Efficient Autocomplete for Persons

Working principle for person search via autocomplete A one-step infix-capable auto-completion is offered javascript (todo: check what is used in bibsonomy or use http://jqueryui.com/autocomplete/#custom-data ). This means a user may Enter a first name and the beginning of some word in the title of a thesis and will directly get a list of thesis entries containing both the first name and one of the most likely completions of the word that has been begun. Internally this is handled by a two-step process:

a suggestion-tree containing all words and names returns the N most frequent completion of the word that the user is currently typing
lucene is queryed N times - ie once for each completion. Each query also contains the words that already have been completed by the user.

For limiting to one displaye publication per user the following options are possible:

* Option 1: What is queried is the (almost) existing index on resources(posts), as each person is identified by a publication. However, the index will require an additional marker-field for “qualifying a user” (as > 50 in the “qualifying” field of pub_person). For persons without publications, additional empty posts are required (which will not be found by existing lucene-queries if filtering by not(qualifying=52)
* Option 2: create a separate lucene index for the (resource, pub_person, person)-join where qualifying > 50

Comments (6)