Home Depot Finds DIY Success with Vector Search | Hacker News Reader

Home Depot Finds DIY Success with Vector Search

datanami.com

138 pointsgk189 comments

I’m a sales associate at Home Depot - I have to look up items all the time. I usually just google “item Home Depot”

This usually gets me the item or a similar item. If the item is similar and not what they want, I have a tree categorization and I can usually go up a level. The interesting problem w the Home Depot I think is figuring out how to categorize and when to make a category. For example “screws” is an obvious category in which there will be sub-categories and parameters on which to filter like “length” or “color” (I use this a lot). What filter parameters to have is also interesting as is how to display the results. Most items are returned in a list form but, annoyingly, some types of items have their own splash page that is kind of unique to that item.

If the item is weird, it might be really hard to categorize, usually these go into _____ accessories. Overall I think Home Depot does a good job w the search except the UI is laggy as someone mentioned and as they also mentioned the location setting is ALWAYS WRONG. FIX THIS NOW! Every. Single. Time I have to check if an item is in stock at my store, it puts me in the wrong store and I have to manually change it myself. Am I just a moron or is there really no way to set it and forget it?

The store location selector forgets all the time which you had selected last. What's best is it appears to try to infer location from ip address, so when you are in a home Depot on a mobile, it picks a different store from the one you are standing in.

I think a lot of stores incorporate a "cheater" mechanism where if you're on the store wifi they use that as the location. You could pick it up via IP or hardware MAC (even if you are on a VPN the local router still knows your MAC from when you passed the ToS accept portal).

Doesn't help cell service of course, and cellular data clients are REALLY tricky to geolocate properly. Cell towers are often trunked a considerable distance (apparently?) so your IP can indicate you're quite far from your physical location.

Pages really rely on the browser to cue location in that scenario, but of course many of us turn it off, and at that point you get what you get.

Completely agree, without precise location, they are relying on unreliable signals. You should only have to tell it once explicitly though, they don't seem to keep the cookies around for the store selection.

Our HD can’t get their internet to work, so we have to use our own data to look shit up. Idk how cellular data works but I’m usually some store in Chicago (despite being in Madison), it seems to vary depending on where I am in the store, though I may be making connections that aren’t there.

You would probably find it interesting to take a look at McMaster-Carr's website for an example of the hardware-store UX done right.

They do the categorization thing really well, there's "catalog pages" that reference item categories, and item categories can be attached to multiple different catalog pages, and items are categorized by tons of different metrics (length, diameter, thread pitch, head shape, material, color...).

And while that's a pretty generic statement - there's lots of stores that let you filter by material or color, after all - it's all executed SUPER well. I often use it just to get a broad idea of what hardware actually exists/what the terminology I need to search for elsewhere.

https://www.mcmaster.com

I knew someone who worked on their system and apparently it's all COBOL on the backend too lol. I'm sure there's some modern tech on the frontend but I bet all the inventory stuff/etc and the warehouse terminals are running on COBOL.

I have never worked even close to any kind of industrial procurement, and I still know about McMaster-Carr, just because of how much people love using their website.

That's some good UX right there.

I'm sure it's a massive boost to their bottom line too. Once you get people on the site, and once they start putting together their order, odds are good they're not going to go elsewhere. The UX literally makes the sale for them.

(of course not every shopping cart converts to an order, and people still order things you don't have elsewhere, etc. But getting people on the site and engaged is half the battle, assuming you don't give them a reason to shop elsewhere.)

I agree that their website is great. For the olds like me, we had just as big a love affair with their printed catalog before the website. Their shipping was on a different plane also, long before e-commerce changed the shipping game. They have always taken UX seriously. I had an engineering internship where the president had a saying: "If you can't design a machine from the McMaster-Carr catalog, you aren't an engineer."

How about rockauto.com?

I mean, I can drill down to exactly what I need easily.

> How about rockauto.com?

It looks old and outdated but it's so easy to use

Weirdly, they don't sell whole replacement transmissions? I get they're heavy and the core return thing, but it's a big-ticket item and it'd be nice to be able to price-shop the local mechanics on it.

Not only the website is perfect, I’d say the same about delivery and customer service.

