency of our focused crawler by a maximum of approximately 5%.