Exploration versus Exploitation in Topic Driven Crawlers Gautam Pant@ Padmini Srinivasan@! Filippo Menczer@ @ Department of Management Sciences ! School of Library and Information Science The University of Iowa Iowa City, IA 52242 {gautam-pant,padmini-srinivasan,filippo-menczer}@uiowa.edu Abstract ered by search engines has not improved much over the past few years [16]. Even with increas- The dynamic nature of the Web highlights the ing hardware and bandwidth resources at their scalability limitations of universal search en- disposal, search engines cannot keep up with the gines. Topic driven crawlers can address the growth of the Web and with its rate of change problem by distributing the crawling process [5]. across users, queries, or even client computers. These scalability limitations stem from search The context available to a topic driven crawler engines’ attempt to crawl the whole Web, and to allows for informed decisions about how to prior- answer any query from any user. Decentralizing itize the links to be visited. Here we focus on the the crawling process is a more scalable approach, balance between a crawler’s need to exploit this and bears the additional benefit that crawlers information to focus on the most promising links, can be driven by a rich context (topics, queries, and the need to explore links that appear sub- user profiles) within which to interpret pages and optimal but might lead to more relevant pages. select the links to be visited. It comes as no We investigate the issue for two different tasks: surprise, therefore, that the development of topic (i) seeking new relevant pages starting from a driven crawler algorithms has received significant known relevant subset, and (ii) seeking relevant attention in recent years [9, 14, 8, 18, 1, 19]. pages starting a few links away from the relevant Topic driven crawlers (also known as focused subset. Using a framework and a number of qual- crawlers) respond to the particular information ity metrics developed to evaluate topic driven needs expressed by topical queries or interest crawling algorithms in a fair way, we find that a profiles. These could be the needs of an indi- mix of exploitation and exploration is essential vidual user (query time or online crawlers) or for both tasks, in spite of a penalty in the early those of a community with shared interests (top- stage of the crawl. ical search engines and portals). Evaluation of topic driven crawlers is diffi- 1 Introduction cult due to the lack of known relevant sets for Web searches, to the presence of many conflict- A recent projection estimates the size of the vis- ing page quality measures, and to the need for ible Web today (March 2002) to be around 7 fair gauging of crawlers’ time and space algorith- billion “static” pages [10]. The largest search mic complexity. In recent research we presented engine, Google, claims to be “searching” about an evaluation framework designed to support the 2 billion pages. The fraction of the Web cov- comparison of topic driven crawler algorithms 88 1 under specified resource constraints [19]. In this Balancing the exploitation of quality esti- paper we further this line of research by inves- mate information with exploration of subopti- tigating the relative merits of exploration versus mal pages is thus crucial for the performance of exploitation as a defining characteristic of the topic driven crawlers. It is a question that we crawling mechanism. study empirically with respect to two different The issue of exploitation versus exploration is tasks. In the first, we seek relevant pages start- a universal one in machine learning and artificial ing from a set of relevant links. Applications of intelligence, since it presents itself in any task such a task are query-time search agents that where search is guided by quality estimations. use results of a search engine as starting points Under some regularity assumption, one can as- to provide a user with recent and personalized sume that a measure of quality at one point in results. Since we start from relevant links, we the search space provides some information on may expect an exploratory crawler to perform the quality of nearby points. A greedy algo- reasonably well. The second task involves seek- rithm can then exploit this information by con- ing relevant pages while starting the crawl from centrating the search in the vicinity of the most links that are a few links away from a relevant promising points. However, this strategy can subset. Such a task may be a part of Web min- lead to missing other equally good or even bet- ing or competitive intelligence applications (e.g., ter points, for two reasons: first, the estimates a search starting from competitors’ home pages). may be noisy; and second, the search space may If we do not start from a known relevant subset, have local optima that trap the algorithm and the appropriate balance of