Smart search in the online store
Search in the online store is always included in the general package of creating an online store. And it has only recently been offered as a separate product, for example, the findologoc service or as plugins to cms searchanise.com. It cannot be said that clients immediately rushed at them, but a start was made. And the emergence of such services is inevitable for one reason. A year ago, one digital equipment retailer (Kay), before making his new front-end site to improve usability, conducted a focus group on identifying the most important elements of a product search.
Oddly enough, in the first place was neither a catalog, nor filters, nor an online consultant and not blocks with special offers, but a search line. Deeper interviewing revealed that the reason for such attention to the line is simple - it is the transfer of experience from search engines. People are increasingly getting used to looking for everything they need right from the search bar. I'm not sure that this will lead to the disappearance of the catalog in the online store, as it disappeared in the search engines (many still remember that all the search engines started, including as catalogs of Internet resources). The catalog is needed as a navigation map for the buyer (“what do you have here?”). But the value of the search line in the online store is clearly increasing.
In principle, it’s more convenient and faster to write “Asus 15 inch laptop” directly in the search bar instead of searching the Asus notebooks first in the catalog, then selecting inches in the filters. Especially for "blondes" and "locksmiths".
That is, it is an opportunity to search directly from the search string by categories, characteristics and their meanings, which the visitors seem to want. But existing solutions cannot process such a request. The only one who can handle such results is the Utinet online store. But they made it only for themselves, that is, they can not offer such a solution to anyone else.
')
Usually, the search is constructed as a full-text index, for example in Sphinx, for certain fields of the product database. Most often this is just a search by product name. But some extend it to search by category (tags) and even by characteristics. And there is a problem. If you do a search on all the fields of the “card” of a product, then the search results may contain a lot of garbage that the visitor will not like. Let me remind you that usually the words from the query are subjected to stemming, otherwise the problem is unsolvable, for example, pl. and unit numbers in the visitor's request. But if I write a laptop with a hard disk of 1 terabyte, such a search will give out both laptops and hard drives. In fact, everything is even worse, since there will be a lot of other “garbage”, for example, hard drive covers, laptop charging, etc. You can rank, of course, by assigning “top priority” categories, but this is a wry decision, because sometimes a visitor searches for a hard disk case.
Search by exact name - everyone can do it, only by one category, for example, a TV set (by a synonym) - they have learned. But something more complicated (category plus characteristics) is already gone. The fact is that some single word from such a query separately will definitely be found somewhere not at all where it should be. And all the goods with such a word somewhere in the description will fall to the visitor. He won't like it. Therefore, as I know (and if something is unknown, I will be happy for the comments), everyone is trying to find a compromise option and select some separate fields for the search from the line. Like searchanise, searches by name, and by brand. Everything. A request like "telly up to 30 000 rubles" is no longer a ride. But it is also possible to find by such a request. If you conjure it with a query, and not with search fields and labels for rankings. More specifically, this requires parsing such a query. And something else. I will not describe the details, so as not to clog the brain (we are still not on Habré, but on Megamind). Moreover, those who do not want to think long and do half a year can contact us immediately.
But the question is - how much do you need such a search, or what kind of search does a visitor need for an online store? This article is an attempt to collect opinions for and against . Perhaps someone himself had experience with Yutinet, or with Yulmart and Kay (we set up such a search for them), maybe someone has his own statistics. I will be glad to opinions.
Oddly enough, in the first place was neither a catalog, nor filters, nor an online consultant and not blocks with special offers, but a search line. Deeper interviewing revealed that the reason for such attention to the line is simple - it is the transfer of experience from search engines. People are increasingly getting used to looking for everything they need right from the search bar. I'm not sure that this will lead to the disappearance of the catalog in the online store, as it disappeared in the search engines (many still remember that all the search engines started, including as catalogs of Internet resources). The catalog is needed as a navigation map for the buyer (“what do you have here?”). But the value of the search line in the online store is clearly increasing.
In principle, it’s more convenient and faster to write “Asus 15 inch laptop” directly in the search bar instead of searching the Asus notebooks first in the catalog, then selecting inches in the filters. Especially for "blondes" and "locksmiths".
That is, it is an opportunity to search directly from the search string by categories, characteristics and their meanings, which the visitors seem to want. But existing solutions cannot process such a request. The only one who can handle such results is the Utinet online store. But they made it only for themselves, that is, they can not offer such a solution to anyone else.
')
Usually, the search is constructed as a full-text index, for example in Sphinx, for certain fields of the product database. Most often this is just a search by product name. But some extend it to search by category (tags) and even by characteristics. And there is a problem. If you do a search on all the fields of the “card” of a product, then the search results may contain a lot of garbage that the visitor will not like. Let me remind you that usually the words from the query are subjected to stemming, otherwise the problem is unsolvable, for example, pl. and unit numbers in the visitor's request. But if I write a laptop with a hard disk of 1 terabyte, such a search will give out both laptops and hard drives. In fact, everything is even worse, since there will be a lot of other “garbage”, for example, hard drive covers, laptop charging, etc. You can rank, of course, by assigning “top priority” categories, but this is a wry decision, because sometimes a visitor searches for a hard disk case.
Search by exact name - everyone can do it, only by one category, for example, a TV set (by a synonym) - they have learned. But something more complicated (category plus characteristics) is already gone. The fact is that some single word from such a query separately will definitely be found somewhere not at all where it should be. And all the goods with such a word somewhere in the description will fall to the visitor. He won't like it. Therefore, as I know (and if something is unknown, I will be happy for the comments), everyone is trying to find a compromise option and select some separate fields for the search from the line. Like searchanise, searches by name, and by brand. Everything. A request like "telly up to 30 000 rubles" is no longer a ride. But it is also possible to find by such a request. If you conjure it with a query, and not with search fields and labels for rankings. More specifically, this requires parsing such a query. And something else. I will not describe the details, so as not to clog the brain (we are still not on Habré, but on Megamind). Moreover, those who do not want to think long and do half a year can contact us immediately.
But the question is - how much do you need such a search, or what kind of search does a visitor need for an online store? This article is an attempt to collect opinions for and against . Perhaps someone himself had experience with Yutinet, or with Yulmart and Kay (we set up such a search for them), maybe someone has his own statistics. I will be glad to opinions.
Source: https://habr.com/ru/post/290812/
All Articles