Uhh on McMaster I think I need a flow meter with range around 0.3 L/min since I’m designing something new and not replacing, and there are separate check boxes for 0.25-0.5, 0.2-1, 0.1-0.6, … etc. with 3 different units mixed in.

If you know exactly what you need, then yes. But if you’re messing around on home depot needing like a 3 7/8 “ish” screw but-maybe-a-4-will-do the McMaster system would stink.

While true, this is unfortunately not unique, I see the same issue on (eg) Mouser/Digikey/Arrow for electronics parts.

Obviously in some cases it can be replaced by a range slider, but there's also cases where it really can't. In some cases it's the part itself that has the range - like let's say a pump is designed to produce a flow rate of 0.5-5 gal/min, you can't express that with the usual range slider.

I guess what you really need is an "inverse range slider" where instead of saying "I want a connector that carries at least 5 amps but not more than 15" you say "my project needs a flow rate of 2.5 gpm, what parts have that value within their range".

But I mean... compare the McMaster-Carr site to Home Depot, or Target, or whatever, and it's just a breath of fresh air. Like, if inverse-ranges are the biggest thing you can complain about, mission accomplished ;)

Flow meters need an operations range though. So any of those ranges would work for your application.

But trying to find a screw on Home Depot or Fastenal .. ugh

I would be surprised (but not totally amazed) if they had fixed it in the hour since your comment was posted but...

I now see separate selectors for maximum flow, minimum flow, and graduations here: https://www.mcmaster.com/flowmeters/

This. I recently wrote something about why I love McMaster.com so much

https://www.bedelstein.com/post/mcmaster-carr

I talk to random people outside of tech who love that site too. I did a bit of googling to get any details on how it came about and came up with this:

https://www.quora.com/Who-designed-UI-UX-and-developed-the-w...

It was heavily influenced by Tufte!

I think they achieved mainstream notoriety due to Mythbusters.

I learned about them through word-of-mouth through various "maker" types of people.

Can confirm as an ex-mcmaster-carr software developer that they put a lot of effort into their websites UI/UX

I appreciate companies like this so much— they sell widgets, so they invest in software and do it right. I worked in the printing industry circa 2009 and the desire to fix antiquated systems was split (and rare). The companies that did are still around; the ones that didn't didn't make it.

I wish someone at Home Depot would fix come of their website's basic usability issues. The layout shifts, bad location UI, and slow speed really sour the experience.

Interestingly, I recently did what I thought was a basic search for a 2'x4' piece of birch plywood and the results were pretty wonky. I ended up drilling down through menus instead.

If you contrast this with McMaster Carr's website, you'll wonder what went wrong at HD. I've been ordering online from McMaster for probably close to two decades, and it's better in every way from speed to interface to product specs.

I think the problem is the vendors loading different information into whatever platform they use. I’ve seen it at other retailers and customer support said the vendors load all the info, in my case she said they entered it wrong.

The results here look fine to me in positions 3 and 4 - https://www.homedepot.com/s/2'x4'%2520piece%2520of%2520birch...

What are you expecting to see? I worked on this search engine so interested to get feedback, good or bad (bad is actually more useful!).

Well almost half the results on the page are 4x8's, when explicitly searching for 2x4's for one.

It does semantic matching. We don't have a lot of exact matches for that search for birch wood (given the exact dimensions) so the engine broadens the search criteria automatically. You never want to just show exact matches, so long as they are ranked near the top over fuzzier matches. People often assume the search is bad if they see a lot of broader matches, without realizing that there might not be many perfect matches for the exact query given.

There's no indication it's doing this on purpose. It just looks broken and confusing. There also doesn't appear to be a way to tell it to only show exact matches. The obvious place would be the refinement sidebar, but dimensions are absent.

edit: for comparison, have a look at B&H's search where you can get so specific that it only shows one result

https://www.bhphotovideo.com/c/products/laptops/ci/18818/N/4...

Returning things that are fundamentally incompatible with the search (like different dimensions than being searched) is actually worse than useless, as it clutters up everything and makes me spend more mental time and effort seeing if you have something in stock.

Which sucks.

HD search has been basically useless to me in the past for this exact reason. If I’m searching for a 3/8 lag bolt, returning results for 1/4 and 1/2 lag bolts mixed in with random 3/8 lag bolts are literally worse than just showing me 1 lag bolt result.