exploration vs. ex- keep it from locating global optima. In other ploitation becomes an empirical question. words, it may be necessary to visit some “bad” points in order to arrive at the best ones. At the other extreme, algorithms that completely 2 Evaluation Framework disregard quality estimates and continue to ex- plore in a uniform or random fashion do not risk 2.1 Topics, Examples and Neighbors getting stuck at local optima, but they do not In order to evaluate crawler algorithms, we use the available information to bias the search need topics, some corresponding relevant exam- and thus may spend most of their time exploring ples, and neighbors. The neighbors are URLs suboptimal areas. A balance between exploita- extracted from neighborhood of the examples. tion and exploration of clues is obviously called We obtain our topics from the Open Direc- for in heuristic search algorithms, but the op- tory (DMOZ). We ran randomized Breadth-First timal compromise point is unknown unless the crawls starting from each of the main cate- topology of the search space is well understood gories on the DMOZ site.1 The crawlers identify — which is typically not the case. DMOZ “leaves,” i.e., pages that have no children Topic driven crawlers fit into this picture very category nodes. Leaves with five or more exter- well if one views the Web as the search space, nal links are then used to derive topics. We thus with pages as points and neighborhoods as de- collected 100 topics. fined by hyperlinks. A crawler must decide which A topic is represented by three types of infor- pages to visit based on the cues provided by links mation derived from the corresponding leaf page. from nearby pages. If one assumes that a rele- First, the words in the DMOZ hierarchy form the vant page has a higher probability to be near topic’s keywords. Second, up to 10 external links other relevant pages than to any random page, form the topic’s examples. Third, we concate- then quality estimate of pages provide cues that nate the text descriptions and anchor text of the can be exploited to bias the search process. How- target URLs (written by DMOZ human editors) ever, given the short range of relevance clues on to form a topic description. The difference be- the Web [17], a very relevant page might be only 1 a few links behind an apparently irrelevant one. http://dmoz.org 89 2 Table 1: A sample topic. The description is truncated for space limitations. Topic Keywords Topic Description Examples Recreation Aerostat Society of Australia Varied collec- http://www.ozemail.com.au/~p0gwil Hot Air Ballooning tion of photos and facts about ballooning http://www.hotairballooning.org/ Organizations in Australia, Airships, Parachutes, Balloon http://www.ballooningaz.com Building and more. Includes an article on http://www.aristotle.net/~mikev/ the Theory of Flight. Albuquerque Aerostat http://www89.pair.com/techinfo/ABAC/abac.htm Ascension Association A comprehensive site http://www.prairienet.org/bagi covering a range of ballooning topics includ- http://www.atu.com.au/~balloon/club1.html ing the Albuqeurque Balloon Fiesta, local ed- http://communities.msn.com/BalloonCrews ucation and safety programs, flying events, http://www.bfa.net club activities and committees, and club his- http://www.ask.ne.jp/~kanako/ebst.html tory. Arizona Hot Air Balloon Club [...] tween topic keywords and topic descriptions is guide the selection of frontier URLs that are to that we give the former to the crawlers, as mod- be fetched at each iteration. For a given topic, els of (short) query-like topics, while we use the a crawler is allowed to crawl up to MAX PAGES = latter, which are much more detailed representa- 2000 pages. However, a crawl may end sooner if tions of the topics, to gauge the relevance of the the crawler’s frontier should become empty. We crawled pages in our post-hoc analysis. Table 1 use a timeout of 10 seconds for Web downloads. shows a sample topic. Large pages are chopped so that we retrieve only The neighbors are obtained for each topic the first 100 KB. The only protocol allowed is through the following process. For each of the HTTP (with redirection allowed), and we also examples, we obtain the top 20 inlinks as re- filter out all but “static pages” with text/html turned by Google.2 Next, we get the top 20 content. Stale links yielding HTTP error codes inlinks for each of the inlinks obtained earlier. are removed as they are found (only good links Hence, if we had 10 examples to start with, we are used in the analysis). may now have a maximum of 4000 unique URLs. We constrain the space resources a crawler al- A subset of 10 URLs is then picked at random gorithm can use by restricting the frontier size to from this set. The links in such a subset are MAX BUFFER = 256 URLs. If the buffer becomes called the neighbors. full then the crawler must decide which links are to be replaced as new links are added. 