Help us make search produce better results

The clearest suggestion I can make is to allow the user to choose between 'most relevant results' and 'most recent results' and default to the latter.  There have been some embarrassing bumps lately to years old thread because of this choice.

 That's a good one Sam, thanks. Any other thoughts?

Not from me, others might tho I hope!

Just wanted to update you... we're looking into a search that will hopefully produce better quality results than recency + keywords... after testing this, the results weren't very good :( That sorting is already available in the advanced search area, too, so we're hoping to come up with something a bit more intelligently weighted.

As Senor Gruber would say: "Good luck with that"

I wish I had more knowledge to make good suggestions for how to improve the search engine...but I know very little about how they work.

One issue I always have as a user is whether or not the old boolean (sp?) parameters are still in play (AND, OR, etc.)  I should probably just forget those things...but if I enter two keywords, in a title-only search, I inevitably gets pages of results of threads which only have one of the words in the title and wonder if AND would have helped.  Under that last hypothetical I always figure the threads with both words in the title should be listed first (after all, they would be more "relevant", right?!?)  but it doesn't always seem to work out that way

Most of the results I have gotten from searches in the new software have been, for some reason, weighted to older threads when relevancy is the key parameter.  I am usually looking for a much more recent thread and end up with results starting in 2003. 

If I get less-than-satisfactory search results in the future I will try to remember to post those results here. 

Formatting issues can really trip up the search. For instance, show threads in the TV forum that put quotes around the show name in the title often won't show up in the search for you search for that show. Sometimes the thread for the latest "Monk" season falls back a few pages. When I search for Monk, it should bring up the season 8 thread. But neither that thread nor any of the season threads for the show are on the first page of results.

Greetings. Search appears to be broken. It doesn't return links to the terms being search; only links to the first page, first post of threads that presumably (but who knows) have the terms in them.

For example, I'm looking for discussions about "Lost" where people discuss "retcon" in the plot. So I search for "Lost retcon". Search returns links to the beginning of threads, none of which actually have both "Lost" and "retcon" in them. There are no links directly to the posts or even pages with these terms.

How could search be better? Return results with the search terms.

How should Search should work? That's a black art, and I think that asking us is like asking us what might go into a quality cryptology system :) We really don't know. But I'll make some plausible suggestions, which are probably all wrong, since Google has already shown that you need deep knowledge to make a great search system:

* Newer is better than older
* More replies are more important than fewer replies
* Titles are more important than user comments (for search terms)
* Search terms in close proximity is more important than search terms being far apart
* Posts with a quote that has the search terms are more important than posts without the terms quoted (that is, if it's likely an answer to a question about the terms being searched, it's good)

In terms of Interface, if that's the questions, here's some ideas.
* Showing search results preview to give context can help user know if search results are actually worth following
* Visually cluster all results from a single thread, showing them in chronological order (or reverse chron)
* Visually cluster all results from a single forum

These are critical (and sorely missed):
* In-thread "Search this Thread"
* In-forum "Search this Forum"
* Top level "Search all Discussions" and/or search wikis, reviews, etc.

I rarely use Advanced Search. For me, Google and vBulletin showed such a thing is largely not needed.

But most basically, I want search results linking to posts or pages with the search terms. Particular interface and results, again as google shows, is not that important. Or rather, the lesson is that if you do great search, all you need is an input field and output links listed in priority. (And people are welcome to tell me why I'm wrong. You've got domain knowledge here. And you don't have cross-linking, which Google builds on. Etc. That's fine.)

Here's another suggestion: Google Site Search

I was just using site-specific search to look for HTF discussions. I'm not sure it's better than the old vbulletin system, particulary with a search-this-thread tool, but it works, unlike the current search system (at least today. perhaps something is turned off by accident. I recall having better success with the tool previously)


EDIT

I see the issue. Search only returns the start of threads, and not the location of specific results. Searching for anything that is mostly found in single-page threads is fine.
Example: Onkyo Receiver Heat

But, alas, it's worthless for anything related to multi-page threads
Example: Lost Retcon

A question on search:
Searching multiple terms will return results without one or more of those terms. For example, search "itunes xbox" gives many results that don't have "xbox" anywhere in them. Trying "itunes and xbox" doesn't change this.

Is there a way to force an "AND" search in the site search? That is, to find all results that have both "itunes" and "xbox" in the thread (or display page or post...)? Thanks!

I brought this up in Post #7.  Well, the idea of adding the AND ourselves. 

I was wondering the other day about the idea of being able to sort the results after-the-fact:  by relevancy, date, etc.

Weighting of search results really frustrates me on every site that does it. I want results ordered by date every time. I am smart enough to determine relevancy on my own.

Once again, I wanted to search for a thread I participate in, so I search on two terms "refinance davef" to find it. But I get strange results. The top results don't have "refinance" anywhere in the thread. And it doesn't have the thread I want anywhere in its 42 results.