>You never want to just show exact matches

I am not sure how well this holds up when searching a product catalog. I would believe it for a general purpose search engine, but for a product catalog, I am searching for a very specific thing.

Qualifying keywords on a product catalog are likely more common. Compare the searches for "'2' x '4' birch plywood" vs "birch plywood" and you can see why exact matching can be prefered, as it's very easy for the user to generalize their search.

I agree that it looks confusing, when you combine those two.

What you can do is to show the best matches on top, and add other results, which are not an exact match, but visually separate them ("You might also like these..").

That is how we usually do it. You can use this approach also when recommending visually similar products to the one you searched for.

> It does semantic matching.

Matching the semantics of "it's wood and it's for building and it has a 2 in it" isn't very useful when you have something that's absolutely not semantic, like a specific size definition that's an industry standard, and not really interchangeable.

It's like searching for a 3/4" socket and getting a list of 1/4", because there's a four in each. It's not helpful. It's completely useless and makes the search results look incompetent. Put those in "related searches", with a divider or different background color, or something to separate the "right" from "obviously wrong".

The search experience for wood and sockets should be driven by usefulness, not broad SEO advertising rules of thumb where things are fuzzy.

There's definitely better ways of doing this. Like the idea is good, but it is very confusing when it happens automatically. A simple UX fix might be to show the relevant results, and then have a spacer with message like "we didn't find a lot of good matches, but maybe these will be useful".

Immediately a lot less frustrating and confusing.

The advantage of a computer and flexible web page designs are to make search results easy to view and understand. In the 2'x4' piece of birch plywood example above, I saw a mix of results and only items 2 & 3 on the first row and item 2 on the second row matched what I expected. I stopped looking at the results after the second row.

"People often assume the search is bad if they see a lot of broader matches". Yes, because it usually is. Non-matching results should always be indicated separately and secondarily from perfect matches.

"...without realizing that there might not be many perfect matches...". Um, I can decide if zero or more perfect matches are good enough for me. If I am looking for a specific item, related results are almost never wanted and are a waste of time. On the other hand, if I don't have something exact in mind, then browsing is OK (since I'm choosing to go broad -> narrow).

If you are old enough, remember searching for a particular word in a printed dictionary? If I didn't find the word I wanted (ie an exact search), then I knew that my spelling (ie my exact search term) was wrong. At that point I could be be finished or I could choose to do more searching. With a printed dictionary I could laboriously expand the search using alternate spellings or synonyms (ie a fuzzy search). The exact match is required for crossword puzzles. The fuzzy search is good when writing poetry and articles.

A while ago I was researching and pricing generators. I had a set of specific parameters in mind and my friend's generator that I was using at the time to compare against. Home Depot's website was partially helpful. It gave me a good overview of what Home Depot has to offer. But, there were far too many extra results and I couldn't tell if my specific item would be included without wasting a lot of time scrolling around. After a little bit of scrolling I moved on to other sites. I didn't buy a generator from Home Depot. During my searches at other sites I stopped as soon as I got an exact match.

In my generator search, the broad results were OK to get a general idea of generator price ranges for the type I wanted. It was too time consuming when I had an exact generator I wanted to compare against and couldn't limit the search. Now that I am writing this, I can honestly say I have never used Home Depot's website again.

I read the article. If I could specify some kind of specificity or weighting to control the amount of vectorization (is that the right way to say this?), I'd think this was good news. This article just confirmed I won't bother to try. Reading the article just makes the Home Depot's website sound like another Amazon search site.

Hopefully this does not sound snarky, but I expect to see 2'x4' birch plywood, in 1/2" and 3/4", in the first two slots!

4'x8' sheets are fine, but later in the results.

2"x4" lumber should not appear and is like showing me a toilet when I want a sink. Also see my reply below.

If you install NoScript and then enable javascript on only the Home Depot domains (as opposed to the ~17 or so surveillance domains), the site works much better. No great, of course. But much better.

FWIW McMaster and Digikey are themselves the exceptions. I've never seen a good parametric search engine anywhere else - they all compensate by leaning on product popularity (read: profitability), and you can never be sure it's actually surfaced all of the relevant results. I usually end up searching by one or two basic keywords, then linearly iterating through a few pages of results.