2.2 Architecture We use the a previously proposed evaluation 3 Crawling Algorithms framework to compare different crawlers [19]. In this paper we study the notion of explo- The framework allows one to easily plug in mod- ration versus exploitation. We begin with a ules implementing arbitrary crawling algorithms, single family of crawler algorithms with a sin- which share data structures and utilities to op- gle greediness parameter to control the explo- timize efficiency without affecting the fairness of ration/exploitation behavior. In our previous ex- the evaluation. periments [19] we found that a naive Best-First As mentioned before, we use the crawlers crawler displayed the best performance among for two different tasks. For the first task, the three crawlers considered. Hence, in this study crawlers start from the examples while for the we explore variants of the Best-First crawler. second the starting points are the neighbors. In More generally, we examine the Best-N-First either case, as the pages are fetched their com- family of crawlers where the parameter N con- ponent URLs are added to a list that we call the trols the characteristic of interest. frontier. A crawler may use topic’s keywords to Best-First crawlers have been studied before 2 http://google.yahoo.com [9, 14]. The basic idea is that given a frontier of 90 3 Best_N_First(topic, starting_urls, N) { 4 Evaluation Methods foreach link (starting_urls) { enqueue(frontier, link); } while (#frontier > 0 and visited < MAX_PAGES) { Table 2 depicts our overall methodology for links_to_crawl := dequeue_top_links(frontier, N); foreach link (randomize(links_to_crawl)) { crawler evaluation. The two rows of Table 2 in- doc := fetch_new_document(link); dicate two different methods for gauging page score := sim(topic, doc); quality. The first is a purely lexical approach foreach outlink (extract_links(doc)) { if (#frontier >= MAX_BUFFER) { wherein similarity to the topic description is used dequeue_link_with_min_score(frontier); to assess relevance. The second method is pri- } enqueue(frontier, outlink, score); marily linkage based and is an approximation of } the retrieval/ranking method used by Google [6]; } } it uses PageRank to discriminate between pages } containing the same number of topic keywords. Figure 1: Pseudocode of Best-N-First crawlers. The columns of the table show that our mea- sures are used both from a static and a dynamic links, the best link according to some estimation perspective. The static approach examines crawl criterion is selected for crawling. quality assessed from the full set of (up to 2000) Best-N-First is a generalization in that at each pages crawled for each query. In contrast the iteration a batch of top N links to crawl are se- dynamic measures provide a temporal charac- lected. After completing the crawl of N pages terization of the crawl strategy, by considering the crawler decides on the next batch of N and the pages fetched while the crawl is in progress. so on. As mentioned above, the topic’s keywords More specifically, the static approach measures are used to guide the crawl. More specifically coverage, i.e., the ability to retrieve “good” pages this is done in the link selection process by com- where the quality of a page is assessed in two puting the lexical similarity between a topic’s different ways (corresponding to the rows of the keywords and the source page for the link. Thus table). Our static plots show the ability of each the similarity between a page p and the topic crawler to retrieve more or fewer highly relevant is used to estimate the relevance of the pages pages. This is analogous to plotting recall as a linked from p. The N URLs with the best es- function of generality. timates are then selected for crawling. Cosine The dynamic approach examines the quality of similarity is used by the crawlers and the links retrieval as the crawl progresses. Dynamic plots with minimum similarity score are removed from offer a trajectory over time that displays the dy- the frontier if necessary in order to not exceed namic behavior of the crawl. The measures are the MAX BUFFER size. Figure 1 offers a simplified built on average (quality-based) ranks and are pseudocode of a Best-N-First crawler. generally inversely related to precision. As the Best-N-First offers an ideal context for our average rank decreases, an increasing proportion study. The parameter N controls the greedy be- of the crawled set can be expected to be relevant. havior of the crawler. Increasing N results in crawlers with greater emphasis on exploration It should be noted that scores and ranks used and consequently a reduced emphasis on ex- in each dynamic measure are computed omni- ploitation. Decreasing N reverses this; selecting sciently, i.e., all calculations for each point in a smaller set of links is more exploitative of the time for a crawler are done using data generated evidence available regarding the potential mer- from the full crawl. For instance, all PageR- its of the links. In our experiments we test five ank scores are calculated using the full set of mutants of the crawler by setting N to 1, 16, 64, retrieved pages. This strategy is quite reason- 128 and 256. We refer to them as BFSN where able given that we want to use the best possible N is one of the above values. evidence when judging page quality. 