Suggestions for Advanced Search Features

http://forums.station.sony.com/swg/board/message?board.id=communityrelations&message.id=62

2005/01/19

Here are some features an Advanced Search function should have.

DATE SELECTION

This would return only messages posted between two user-specified dates.

The GUI would consist of two datefields and the appropriate labels. If a date is entered into the first datefield (preferably from a calendar popup box, which would eliminate data entry errors), it is used as a starting date for returned messages -- only those messages posted on or after the given date would be returned by a search. If a date is entered into the second datefield, it is used as a final date for returned messages -- only messages between the two dates (inclusive) would be returned.

Appropriate error handling logic would deal with the case of a date entered into the second datefield but not the first, and the case of a date entered into the second datefield being earlier than a date entered into the first datefield.

UNREAD MESSAGE SELECTION

This would return only those messages which the user has not yet read. By itself it wouldn't be very useful, but in combination with some other specifications it could help reduce the time spent searching for messages the user might want to reply to.

The GUI would consist of an "Unread messages only" checkbox. Checking this box would activate a test that checks whether a message has been read by the user who performed the search. If the check box is selected (checked), and the message has not been read by the requesting user, it is returned; otherwise it is not returned.

This checkbox would be unchecked by default. A nice-to-have feature would be for this checkbox to "remember" the user's preference between sessions.

Another nice touch would be to use the icons displayed next to thread titles on each forum's header page (the open/closed/arrowed "file folder" icons) in front of returned messages. This would be another effectively way to tell the user whether or not she's read a message returned by a search. Currently you can tell whether a message has been read or not by the brightness of the returned message's title, but this can be hard to read. The thread icons would make this information much more obvious.

MULTI-WORD (PHRASE) SEARCHES

This feature would allow searching for multi-word phrases, rather than just conjunctions of individual words.

Currently if you enter multiple words into the "With all the words" text field on the Advanced Search page, you get every message that has all of those words no matter where they occur in the message. Sometimes that's what you want... but usually it's not.

What I'm looking for is an addition to the Advanced Search GUI that would be structured as the label "With the exact phrase" and a text field. Anything you enter into this exact phrase text field would be searched for within messages as (surprise!) a single phrase exactly as the user entered it, spaces and all.

Consider the current "With all the words" field. If you enter "smart bomb" into this field, a message that includes the following sentence would be found: "They made a movie from Get Smart called 'The Nude Bomb'." The thread containing this sentence almost certainly has nothing to do with what the person who entered "smart bomb" was really looking for because the text entered in the search field is not being treated as an exact phrase. Any message that has those two words anywhere will be returned.

With an exact phrase search, the above sentence would not be found, but "They dropped a smart bomb down the chimney" would be found, which is probably part of a thread much more closely related to what the searcher really wanted to find. This whole-phrase search capability would be a huge help in cutting down on the number of completely irrelevant messages returned from a multiword search.

IMPROVED RELEVANCE SCORE CALCULATION

The calculation for relevance scores in search results should working properly. If a message is scored at 99% or 100%, it should be extremely close to what the user was searching for.

Multi-word searches should have a much broader range of result scores. If I search on "vehicle speed" in the "With all the words" text field, messages that have the exact phrase "vehicle speed" ought to be scored at 100%, while messages with both the word "vehicle" and the word "speed" separated by other words ought to receive scores considerably lower than 100%. In an exact phrase search, only messages with the exact phrase "vehicle speed" somewhere in them would be returned at all, so every message returned should have a relevance score of 100%.

In a "With zero or more of the words" search, messages containing both of the submitted words exactly as entered ("vehicle speed") would be scored at 100%; messages with both the words separated by other words would be scored between (say) 50% and 80%; messages with either but not both of the words would be scored at between 20% and 50%, and messages with neither of the words would be scored at 0% (and thus would not be returned at all).

This scoring calculation doesn't have to use fuzzy logic (although that would be effective). A simple distance calculation would be far superior to the completely dysfunctional scoring system we have now.

It also doesn't need to use Soundex codes, although that might be useful as well.