Yes. Just try to use it on a slightly poor mobile signal, like when actually in a Home Depot store.

I'm remodeling a house and I end up having to go to one of three HD stores that are each about five miles from the house. Each is in a slightly different area demographically and thus has a different selection of stuff, and each has a very different layout.

Their listing of stock and aisle/bin makes it easy to find what I need, but the site is so awful on mobile that I'll often look things up at home on a full PC, write down locations, then go.

On mobile I've seen the layout shift all over, the search break if I touch something or scroll after I submit a query, checking the "in stock items" box sometimes breaks my session. It's awful.

I'd suggest using the app over the mobile website. If you are in store, it will tell you where in the store the items are (Bay and Aisle) which is super useful.

The site will do that too, without being an app installed on my device collecting... stuff? If only it'd work well.

(I tend to eschew apps however much I can. But maybe I should just suck it up until this project is done.)

The Home Depot mobile app is pretty nice. It's significantly easier and faster to use, especially while in a store. Along with the quick location toggles/searches, it's worth checking out.

@tbran, curious what your initial search text was. Was it "2'x4' piece of birch plywood"? That returns relevant results in the top row for me, though I admit, it also returns two not-quite-relevant items in the first slots.

Full disclosure: I work in THD Online Search.

I'm not quite sure what it was, it was maybe 3-4 weeks ago. I think part of the problem is that different searches return different results:

"2x4 birch plywood" returns the correct results in the first two slots - 2'x4' birch plywood in 1/2" and 3/4".

"2'x4' birch plywood" returns those two correct results above in the 3rd/4th positions as you noted. As a customer, this is a signal that I need to be real good at searching on HD's website and start thinking up clever permutations for my searches.

Also, IIRC, at the time it also returned 2"x4" lumber either at or near the top of the results. Well, if I put "plywood" in the search string, I expect plywood, not results for every length of 2"x4" lumber that HD carries. I was pretty irritated at the time and I'm actually surprised that "2x4 birch plywood" even appears in the top row of the results.

I've stopped shopping at Home Depot and now always go to Lowe's, due to Home Depots inability to track and pack their own stock, in their app/website.

There's no way to know if something actually exists in the store, before driving there. If you call, they look at the app and say yes. If you ask for help finding where the pile of "40 in store, aisle 12, bay 15" is, they look at the app and say it's out of stock, where you can sometimes then go find it a few aisles away, but almost always not. I've wasted so many hours, with so much frustration walking around those damn aisles.

I ordered some items from them I couldn't find in my local stores. One of them was sent from a store but it was an older model different from the one I had ordered from the site. It turns out they reuse the same SKU when products are updated. Phone support refuses to provide customer service for anything delivered from a store so I had to go back and exchange it for the proper model. Except that the price had risen since I placed my order and the drone at the customer disservice desk insisted that I had to pay the difference. I had to argue with them to do what they were supposed to do with printouts from the web site showing their foolishness.

Never use the Home Depot site to buy anything.

I’m on my second home remodel and spend a lot of time at HomeDepot. I generally avoid searching via their website or app and just use either DDG/google with the “site:homedepot.com” search operator. I get way more reliable results that way.

And at my local stores the available quantity shown on the website is rarely accurate. Fairly frustrating experience to show up when the app lists “20 in stock” and there isn’t anything there.

>And at my local stores the available quantity shown on the website is rarely accurate. Fairly frustrating experience to show up when the app lists “20 in stock” and there isn’t anything there.

Seen the same and now I draw an arbitrary line - if they have less than 10, you're taking a chance - don't bet your project on it. If they have less than 5, you can almost guarantee they don't actually have it. If they have 10+, you can often find them, but they've not been unpacked or stocked on shelves yet. Usually an associate can locate them up high on the shelves and if you really need it and ask nice, they'll often get it down for you if they aren't otherwise busy*

*I wouldn't expect this to work on a Saturday at 11AM, though.

If you're into this kind of thing check out the #vectors-in-search channel at https://opensourceconnections.com/slack -- Us vector search nerds hang out there (and some of the Home Depot search nerds too) and share insights and ideas.