91 4 Table 2: Evaluation Schemes and Measures. The static scheme is based on coverage of top pages (ranked by quality metric among all crawled pages, S). Scrawler is the set of pages visited by a crawler. The dynamic scheme is based on the ranks (by quality metric among all crawled pages, S) averaged over the crawl sets at time t, Scrawler (t). Static Scheme Dynamic Scheme |Scrawler ∩ toprankSM ART (S)| P Lexical rankSM ART (p)/|Scrawler (t)| Pp∈Scrawler (t) Linkage |Scrawler ∩ toprankKW,P R (S)| p∈Scrawler (t) rankKW,P R (p)/|Scrawler (t)| 4.1 Lexical Based Page Quality rank pages. PageRank in particular estimates the global popularity of a page. The computa- We use the SMART system [23] to rank the re- tion of PageRanks can be done through an it- trieved pages by their lexical similarity to the erative process. PageRanks are calculated once topic. The SMART system allows us to pool after all the crawls are completed. That is, we all the pages crawled by all the crawlers for a pool the pages crawled for all the topics by all topic and then rank these against the topic de- the crawlers and then calculate the PageRanks scription. The system utilizes term weighting according to the algorithm described in [13]. We strategies involving term frequency and inverse sort the pages crawled for a given topic, by all document frequency computed from the pooled crawlers, first based on the number of topic key- pages for a topic. SMART computes the simi- words they contain and then sort the pages with larity between the query and the topic as a dot same number of keywords by their PageRank. product of the topic and page vectors. It out- The process gives us a rankKW,P R for each page puts a ranked set of pages based on their topic crawled for a topic. similarity scores. That is, for each page we get Once again, our static evaluation metric mea- a rank which we refer to as rankSM ART (cf. Ta- sures the percentage of top n pages (ranked by ble 2). Thus given a topic, the percentage of top rankKW,P R ) crawled by a crawler on a topic. In n pages ranked by SMART (where n varies) that the dynamic metric, mean rankKW,P R is plotted are retrieved by each crawler may be calculated, over each Scrawler (t) where t is the number of yielding the static evaluation metric. pages crawled. For the dynamic view we use the rankSM ART values for pages to calculate mean rankSM ART at different points of the crawl. If we let 5 Results Scrawler (t) denote the set of pages retrieved up to For each of the evaluation schemes and metrics time t, then we calculate mean rankSM ART over outlined in Table 2, we analyzed the performance Scrawler (t). The set Scrawler (t) of pages increases of each crawler on the two tasks. in size as we proceed in time. We approximate t by the number of pages crawled. The trajectory of mean rankSM ART values over time displays 5.1 Task 1 : Starting from Examples the dynamic behavior of a crawler. For the first task the crawlers start from a rele- vant subset of links, the examples, and use the 4.2 Linkage Based Page Quality hyperlinks to navigate and discover more rele- vant pages. The results for the task are sum- It has been observed that content alone does not marized by the plots in Figure 2. For read- give a fair measure of the quality of the page ability, we are only plotting the performance of [15]. Algorithms such as HITS [15] and PageR- a selected subset of the Best-N-First crawlers ank [6] use the linkage structure of the Web to (N = 1, 256). The behavior of the remaining 92 5 a) Static Lexical Performance b) Dynamic Lexical Performance 80 3400 BFS1 BFS256 75 3200 3000 70 2800 65 2600 average rank % crawled 60 2400 55 2200 50 2000 45 1800 40 1600 BFS1 BFS256 35 1400 0 100 200 300 400 500 0 500 1000 1500 2000 number of top pages pages crawled c) Static Linkage Performance d) Dynamic Linkage Performance 65 3100 BFS1 BFS256 3000 60 2900 2800 55 2700 average rank % crawled 50 2600 2500 45 2400 2300 40 2200 BFS1 BFS256 35 2100 0 100 200 300 400 500 0 500 1000 1500 2000 number of top pages pages crawled Figure 2: Static evaluation (left) and dynamic evaluation (right) of representative crawlers on Task 1. The plots correspond to lexical (top) and linkage (bottom) quality metric. Error bars correspond to ±1 standard error across the 100 topics in this and the following plots. crawlers (BFS16, BFS64 and BFS128) can be BFS256 still does significantly better than other extrapolated between the curves corresponding crawlers