3. What are People Looking For?

Out of the 1.5 million hits recorded on this site over the one year interval, 393,467 of them were to primary html and shtml files. This latter count more accurately represents the number of mouse-clicks performed by users than the total hits, since most of the remaining hits are on image files associated with these pages. Therefore, we will limit further analysis to hits on these files.

Because user ID information is not logged on our WWW server, it is difficult to judge exactly how many different people accounted for these hits. Based on the roster of machines which accessed the site, I estimate that roughly 40,000 different people paid a visit during this 10 week study. Some fraction of hits came from webcrawler robots instead of human users, however I believe they had only a minor effect on our statistics. Observe that the least frequently clicked shtml file (containing the copyright notice for the site) was hit only 298 times versus 17,733 hits for the most frequently accessed page (the front page).

 

Table 1: Hits by Major Section Index

Problem Category	Index Hits	Subsection Hits	Problems
Data Structures	6698	14492	6
Numerical Problems	4499	12812	11
Combinatorial Problems	3419	10114	10
Graph Problems: Polynomial	4496	17492	12
Graph Problems: Hard	3849	13447	11
Computational Geometry	7031	25129	16
Set and String Problems	3003	10776	9
Totals	32995	104262	75

Table 1 reports the number of hits distributed among our highest level of classification - the seven major subfields of algorithms. Two different hit measures are reported for each subfield, first the number of hits to the menu of problems within the subfield, and second the total number of hits to individual problem pages within this subfield. Computational geometry proved to be the most popular subfield by both measures, although outweighed by the interest in graph problems split across two subtopics. Data structures recorded the highest ``per-problem'' interest, but I was surprised by the relative lack of enthusiasm for set and string algorithms.

 

Table 2: Hits by Programming Language Index

Programming Language	Index Hits	Implementations
C language	4776	37
C++	5397	11
Fortran	868	6
Lisp	698	1
Mathematica	652	3
Pascal	1556	5
Totals	13947	63

Table 2 reports the number of hits distributed among the various programming language submenus. C++ seems to have supplanted C as the most popular programming language among developers, although there is clearly a lag in the size of the body of software written in C++. C remains the source language for over half the implementations available on the Algorithm Repository. User interest in Mathematica rivals that of Fortran, perhaps suggesting that computer algebra systems are becoming the language of choice for scientific computation. There was no submenu associated with Java, reflecting what was available when I built the repository. The total number of implementations in Table 2 is greater than 56 because seven codes are written in more than one language.

 

Table 3: Most and least popular algorithmic problems, by repository hits.

Most Popular Problems	Hits	Least Popular Problems	Hits
shortest-path	3660	generating-subsets	854
kd-trees	3198	edge-coloring	817
dictionaries	3022	satisfiability	792
traveling-salesman	2963	independent-set	789
convex-hull	2963	cryptography	786
nearest-neighbor	2936	text-compression	767
minimum-spanning-tree	2922	maintaining-arrangements	727
voronoi-diagrams	2815	set-packing	703
triangulations	2786	planar-drawing	678
sorting	2734	median	671
graph-data-structures	2596	bandwidth	629
string-matching	2304	factoring-integers	628
suffix-trees	2213	shortest-common-superstring	520
priority-queues	2208	determinants	520
geometric-primitives	2162	feedback-set	483

Table 3 reports the 15 most popular and least popular algorithmic problems, as measured by the number of hits the associated pages received. Hit counts for all of the 75 problems appears in Table 6. Several observations can be drawn from this data:

• Shortest path (with 3660 hits) was the most popular of the algorithmic problems over the course of the study. Much of this interest was no doubt from students in algorithms courses seeking an edge. However its course-partner minimum spanning tree proved less popular, finishing in seventh place with 2922 hits.

• Of the six data structure problems, only set union-find (number 31) failed to make the top 15 cut.

• There is more interest in kd-trees (3198 hits) than nearest-neighbor searching (2936 hits), even though the former is most often used as a means to solve the latter.

• People seem much more interested in generating permutations than subsets (1474 hits to 854 hits), presumably reflecting the perceived difficulty of the task.

• Surprisingly popular problems include traveling salesman (number 4), suffix trees (number 13), the knapsack problem (number 18). These might reflect educational interest, although Table 5 shows that almost twice as many total .com hits were recorded than total .edu hits.

• Surprisingly unpopular problems include independent set (number 64), planar drawing (number 69), and satisfiability (number 63). Such obviously commercial problems as cryptography (number 65) and text compression (number 66) proved unpopular presumably because better WWW resources exist elsewhere for these problems.

• My users (people seeking programs) rated NP-complete problems substantially differently than users of the compendium of optimization problems [#!CK-98!#] (people seeking references). Our five most popular hard problems, in order were traveling salesman (2963 hits), knapsack (1926 hits), graph coloring (1847 hits), Hamiltonian cycle (1681 hits), and bin packing (1496 hits). Their most popular hard problem was vertex cover, which with 876 hits ranked 58th of all problems on our list.

It is interesting to note that only 17,733 hits occurred to the front-page of the site, which suggests that most visitors never saw the main index of the site. This implies that most users initially entered the site through a keyword-oriented search engine, and gives credence to the notation that these hits measure problem interest more than just directionless wandering through the site.