If you're interested in the detail behind this check out Simon Hughes' talk at our Haystack search relevance conference last year https://haystackconf.com/2021/ - the video is here https://www.youtube.com/watch?v=bWc4DOM-CHk . Haystack US 2022 is open for business (and talk submissions) too at www.haystackconf.com if you're into search!

While I appreciate the sentiment of the topic of fuzziness and AI, this also breaks search for people who are looking for precision. For people who are used to a McMaster-Carr or Digi-Key style search where precision matters in results, the fuzziness introduces ambiguity and hampers repeatability. These are both very important when designing products. Amazon, Home Depot, and Lowes are a bear when trying to find something with very specific specifications, it's quite literally rage inducing to add a more specific keyword to q query and get more (wrong) results in the output.

Nice to see Huiming Qu getting a mention.

Home Depot seemed to have a tradition of good, practical engineering when I was there. Emphasis on solving the problem incrementally and reliably, rather than trying to boil the ocean.

(Disclaimer: Former employee)

As a long-time Home Depot customer who works on everything DIY (cars/gardens/home improvements/etc) I have found over the past few years that Home Depot search will consistently get me to the product that I am looking to buy or close enough that it is easy to comb through the results to refine the search.

This is in stark contrast with Lowe's where I would describe the difference as being akin to searching via Google (Home Depot) or by Cuil (Lowe's).

For those unfamiliar with Cuil:

https://knowyourmeme.com/memes/sites/cuil-theory

If you know her please send some high praise from me. I am one of the people who has long experience with making things and knows several ways to describe common tools, materials, parts, etc but who hates to have to cycle through each of them hoping that the inventory management engine recognizes the product and lists it near the top when it delivers search results.

As a customer, I could tell that results from Home Depot searches delivered quality results by a simple search for a 2x4 of a specific length.

Excellent work!

After building 3 houses, some of them doing all electrical, windows, doors, trim, flooring, paint and endless projects - I too have been impressed with the search results.

Lowe's, on the other hand blows HD away with speed. HD could also work on their funky layout issues. On an up to date mobile chrome, sometimes it decides that I'm really on a desktop (no, I didn't hit desktop mode) and it renders what is nearly useless on a small screen.

>On an up to date mobile chrome, sometimes it decides that I'm really on a desktop (no, I didn't hit desktop mode) and it renders what is nearly useless on a small screen.

Thanks for reminding me of this. I'm still using a Windows phone. It will be obsoleted soon though, and I had blamed most of the poor rendering of HD site on a phone on the phone itself since I commonly get the "unsupported browser - please upgrade" messages from lots of sites.

Like Best Buy, their mobile version is not good and borders on unusable. I had always blamed that on my own decision to use an obsolete phone as long as possible instead of on poor mobile website performance. I tend to locate the part that I need on my desktop, send a link or screenshot to my phone. That way I know which aisle/bin to find it once I make it to the store and don't have to waste time exploring all the things that I want but don't need.

Lowe's does throw up search results a bit faster but to me, the time wasted trying to find the right combination of terms to use in searching is not worth the small price difference between the two stores. I buy from Lowe's when I am looking for a specific brand that is not carried at Home Depot.

I also prefer Home Depot search, but having something labeled as "in store", driving there, and finding that it doesn't exist (appliances are especially bad) is infuriating. I'm basically right between two Home Depot and one Lowes. With Home Depot, I almost always have to go to both, and usually end up saying "f*ck it" and then go to Lowes and pay a couple dollars more. I now just go straight to Lowes because my time is worth more than a few dollars in savings. My wife also refuses to go to Home Depot for anything, from her own experiences, saying they never have anything in stock.

Yes! This is always a big problem with both stores locally. The inventory they show online may not be close to the actual inventory at the store.

We recently had problems buying a couple of appliances from Lowe's due to discrepancies in their online versus actual inventory. If Home Depot had carried a similar model we would've gladly bought there but their inventory was slim.

I tried to buy a dishwasher from Home Depot. The app listed around 15 models in the store, ready to pick up. I went to the person at the appliance section, pointed to the one I wanted, and they said "We actually don't have any in stock, except for some manager specials". I asked him if the app was wrong, and he very weasily avoided the question, multiple times, so I think the staff is aware. I ended up leaving after they said they didn't have a list of the manager specials, they would have to look them up.