on the lexical metric (cf. Figure 2b). to BFS1 and BFS256. However, the linkage metric shows that BFS256 The most general observation we can draw pays a large penalty in the early stage of the from the plots is that BFS256 achieves a sig- crawl (cf. Figure 2d). However, the crawler re- nificantly better performance under the static gains quality over the longer run. The better evaluation schemes, i.e., a superior coverage of coverage of highly relevant pages by this crawler the most highly relevant pages based on both (cf. Figure 2c) may help us interpret the im- quality metrics and across different numbers of provement observed in the second phase of the top pages (cf. Figure 2a,c). The difference be- crawl. We conjecture that by exploring subop- tween the coverage by crawlers for different N timal links early on, BFS256 is capable of even- increases as one considers fewer highly relevant tually discovering paths to highly relevant pages pages. These results indicate that exploration that escape more greedy strategies. is important to locate the highly relevant pages when starting from relevant links, whereas too 5.2 Task 2: Starting from Neighbors much exploitation is harmful. The success of a more exploratory algorithm on The dynamic plots give us a richer picture. the first task may not come as a surprise since (Recall that here lowest average rank is best.) we start from known relevant pages. However, in 93 6 a) Static Lexical Performance b) Dynamic Lexical Performance 40 7500 BFS1 BFS256 38 BreadthFirst 7000 36 6500 34 32 6000 average rank % crawled 30 5500 28 26 5000 24 4500 22 4000 20 BFS1 BFS256 BreadthFirst 18 3500 0 100 200 300 400 500 0 500 1000 1500 2000 number of top pages pages crawled c) Static Linkage Performance d) Dynamic Linkage Performance 40 7000 BFS1 BFS256 BreadthFirst 35 6500 30 6000 average rank % crawled 25 5500 20 5000 15 4500 BFS1 BFS256 BreadthFirst 10 4000 0 100 200 300 400 500 0 500 1000 1500 2000 number of top pages pages crawled Figure 3: Static evaluation (left) and dynamic evaluation (right) of representative crawlers on Task 2. The plots correspond to lexical (top) and linkage (bottom) quality metric. the second task we use the links obtained from general result we find that exploration helps an the neighborhood of relevant subset as the start- exploitative algorithm, but exploration without ing points with the goal of finding more relevant guidance goes astray. pages. We take the worst (BFS1) and the best Due to the availability of relevant subsets (ex- (BFS256) crawlers on Task 1, and use them for amples) for each of the topics in the current task, Task 2. In addition, we add a simple Breadth- we plot the average recall of the relevant exam- First crawler that uses the limited size frontier ples against number of pages crawled (Figure 4). as a FIFO queue. The Breadth-First crawler is The plot illustrates the target-seeking behavior added to observe the performance of a blind ex- of the three crawlers if the examples are viewed ploratory algorithm. A summary of the results as the targets. We again find BFS256 outper- is shown through plots in Figure 3. As for Task forming BFS1 while Breadth-First trails behind. 1, we find that the more exploratory algorithm, BFS256, performs significantly better than BFS1 under static evaluations for both lexical and link- 6 Related Research age quality metrics (cf. Figure 3a,c). In the dy- namic plots (cf. Figure 3b,d) BFS256 seems to Research on the design of effective focused bear an initial penalty for exploration but recov- crawlers is very vibrant. Many different types ers in the long run. The Breadth-First crawler of crawling algorithms have been developed. For performs poorly on all evaluations. Hence, as a example, Chakrabarti et al. [8] use classifiers built from training sets of positive and negative 94 7 0.45 long-term gains. Their solution is to build clas- 0.4 sifiers that can assign pages to different classes 0.35 based on the expected link distance between the 0.3 current page and relevant documents. The area of crawler quality evaluation has average recall 0.25 also received much attention in recent research 0.2 [2, 9, 8, 19, 4]. For instance, many alterna- 0.15 tives for assessing page importance have been 0.1 explored, showing a range of sophistication. Cho 0.05 BFS1 et al. [9] use the simple presence of a word such BFS256 0 BreadthFirst as “computer” to indicate relevance. Amento et 0 500 1000 1500 2000 pages crawled al. [2] compute the similarity between a page and the centroid of the seeds. In fact content- Figure 4: Average recall of examples when the based similarity assessments form the basis of crawls start from the neighbors. relevance decisions in several examples of re- search [8, 19]. Others exploit link information to estimate page relevance with methods based on example pages to guide their focused crawlers. in-degree, out-degree, PageRank, hubs