How does relevancy ranking work?
AquaBrowser ranks search according to information about the matching search terms in the results and the fields they occur in. The weights for all occurrences of search terms in a record are combined, and the result is used to rank the results. The weighting of search terms takes into account: 1. which field(s) a term appears in, for example, an occurrence of a search term in the title is considered more significant than a term in a note field; 2. how significant the term is in the entire database, according to its term frequency/inverse document frequency ; and 3. whether the matching term in an exact match or an algorithmically-determined spelling variations, exact matches being given more weight than spelling variations.
When running a job search the default search result is to order the results most relevant to least relevant. • Relevancy is determined based on terms entered into the search. • Those terms are weighted based on the frequency that they appear in all the jobs in the database. • Infrequently used words have a higher weight than frequently used words causing jobs that match the infrequently used word to have a higher relevancy. An example of this would be an “any word” keyword search for “nanotechnology analyst.” The jobs with the word “nanotechnology” would be returned first, because that word has a higher ranking than “analyst”. Because this was run as an “any word” search, jobs with just the word “analyst” would follow the nanotechnology jobs. I have a link to a job and when I click on it, your system says it doesn’t exist. Employers control how long each job stays up. A job could be up for a day or for months. Please print the job if you need it for future reference. We recommend check
The ranking all happens in lib/rank.pl. This piece of Perl takes the query and the templates handles that have been matched by the search mechanism. It first removes all booleans, WHOIS++ constraints, punctuation and other guff from the query string. It then cycles over all the templates handles and looks for occurances of each term in the munged query in each line of the template. A counter is incremented for each match and at the end of the template the handle is placed in a hashed array with a key equal to this count (there maybe handles with this key already; if so the handle is appended to the existing list). Once all the templates have been checked, a normal array is built by doing a reverse sort on the keys of this hashed array (so that the largest counts, i.e. most hits, come first) and pushing the handles from the hash array’s values on the normal array. This array is then returned for use by the rendering routines. Note that if you wish you can put your own ranking algorithm