This is our experience with Lowe's almost to a T. We checked refrigerator inventory online and found 3 in stock. Tried to buy one online. No luck as the purchase would not complete. We went to the store to have the appliance person verify that they had them in stock ready to buy. The appliance department person confirmed that they had 3 but told us that none were available for immediate purchase and delivery. They claimed that all were either on hold or already purchased awaiting delivery, therefore we would need to order one.

Once we arrived back home, we tried to buy online again and it failed so we called the customer service desk to relate the tale about trying to buy one online, trying to buy one in the store, etc and the person checked inventory for us. She discovered that they had 6 of the model in stock and we could definitely buy one and have it delivered the next day. We completed the purchase over the phone and scheduled delivery for the next day.

That part went off without a hitch since I just had the delivery people roll it into the kitchen and I set it all up and swapped everything over.

Aggravating.

I needed some 40 lbs sacks of rocks to finish up a project. Just a few more. There is store between my home and where I worked at the time. I checked online, it said over 500 in stock. This would be over 20,000 lbs of rocks sitting around in bags. They had exactly zero in stock in the store.

The website said "over 500", the store had zero. This is the equivalent of thinking you're a millionaire but finding out you can't even purchase a McDonald's dollar menu item.

I really appreciated the insight into the KPI's they are using to measure the impact of Intent Search:

* counted cumulative gain - a measure of ranking quality * query reformulations - a measure of search friction * complaints about search relevance * engagement with top search results

It's great to be able to define concrete KPIs that have an objective impact on user experience.

If anyone from Home Depot is there:

I normally search for T&G Plywood and I get what I expect.

Right now that search is broken at the & and it thinks my entire search string is just "T". Instead I have to enter "T&G Plywood" into the search - literally adding the html entity and that actually works correctly for the search.

Thanks that is definitely broken. Logging the issue right now.

:heart:

> Home Depot has a list of IT projects that it wants to tackle

I'm outside of Canada. Trying to go to homedepot.ca just gives me an 'Access Denied". I legitimately want to purchase things to send to people back home and cannot. There's not even some sort of graceful fallback. I don't understand the logic and wonder if it's just an honest IT routing mistake.

That's interesting. A few months ago my father's credit card was charged by Home Depot in Canada, and we don't live in that country. It was certainly fraud, and the payment was reverted. Maybe their systems are compromised and that's the reason to deny foreign access?

I just tried from the US (Massachusetts): worked just fine. (I let it have my location, and it suggested I shop in Sherbrooke, QC, which actually could be the closest Canadian HD to me.)

I'm in the US Virgin islands and it exists in some sort of inbetween. (Google maps denies some access because it has erroneously labeled the 'country' code as VI)

So yeah, it's kinda absurd to deny access.

Not really sure how to fix the various errors I've encountered

I hate this shit. I hate when companies try to optimize their results to what they think I want. I don't care if it's Home Depot or Amazon, nothing infuriates me more than searching for something I know exists and getting contextualized results that are not what I asked for. If you're going to do this, at the very least allow me to turn it off so I can see what I'm missing.

I wanted to see if HD had a outdoor jbox extension ring in stock before I drove down there, so what did I search for? Well "outdoor jbox extension ring" of course. What did I get results for? A sprinkler spike, various types of romex, a Wyze flood light, and extension cord protectors.

I guess at least now I know why their search is useless.

* What would be helpful is knowing which terms, if any, didn't get hits. Maybe it was jbox and I need to type "junction box", idk. But providing me garbage results is just infuriating.

Home Depot does some things right and some things so poorly I wonder if it's intentional.

Here's one I ran into just yesterday. I wanted to know the ballpark cost of a belt-driven garage door opener. So I google "home depot belt drive garage door opener". Landed on this page: https://www.homedepot.com/b/Doors-Windows-Garage-Door-Opener...

So... where's the garage door openers? I know you carry them. No amount of modification of the search options surfaces any belt-driven garage door openers. Thinking I was crazy, I went up a category to "Garage Door Openers" and clicked the "Belt Drive" category at the top. Same result.

WTH?

I will say that the accurate listing of item location in the store has saved me hours and hours of wandering and even if I can't find the specific item I'm looking for (see above) I can look up generic garage door openers and find what aisle it's likely to be in.