and au- Fetuccino [3] and InfoSpiders [18] begin their thorities [2, 3, 4, 8, 9, 20]. For example, Cho et focused crawling with starting points generated al. [9] consider pages with PageRank score above from CLEVER [7] or other search engines. Most a threshold as relevant. Najork and Wiener [20] crawlers follow fixed strategies, while some can use a crawler that can fetch millions of pages per adapt in the course of the crawl by learning to day; they then calculate the average PageRank estimate the quality of links [18, 1, 22]. of the pages crawled daily, under the assumption The question of exploration versus exploita- that PageRank estimates relevance. Combina- tion in crawler strategies has been addressed tions of link and content-based relevance estima- in a number of papers, more or less directly. tors are evident in several approaches [4, 7, 18]. Fish-Search [11] limited exploration by bound- ing the depth along any path that appeared sub- optimal. Cho et al. [9] found that exploratory 7 Conclusions crawling behaviors such as implemented in the Breadth-First algorithm lead to efficient discov- In this paper we used an evaluation framework ery of pages with good PageRank. They also dis- for topic driven crawlers to study the role of ex- cuss the issue of limiting the memory resources ploitation of link estimates versus exploration of (buffer size) of a crawler, which has an impact on suboptimal pages. We experimented with a fam- the exploitative behavior of the crawling strategy ily of simple crawler algorithms of varying greed- because it forces the crawler to make frequent iness, under limited memory resources for two filtering decisions. Breadth-First crawlers also different tasks. A number of schemes and qual- seem to find popular pages early in the crawl [20]. ity metrics derived from lexical features and link The exploration versus exploitation issue contin- analysis were introduced and applied to gauge ues to be studied via variations on the two major crawler performance. classes of Breadth-First and Best-First crawlers. We found consistently that exploration leads For example, in recent research on Breadth-First to better coverage of highly relevant pages, in focused crawling, Diligenti et al. [12] address spite of a possible penalty during the early stage the “shortsightedness” of some crawlers when as- of the crawl. An obvious explanation is that ex- sessing the potential value of links to crawl. In ploration allows to trade off short term gains for particular, they look at how to avoid short-term longer term and potentially larger gains. How- gains at the expense of less-obvious but larger ever, we also found that a blind exploration 95 8 when starting from neighbors of relevant pages, A goal of present research is to identify op- leads to poor results. Therefore, a mix of ex- timal trade-offs between exploration and ex- ploration and exploitation is necessary for good ploitation, where either more exploration or overall performance. When starting from rele- more greediness would degrade performance. A vant examples (Task 1), the better performance large enough buffer size will have to be used of crawlers with higher exploration could be at- so as not to constrain the range of explo- tributed to their better coverage of documents ration/exploitation strategies as much as hap- close to the relevant subset. The good perfor- pened in the experiments described here due to mance of BFS256 starting away from relevant the small MAX BUFFER. Identifying an optimal pages, shows that its exploratory nature com- exploration/exploitation trade-off would be the plements its greedy side in finding highly rele- first step toward the development of an adaptive vant pages. Extreme exploitation (BFS1) and crawler that would attempt to adjust the level of blind exploration (Breadth-First), impede per- greediness during the crawl. formance. Nevertheless, any exploitation seems Finally, two things that we have not done in to be better than none. Our results are based this paper are to analyze the time complexity of on short crawls of 2000 pages. The same may the crawlers and the topic-specific performance not hold for longer crawls; this is an issue to be of each strategy. Regarding the former, clearly addressed in future research. The dynamic eval- more greedy strategies require more frequent de- uations do suggest that for very short crawls it cisions and this may have an impact on the ef- is best to be greedy; this is a lesson that should ficiency of the crawlers. Regarding the latter, be incorporated into algorithms for query time we have only considered quality measures in the (online) crawlers such as MySpiders3 [21]. aggregate (across topics). It would be useful to The observation that higher exploration yields study how appropriate trade-offs between explo- better results can motivate parallel and/or ration and exploitation depend on different