[jcrawfordor]

While I think that they're overdramatizing the problem, perhaps Amazon should worry about this since I feel like the search is the only way to actually interact with Amazon. The quality of the metadata they use for browsing by category and filtering on features is so poor that any way of finding things on amazon besides search seems just unusable. I regularly "showroom" on other websites with better product browsing before searching for a specific product on Amazon to check their pricing.

[jrockway]

Yeah, me too actually. Especially when searching for things like PC parts, Newegg does way better (though PC Part Picker is even better). I feel bad not buying from the interface that worked best... but Newegg's order process is pretty evil, so I don't feel that bad. ("Pay $9 to have your order shipped possibly today, instead of waiting an unspecified amount of time." Do they still do that? I saw that once and never shopped there again.)

[gav]

I call this the "massive heterogeneous catalog" problem.

Generally e-commerce retailers have grown from a fairly narrow set of product categories (e.g. books for Amazon) to adding more and more diverse categories. This has a dramatic impact on site search quality.

If you consider a simple example, shoes. You only need a couple of facets to filter products to a reasonable set to browse through: gender and size. Now start adding accessories, athletic clothing, and so on, and the results end up getting harder to navigate with generic search terms like "shoe" giving less relevant results (not having the context of the user's intent hurts here).

I tried "shoe" on Amazon, got over 400,000 results with the first item being a shoehorn. It takes a bunch of clicks to deal with that.

This search problem gets worse as catalog sizes grow even bigger. Personalized results help a lot and Amazon seem to fail me with this, they don't do a good job bubbling up the products I buy to the top.

It's a hard problem to solve but it's not going to kill Amazon.

[lovesdogsnsnow]

Absolutely - this challenge is especially acute for all ecommerce retailers with broad catalogs. Target, macy’s, walmart, jet, etc all face this challenge.

Systems like Solr, elastic search and endeca (out of the box) all assume relevance means keyword frequency in a product page, with some weighting depending of title, description, tag, etc. Delivering relevant results that users might want to purchase requires taking these systems, adding or customizing their NLP techniques, operationalizing historical user search & purchase data to determine intent, personalizing by shopper history, etc.

The challenges of massive heterogenous catalog affect other areas... Chief among them search result personalization… an individual’s gaming purchase history might cause ‘button down’ to return gaming keyboards, rather than oxford shirts, while a pet products purchase history could lead to a search for turkey returning turkey dog food.

The fact that Amazon fails to personalize search results is evidence of the difficulty & opportunity here. The sort of pervasive personalization found in AirBnb, facebook, google are simply out of reach of most ecommerce retailers…

[gav]

> Target, macy’s, walmart, jet, etc all face this challenge.

I've been buying more household items from Jet recently because their smaller category is easier to navigate. Plus there's no third-party sellers and no pricing confusion like there is with Amazon vs. Amazon Fresh vs. Amazon Pantry.

> among them search result personalization

Agree 100%. Retailers need to consider the sometimes overlapping contexts of browsing history, purchases, and importantly the current browsing session (with weighting given to cart contents). Somebody currently browsing for food items should see food items when searching for "turkey", not dog food.

My favorite search example that fails without context is "dress". Does it mean "dress", "dress socks", or "dress shirts"? Even if it means "dress", are we talking about women's or girl's dresses?

I did an experiment a few years ago and found that it was possible to improve search relevancy dramatically by keeping track of items looked at and purchased, bucketing by category/sub-category, with an exponential decay and using this to boost popular categories in results. It's terribly low tech, but it gives a lot better results than no personalization.

There's a bunch of retailers that I visit frequently (and purchase from) that force me to search, the filter by men's, and do this for _every_ search. It would be great if they could just learn this coarse-grained level of personalization.

> The fact that Amazon fails to personalize search results is evidence of the difficulty & opportunity here

The opportunity for Amazon is massive. They don't seem to consider my purchase history at all when ranking products, for example if I search for "olive oil", the 31st item is the one that I've purchased three times in the last couple of years and the _only_ olive oil I've purchased.

I've spent a big chunk of the last decade trying to improve ecommerce search and it's a very neglected area across the board.

[deaps]

I do this often myself - search another site for the correct item, and then take those results to Amazon to actually purchase.

For me, personally, while "Joe's Truck Parts dot com" might do a better job of finding me the floor mats that fit my truck (or the manufacturer) - Amazon already has my credit card on file, and my address, and (for the most part) ships anything I order to my doorstep within two days. So I know I'm not transferring my credit card data to some random, sketchy company who stores it on some Windows 98 server, and takes 2 weeks to ship me my product, and when they finally do ship it, it could take 7 days to get to me.

In either event, I do like to support local business and small business - but I enjoy the convenience and familiarity that Amazon offers...so even if their search isn't the greatest, I'll still be clicking purchase on their site, for the most part.

[panarky]

"Joe's Truck Parts dot com" is probably not the best example of navigation or search. http://www.joestruckparts.com/

But you might be surprised to know that many sites for automotive parts are actually operated by the same company. They have different domains, UI skins and marketing campaigns, but they're the same under the hood (so to speak).

For example, compare https://www.carparts.com/ and https://www.usautoparts.net/

[CommieBobDole]

I always assumed that was the ordering equivalent of the "door close" button on an elevator. I've ordered from them a couple of times, never paid the fee, and it shipped either the same day or the next morning anyway.

[jrockway]

Yeah, that sounds about right.

I would be annoyed with a coin slot in an elevator that said "insert 25 cents to close doors" and then it did nothing, however.