So I do this and the thread I want is #1 result:site:hometheaterforum.com refinance davef

Until HTF gets its search tool working, please (please!) consider implementing a simple Google search tool.
Edited by DaveF - 11/21/09 at 9:15am

Thanks for the feedback here... we're collecting this feedback and looking into all kinds of improvements on search. Although it may seem simple when you use it, google search (and every other search) relies on weighting. google weighs dozens of factors on page rank, relevancy, recency, etc. We're trying to improve ours, but just wanted to say it's not a straightforward ordering.

Right now, we're showing keyword match and density... that's the only dimension. There is no other weighting occurring. I hope that helps in terms of using it more effectively for the time being until we can tune it more.

What's perplexing, as a user, is that the search results don't suffer from being misordered or non-prioritized, so much as simply weird.

Take my "refinance davef" search. #2 result is a thread on the video game "Beyond Good and Evil". It doesn't have "refinance" anywhere in the thread. I've got no idea how this is the #2 result: it it has no relationship to my search query.

Getting lots of undesired results that do have all the search terms can allow that I can understand the search process, adjust my inputs, and get the answer I want. But when the results have no relationship to my query, I'm lost. I'm dealing with an irrational system; I've got no guidance for how to adjust my search query and I can only give up and not use it.

I think it may be better to have a stupid search system than one that is almost, but not quite, smart. Better a dimwitted assistant than a mad scientist helping me out :)

(I've no doubt search is hard. That's why I suggest using Google. Not trying to be like Google, but literally using Google Search for HTF. Perhaps just until you get an HTF-tuned search going.)
Edited by DaveF - 11/23/09 at 9:38am

I think I know the reason... it's keyword density that's making this thread show up. In other words, because you're the primary poster and your ID shows up in two URLs in such a small thread (in other words, the ratio of the word DaveF to the total number of other words in the thread) makes it hit very high on density.

You're totally right... this is not a desired outcome, but I think it's consistent with keyword density results.

I think the quickest way to improve the search (for the short term) is to add the "AND" functionality as mentioned previously. In fact AND should be the default. Even with keyword density, by ensuring that all search terms must show up in any given result will significantly improve the quality of the search.

Frankly simple "OR" searches (where only one search term needs to be present) are the least useful. But, unfortunately it is the most common on forums and various places around the internet.

After this has been implemented, then it would be worthwhile to investigate other search result options than simple keyword density. More advanced searching could include "OR" (like it is now) and more complex queries with a mixture of AND/OR and even NOT.

The ultimate goal would be to have a system where a complex query like this could be entered:

"desired phrase" AND ((this that) OR (these those)) NOT something

Note (this that) is equivalent to (this AND that) since AND filtering should be implied.
Note parentheses controlling precedence.
Note "desired phrase" is in quotes so these two search terms must be together to be included in the result.

Another nifty filter command to use is "NEAR". Search terms must be close together (say within 8-10 words or less).
For example: ham near radio will be less likely to include results about baking hams and listening to the radio.

But, like I said, at least for now, simply making "AND" implied will greatly improve the results right there.

Thanks for this suggestion, Bill. I'll definitely forward it along. To be honest, I believe adding operators like 'and' and 'or' are already on the list being considered for how to evolve search, but there's no harm in passing it along even if it is.

The obvious difference between search results on Google and search results here is that the search results on Google are pulled from countless sites with content posted at different times in different countries. It wouldn't be practical or possible to order them by date. Google also has to worry about fraud in their search results. What could be more relevant than date on a forum?

Google search can be used to search a specific site. It's commonly used at various websites for their internal search (rather rolling their own, inferior search system. It's also trivially done from Google itself, e.g. this search of HTF to find this very thread:
site:hometheaterforum.com better search

Currently HTF search is non functional. Even temporary use of Google Search would be a massive improvement. Until something is implemented, I'm using Google-site-specific search as above.

 I'd like to search thread titles.  Am I missing something?

I tried searching for "star wars holiday special" (which by the way, is a true piece of comedy gold that lucas won't release).

Search results were horrible.  

I got dozens and dozens and dozens of Star Trek links; several Harry Potters, and a few Inglorious Bastards, Public Enemies and Vertigo links for good measure.  Gave up after wading through 7 or 8 pages.

That's actually pretty good. I'll have to remember to try that instead next time I want to use search.

I see that Search has had a silent update for implicit "AND" functionality. Thanks. My previous examples that gave bad results are working better now.

 Came here today to ask/complain about SEARCH, so I'll do it in this thread.
Once upon a time, when I could get around HTF pretty well, I would go to "Advanced Search", type in a key word (the word today was "WIRE") and specify "Topic Header" (well, it was something like that) as opposed to "POSTS".
I could, and still can, limit the search to TV Shows.
Just now, I did not find any Topic Headers selection, so of course I got lots of useless stuff.

Suggestions?

Henry:  Just to the right of the field where you enter the "Keywords," there is a box you can click labeled "Titles Only."

I think that's what you're looking for.

Good luck.

Mike

 Yep, that's it!

Someone must have JUST added that.