char- distributed implementations of topic driven acteristics such as topic heterogeneity. Both of crawlers, since complete orderings of the links these issues are the object of ongoing research. in the frontier, as required by greedy crawler al- gorithms, do not seem to be necessary for good performance. Therefore crawlers based on local References decisions seem to hold promise both for the per- [1] C. Aggarwal, F. Al-Garawi, and P. Yu. In- formance of exploratory strategies and for the ef- telligent crawling on the World Wide Web ficiency and scalability of distributed implemen- with arbitrary predicates. In Proc. 10th Intl. tations. In particular, we intend to experiment World Wide Web Conference, pages 96–105, with variations of crawling algorithms such as 2001. InfoSpiders [18], that allow for adaptive and dis- tributed exploratory strategies. [2] B. Amento, L. Terveen, and W. Hill. Does Other crawler algorithms that we intend ”authority” mean quality? Predicting ex- to study in future research include Best-First pert quality ratings of web documents. In strategies driven by estimates other than lexi- Proc. 23rd ACM SIGIR Conf. on Research cal ones. For example we plan to implement a and Development in Information Retrieval, Best-N-First family using link estimates based pages 296–303, 2000. on local versions of the rankKW,P R metric used in this paper for evaluations purposes. We also [3] I. Ben-Shaul, M. Herscovici, M. Jacovi, plan to test more sophisticated lexical crawlers Y. Maarek, D. Pelleg, M. Shtalhaim, such as InfoSpiders and Shark Search [14], which V. Soroka, and S. Ur. Adding support for can prioritize over links from a single page. dynamic and focused search with Fetuccino. Computer Networks, 31(11–16):1653–1665, 3 http://myspiders.biz.uiowa.edu 1999. 96 9 [4] K. Bharat and M. Henzinger. Improved al- [13] T. Haveliwala. Efficient computation of gorithms for topic distillation in hyperlinked pagerank. Technical report, Stanford environments. In Proc. 21st ACM SIGIR Database Group, 1999. Conf. on Research and Development in In- [14] M. Hersovici, M. Jacovi, Y. S. Maarek, formation Retrieval, pages 104–111, 1998. D. Pelleg, M. Shtalhaim, and S. Ur. The [5] B. E. Brewington and G. Cybenko. How shark-search algorithm — An application: dynamic is the Web? In Proc. 9th Interna- Tailored Web site mapping. In Proc. 7th tional World-Wide Web Conference, 2000. Intl. World-Wide Web Conference, 1998. [15] J. Kleinberg. Authoritative sources in a [6] S. Brin and L. Page. The anatomy of hyperlinked environment. Journal of the a large-scale hypertextual Web search en- ACM, 46(5):604–632, 1999. gine. Computer Networks, 30(1–7):107–117, 1998. [16] S. Lawrence and C. Giles. Accessibility of information on the Web. Nature, 400:107– [7] S. Chakrabarti, B. Dom, P. Raghavan, 109, 1999. S. Rajagopalan, D. Gibson, and J. Klein- [17] F. Menczer. Links tell us about lexical and berg. Automatic resource compilation by semantic web content. arXiv:cs.IR/0108004. analyzing hyperlink structure and associ- ated text. Computer Networks, 30(1–7):65– [18] F. Menczer and R. Belew. Adaptive re- 74, 1998. trieval agents: Internalizing local context and scaling up to the Web. Machine Learn- [8] S. Chakrabarti, M. van den Berg, and ing, 39(2–3):203–242, 2000. B. Dom. Focused crawling: A new approach to topic-specific Web resource discovery. [19] F. Menczer, G. Pant, M. Ruiz, and Computer Networks, 31(11–16):1623–1640, P. Srinivasan. Evaluating topic-driven Web 1999. crawlers. In Proc. 24th Annual Intl. ACM SIGIR Conf. on Research and Development [9] J. Cho, H. Garcia-Molina, and L. Page. Ef- in Information Retrieval, 2001. ficient crawling through url ordering. In [20] M. Najork and J. L. Wiener. Breadth-first Proc. 7th Intl. World Wide Web Confer- search crawling yields high-quality pages. In ence, Brisbane, Australia, 1998. Proc. 10th International World Wide Web Conference, 2001. [10] Cyveillance. Sizing the inter- net. White paper, July 2000. [21] G. Pant and F. Menczer. Myspiders: Evolve http://www.cyveillance.com/web/us/ your own intelligent web crawlers. Au- corporate/white papers.htm. tonomous Agents and Multi-Agent Systems, 5(2):221–229, 2002. [11] P. De Bra and R. Post. Information retrieval in the World Wide Web: Making client- [22] J. Rennie and A. K. McCallum. Using re- based searching feasible. In Proc. 1st Intl. inforcement learning to spider the Web effi- World Wide Web Conference, 1994. ciently. In Proc. 16th International Conf. on Machine Learning, pages 335–343. Morgan [12] M. Diligenti, F. Coetzee, S. Lawrence, Kaufmann, San Francisco, CA, 1999. C. L. Giles, and M. Gori. Focused crawl- [23] G. Salton. The SMART Retrieval System – ing using context graphs. In Proc. 26th Experiments in Automatic Document Pro- International Conference on Very Large cessing. Prentice-Hall, Englewood Cliffs, Databases (VLDB 2000), pages 527–534, NJ, 1971. Cairo, Egypt, 2000. 97 10