Jump to content

Diablo 3 Blue Posts » "Search for similar items" needs some po


STG
It would be more aptly termed "search for identical", because that seems to be what it attempts to do. But I don't think that's what users really want. (I know it's not what I was hoping for.)

It searches too strictly, and more often than not returns zero or very limited results. It's really a toss-up as to whether this feature is a time-saver or time-waster. Not good.

Overall it should:
- work in a more sensible way to provide usable results; try to provide results which are functionally equivalent in the gameplay sense, not parametrically identical in the database sense
- in automatically entering search criteria, start with more relaxed criteria which the user can then "tighten" to lead to more refined results, instead of the other way around, which is how it seems to work right now -- i.e. it enters criteria that are too restrictive often leading to zero or very minimal results, requiring more user intervention to relax them and start seeing some actual helpful results.
- minimize the amount of "reset"ing of search parameters required to do subsequent (manual) searches (by eliminating the inclusion of redundant or unimportant criteria)

Direct suggestions:
-------------------------------------------
Round numeric values down

Trying to match on exact numeric values is too restrictive and leads the tool to often give zero results, which is not really accurate in the practical sense. This is by far the single most important change that would lead to better output from this function.

For example: the way I generally approach searching is to round down very small (~<10) numeric values by 1 or 2, small (~10-30) values to the nearest multiple of 5, medium-sized values (~50-150) to the nearest multiple of 10 or 20, and large values (~>150) to the nearest multiple of 25 or 50. Of course I say "or" a lot, because I apply some situational fuzzy logic based on my output expectations, the item, the affix, etc., but I think there could be a happy medium between this between too-particular, too-automatic-to-be-practically-useful automation and 100% manual labor-intensive searching.

(In a fantasy, I could actually envision users being provided with highly detailed configuration options for just how "loosely" they would like to match any given numeric value, at a given level of magnitude, and on a per-affix basis -- i.e. some kind of config system for specifying rules like I outline in the paragraph above. But it probably wouldn't exactly be considered "panda-friendly", so doubt it would ever happen. Nice to dream tho......)

I tried to "find similar" for a 1-handed axe with socket, 700 avg damage, "168" strength, "948" LOH, and get zero results. But if I relax those values somewhat, to, say, 625, 150, 900 respectively, I start to see 6-7 items, which a little better picture of what is really going on in that space, which helps me assess my item accordingly. (But it's still not a great picture.. :/)

And based on my expectations in experimenting with this feature, I speculate a side-effect of this condition is that when users see these highly refined, highly particular results, they usually contain items which are massively overpriced, which, in turn, is going to cause these users to overprice their items equally ridiculously -- for example the axe above @ 20,000,000g??!! This in turn of course leads to AH stagnation, as it exacerbates the already-rampant problem of users pricing their items at levels that don't reflect their real market value.

-------------------------------------------
Eliminate stats that are too divisive

Similar to the point above. Consider searching as a two-(or more)-phase process: Phase 1 is getting the salient traits of an item represented in the search criteria -- e.g. average damage, crit damage, IAS, main stats, LOH. I don't really care at this point whether the average damage comes from good base rolls, or a weapon damage % affix. Similarly, I don't immediately care what kind of elemental damage the weapon has, if applicable. The system as it currently works will look for an exact match on these types of traits, but in practical use, they are not really important, as far as just ballparking goes. The results are artificially limited, and thus inaccurate; they don't tell the real story.

Now, if, once you have narrowed your results down to a pool of reasonably useful weapons, if you want to refine your results further, to look for cold damage, or lightning damage, or whatever, I would call that "Phase 2" of searching. The automatic fill-in of search criteria via the "search for similar" tool should not include such (refining) criteria by default.

Even the item type is often too restrictive, at least when it comes to weapons. For example: I would appreciate it more if "Search for similar" looked for "all 1-handed weapons", not just "axes". If I can find any one-hander that meets my important standards, I may not even care what type of weapon it is. That, too, is more of a "Phase 2" concern, but right now the system considers it a primary criteria, which, again, "artificially" limits search results and doesn't tell the real story. The inclusion of a single element resist is a similar hindrance. Hiding tons of results that are functionally equivalent, just because they don't have "63 poison resist", for example.

Also, along these lines, I have long thought there should be search criteria for "average elemental damage" (/resist) which search for any kind of additional ele damage/resist respectively, not just a specific element. This would help search for items which are functionally equivalent without needlessly restricting results too much.

In summary: if the user then wants to highly refine their results, let that be the point at which manual intervention is required. Right now, manual intervention is required almost 100% of the time to get any results at all, which basically defeats the purpose of having an automated system to begin with!

-------------------------------------------
Ensure that most-important stats are represented / Eliminate or give low priority to redundant stats.

Because "find similar" tries to include every trait of a weapon, verbatim, it can actually fill all 6 search criteria without even covering the essential traits of the item. In one instance, the system used 3 criteria slots for "average damage", "average (ele) damage", "weapon damage %" -- the other 3 slots were used for relatively important stuff like IAS, socket, and character attributes -- meanwhile discarding the fact that the item also had 175% crit dmg! Not helpful.

I could even make a case for "by default, it should be smart enough to only include one 'main' stat + vit", but I think the AH team has enough on their plate. Of course, such a system, were it to exist, would "intelligently" derive the main stat based firstly on item type (dex for DH weapons, STR for mighty weapons, etc.), and fallback to choosing the one with the largest value, in indeterminate cases.

-------------------------------------------
When restricting by class, it should pick the generic class entry, or at least ensure it picks a Lvl 60 character.

Not sure why, but at least once it chose one of my sub-level-60 characters, which meant I had to change the character criteria manually before performing follow-up searches; I wasn't getting any results because the level of the character it had chosen was too low.

-------------------------------------------
Eliminate the automatic entry for minimum item level / Make better usage of max item level w/r to "Reduced level requirement"

I would wager that almost always, a user only cares that his/her salient search criteria are matched, and couldn't care less what the min ilvl is. If a lower ilvl item somehow winds up meeting my standards, so much the better. I don't see how this helps us get more usable results out of searching.

On a related topic: when it comes to searching for items with reduced level requirement, "Reduced level requirement" should actually NOT be auto-filled as a criteria! Instead, just fill in the "max item level" box appropriately. This method will ensure that any items which are naturally of the target ilvl/clvl, yet just happen to be so fantastic so as to meet the other, non-level-related search criteria, will not be missed just because they don't have "reduced level requirement"

Again, the idea here being -- and hopefully you are picking up on the trend by now! -- to broaden rather than narrow the search criteria, in pursuit of getting more realistic, useful, meaningful-to-the-player-not-so-much-as-the-database-analyst results!

-------------------------------------------

That's all for now. Here's hoping we can get some sensible improved usability out of this feature in the near future! And hopefully this inspires everyone who reads it to extend these approaches -- i.e. pursuit of software that actually helps you instead of confounds you -- to all software you ever encounter, EVER, so that we can avoid a bleak "Terminator 3 Rise of the (not-so-smart) Machines" future for humanity! :P TY for reading.


Lylirra
Community Manager
View Source
Oct 18th, 2012 11:42pm in the General Discussion board on the US forums.
Great feedback! Thanks for sharing; I'll make sure it gets passed on. :)