That sounds more like Google sending you to the wrong page on Home Depot's site, no? If you punch in "belt drive garage door opener" on the actual Home Depot website it returns the results you want.

It's a borked landing page, I suspect. Go up a product level: https://www.homedepot.com/b/Doors-Windows-Garage-Door-Opener...

Click on "Belt Drive" among the Opener Types. Same result. Not terribly related to the topic at hand, but its definitely a gripe I have the the HD website sometimes works great and sometimes is terribly unhelpful.

Doesn't work: https://www.homedepot.com/b/Doors-Windows-Garage-Door-Opener...

Works: https://www.homedepot.com/b/Doors-Windows-Garage-Door-Opener...

Also Works: https://www.homedepot.com/b/Does-Not-Matter/Does-Not-Matter/...

The "seo" part of the url doesn't actually matter. Where they went wrong is not redirecting a page that has no results to a one that does when SEO text matches.

This sounds a lot like semantic web search? Can anyone share how it differs?

I'd guess that the primary difference is in the attribute space and specificity.

Home Depot products are specific in odd ways. E.g. an 8' PVC pipe can be shortened on-site, a 3' downrod for a ceiling fan might not be able to. So substitutability is hard.

The part at the beginning where employees manually loaded possible search terms sounds like the semantic web to me. This project uses deep learning to automatically direct a whole space of possible search terms to an item, instead of just missing all the terms that were not manually added.

I worked on the system. It's a similar idea but it's on e-commerce products and not websites. So you can't use things like page rank when doing product search.

Well semantic web search is less about traditional web search and more about semantic relationships between terms using ontologies.

If you search for "pipe A500", the search engine would deconstruct that. It would see pipes have steel, and a coating for steel, and a grade of steel. It would see A500 is a grade of steel. Pipes don't have a grade A500, but tubes do, and tubes have the other classifications that pipes do. It may then conclude that while 'A500' and 'pipe' are not linked directly, a different term may be very similar and a more direct match ('tube'), and thus return 'A500 tube' results.

It seemed like the machine learning model was building relations between these different concepts and using them to improve the search, but without the intentional taxonomic mapping that semantic web uses. Semantic web is essentially more of a curated database of relationships, whereas the machine learning appears to be using another method to establish the relationships. I wonder if they're not doing basically the same thing.

No expert here, but I assumed the difference was that the model created "hidden connections" that weren't explicitly added by curators.

From what I remember, the Semantic Web was intended to be a big library science type of project, where people described things for computers, so that people could find them. This seems more like computers describing things so that people can find them, and then updating based on how people react.

> Qu’s vector search engine is able to discover hidden connections among products, such as sloped ceilings, ceiling fans, and downrods...

And

> Perhaps Home Depot knows that a particular customer is in the middle of a patio renovation, which instantly narrows the search down to outdoor ceiling fans.

It's not clear to me why that part couldn't be done with conventional searches, but maybe it's something this technological makes it simpler to use. /shrug

Now I just wish they would fix the search field so it would actually submit when I press return on my iPhone.

I bought a house in the past year and have become a regular at Home Depot. Now that I'm aware, I have noticed that I've had very little issue with their online search. And the site tells me exactly the aisle and bay # to get it

Example of common sense HD decision: "Wait, why are we developing an app for our associates to locate items, when we could also just expose that to the customer themselves?"

FTA: Instead of powering the search by attempting a direct one-to-one matching of keywords, a vector search engine attempts to match the input term to a vector, which is an array of features generated from objects in the catalog.

Lowes needs something similar. Their search is worse than useless. It commonly doesn't return relevant results that can be found by searching completely irrelevant terms. It's bizarre.

Super cool article, we work on OSS vector search technology (Weaviate) as well and I can confirm that we see many similar use cases especially with multi-model models and rerankers.

https://www.homedepot.com/ is showing a 403 forbidden page to me since GDPR hit. Such a spit in face left bad taste in my mouth. Not even a page that explains why, just the standard 403 page..

I think a lot of companies are going this direction now with search, what are you all finding are the best language models?

I’m building a tool to measure, track, and improve search and AI models. If you’d like to beta test, please reach out!

vector search is everywhere nowadays. Glad to see another success story. Kudos to the team

This is why Semantle is better than Wordle

Reply on news.ycombinator.com