### Table 2: Results of topical clustering (A/CDC), topological clustering (IHS) and their hybrid, on two datasets. The IHS clustering (and the hybrid) results are obtained after the first and second iterations of heuristic search (hyperlink paths of lengthupto4andupto6respectively).

"... In PAGE 5: ... To compare our results with the ones reported by Bekker- man and McCallum [2005], we use their topical clustering method called Agglomerative/Conglomerative Distributional Clustering (A/CDC), which is a state-of-the-art information- theoretic technique. The results are shown in Table2 (A/CDC result is by Bekkerman and McCallum, 2005). We see that after the first iteration the heuristic search method is compet- itive with A/CDC in precision, but is inferior in recall.... In PAGE 5: ... This is the only result we could obtain that shows usefulness of expanding pages at the third hop. When comparing the heuristic search method with topi- cal clustering, we observe exactly the same trend as for the Web appearance disambiguation task (see Table2 ). The best performance (77.... ..."

### Table 2: Results of topical clustering (A/CDC), topological clustering (IHS) and their hybrid, on two datasets. The IHS clustering (and the hybrid) results are obtained after the first and second iterations of heuristic search (hyperlink paths of length up to 4 and up to 6 respectively).

"... In PAGE 5: ... To compare our results with the ones reported by Bekker- man and McCallum [2005], we use their topical clustering method called Agglomerative/Conglomerative Distributional Clustering (A/CDC), which is a state-of-the-art information- theoretic technique. The results are shown in Table2 (A/CDC result is by Bekkerman and McCallum, 2005). We see that after the first iteration the heuristic search method is compet- itive with A/CDC in precision, but is inferior in recall.... In PAGE 5: ... This is the only result we could obtain that shows usefulness of expanding pages at the third hop. When comparing the heuristic search method with topi- cal clustering, we observe exactly the same trend as for the Web appearance disambiguation task (see Table2 ). The best performance (77.... ..."

### Table 2: Results of topical clustering (A/CDC), topological clustering (IHS) and their hybrid, on two datasets. The IHS clustering (and the hybrid) results are obtained after the first and second iterations of heuristic search (hyperlink paths of length up to 4 and up to 6 respectively).

"... In PAGE 5: ... To compare our results with the ones reported by Bekker- man and McCallum [2005], we use their topical clustering method called Agglomerative/Conglomerative Distributional Clustering (A/CDC), which is a state-of-the-art information- theoretic technique. The results are shown in Table2 (A/CDC result is by Bekkerman and McCallum, 2005). We see that after the first iteration the heuristic search method is compet- itive with A/CDC in precision, but is inferior in recall.... In PAGE 5: ... This is the only result we could obtain that shows usefulness of expanding pages at the third hop. When comparing the heuristic search method with topi- cal clustering, we observe exactly the same trend as for the Web appearance disambiguation task (see Table2 ). The best performance (77.... ..."

### Table 1: Comparison of Backtracking and Heuristic Search

1998

"... In PAGE 5: ... Using the settings (5 j j 85 ; = f0:2; 1:0g), the backtracking (A) and heuristic (B) searches were run on all 300 polygons. The total number of failures (A-fail, B-fail) and total number of states (A-states, B-states) examined to find solutions are given in Table1 . The heuristic search was stopped after 500 iterations.... ..."

Cited by 1

### Table 1: Results of the k-opt local search and the state-of-the-art heuristic algorithms (RLS by Battiti and Protasi, GENE and ITER by Marchiori) for the DIMACS benchmark instances

"... In PAGE 5: ... However, since the k-opt local search performs a random selection when tie-breaking choice, it is expected that the final solution or search is different even if the same initial solution is given for the local search. Table1 summarizes the results of the multi-start k-opt local search for large or hard 15 graphs selected from the well-known 37 benchmark instances studied in [1] and [6]. In the table, we show the best clique size found in the 10 runs of the multi-start method, the average clique size with standard deviation (s.... ..."

### Table 3: multistart heuristic

2005

"... In PAGE 16: ... The reason for this may be that the search is restrictive in the sense that it is hard to move vertices around without violating the constraints. Table3 shows the nal results obtained by the second method, multistart. The rightmost column labeled is the number of iterations made by the heuristic during the time provided by tabu search.... In PAGE 16: ... This number provides us with the number of local minima encountered. As with Table 2, Table3 also shows that results improve with an increasing value of t, but this is not the case for instances pr3a-pr3c. Solutions to instances pr3b and pr3c have the same number... ..."

### Table 8: Comparison between different search heuristics

"... In PAGE 31: ... However, a new covering design is always found. Table8 compares simulated annealing, our one level tabu search and multilevel cooperative search. Tests have been conducted on problems for which multilevel cooperative search found new upper bounds.... In PAGE 31: ... Tests have been conducted on problems for which multilevel cooperative search found new upper bounds. Tabu search and simulated annealing procedures are given the same amount of CPU time (in seconds in Table8 ) as multilevel cooperative search using computers with same clock rates. We report the number of iterations2 performed by each procedure as well as the number of t-subsets that have not been covered.... In PAGE 31: ... We report the number of iterations2 performed by each procedure as well as the number of t-subsets that have not been covered. Results in Table8 show that the tabu search procedure is very competitive. These results also confirm that multilevel cooperation substantially improves the performance of tabu search.... ..."

### Table 9 shows the performance of our local search heuristic for different numbers of iterations within phase (II). DEV denotes the percentage deviation from the optimal objective function value (precedence-based lower bound) for the 10-activity (30-activity) instances. CPU denotes the computation time in CPU-seconds. avg, stddev, and max denote the average, the standard deviation, and the maximum value of DEV and CPU, respectively.

1997

"... In PAGE 18: ... Table9 : Effect of the Number of Iterations on the 10-Activity Instances iterations 10 50 100 500 avg 27.92 24.... ..."

Cited by 10

### Table 3 provides the comparison the average results obtained by three other state of the art approaches on these benchmark problems, namely GDA where SD is used to provide the initial solution [7], GDA where initial solutions drive the adaptation of the parameters of the algorithm throughout the search [8, 9], GDA where the combination of SD, MCD and BT is used to construct an initial solution (this heuristic is proposed in [12]). Again GDA was run for 200*106 iterations, because that was the number of iterations used in the approaches that we compare our approach with. For illustration purpose, we also give the time spent on the search given in seconds. Although each algorithm was allocated the same number of

2003

"... In PAGE 14: ... The price to be paid is of course longer the time spent on case retrieval, which is proportional to the number of cases. Table3 . Comparison of results for benchmark problems obtained by different initialisation of ... ..."

Cited by 4

### Table 7 shows larger instances where n1 = n2 = n3 = 10. RandGen was expected for 500 iterations and GraspGen was executed for three outer iterations due to the extreme time requirements of the local search component. Figure 7 shows that GRASP is again competitive for longer runs, but dominated for shorter CPU times. The table indicates that the deterministic heuristics are much faster, but not reliable.

2003

"... In PAGE 20: ...8 / 7110 (time to nd sol.) (2477) (5085) (2214) (2767) (7110) Table7 : General Costs Obj. / CPU (sec) for 10 10 10, j j=30, =20, B(1) = 2, B(2) = 4 0 0.... ..."

Cited by 2