- Posted by David Urbansky
- On 15. July 2016
- 0 Comments
In this article I will discuss nine types of problematic search queries and two search modifiers.
A user is trying to achieve some purpose with every search query. If this purpose requires a complex search query, answering it correctly can demand a great deal from a product search. While typical search engines work purely on a full-text basis (that is, they simply search for keywords in the product titles or descriptions), SEMKNOX uses an ontology of product specifications, without which correctly answering many query types is not even possible.
The seemingly simplest type of search query is the product category. The user does not yet know which product he wants to buy, just what category it is found in, for example a notebook. The first problem the user runs into is that he needs to use the exact category description of the shop to receive reliable results. If he is searching for a notebook in a shop that sorts notebooks under “laptops,” his chances are not looking too good.
- notebook or laptop
- hairdryer or blowdryer
- coffee cup or coffee mug
A good search understands synonyms and delivers the same results for “notebook” and “laptop.” Since customers using such generic queries are clearly unsure what exactly they want to buy, the search result list should offer as many suggestions as possible to move the customer toward a concrete product.
After looking over the search results, a customer may realize which features he wants in the product and start searching more exactly. For example, if he first searched just for “notebook” and after a bit of research decides that main memory capacity and battery life are important to him, he will now use filters or expand his query to find products that fit his criteria. Thus, the query “notebook” will become “notebook with 8GB RAM and long battery life.”
- notebook mit NVIDIA graphic card
- TV 66 inch
- red dress with sequins in size 38
A good search can read a person’s query and break it up into its individual components for interpretation. Here it is important to handle the category and features separately and not to simply search for occurrences of these keywords in the full text of the products.
A special type of feature query is the range query. In these cases, customers want to limit their results based on a feature in a specific range. For example, a customer might search for a TV with a screen 55 inches or larger or a shirt as a gift between 40 and 80 euros.
- TV bigger than 60 inches
- pants between 50 and 100 euros
- glasses under 20 grams
A good search understands such range queries directly from the search query. Shops often have filter options and sliders to make the results fit the requested features, but particularly in light of the increase in mobile shopping this is no longer an attractive solution.
Similar to the range queries, here the customer has already selected a feature that is especially important to him. The goal now is to sort the products based on this criterion.
- cheap tablets (sorting from least to most expensive)
- light glasses (sorting from lightest to heaviest)
- bright projector (sort in descending order based on lumen / luminosity)
Again, a good search understands directly from the search query how the results should be ranked and immediately shows the relevant results when the customer presses enter. For example, check out the results of the query Schokolade mit viel Kakaogehalt (chocolate with high cocoa content) and notice how the results are sorted correctly and how each result displays the cocoa content. The user does not need to go to every single product detail page to find out the value of the feature he is interested in.
Often the customer already knows what he wants to buy and searches for the product name. One would think it is easy to find the matching product when it is in stock. If you consider the huge variety of names that can be used to describe one product, however, it soon becomes clear that these queries pose a challenge for the search. Let us take a look, for example, at the numerous variants (including factory number) and spellings for the Samsung Galaxy S6 smartphone:
- Samsung Galaxy S6
- Samsung S6 Edge
- S6 EDGE
- Samsung S 6 Edge
A good search must have enough fault tolerance to find the right product regardless of the exact spelling. In the best case, a search can even show products from the same product line when the desired product is no longer in stock. For example, a customer searching for a Samsung Galaxy S6 that is no longer sold could be shown instead its successor, the Samsung Galaxy S7.
A particularly complex class of search queries are queries that deal with product compatibility. Customers often want accessories for devices they already own and they of course want to be sure that the accessory products really fit their device. This query type is especially important in the technical domain.
- lens for EOS 6D
- case for Galaxy S7
- charging cable for iPhone
An exceptionally good search recognizes these kinds of queries and automatically finds the truly compatible products. For this to work it is of course required that the necessary information be in the product data.
Subjective queries are queries that cannot be objectively resolved based on product features.
- stylish shoes
- trendy dresses
- best cell phone
Determining how words like “stylish” or “trendy” should be interpreted in the context of a search query usually requires many data points. For example, customer reviews and ratings can be used to decide which shoes are most likely to be considered stylish and to show these first in the search results.
Natural Language Queries
When you go into a real-life store you probably would not go to a salesperson and say “cell phone 4900mAh battery”. It is much more likely you would ask him to recommend a “cellphone with long battery life.” Google and Siri are spoiling internet users when it comes to interpreting their needs; E-Commerce shops on the other hand are years behind on this trend. You can already find millions of queries on Google showing that customers search for products with queries in natural, everyday language.
- laptop with a lot of storage
- big TV
- long dress
- thin phone for gaming
A good search must understand the intent behind a query just like a real salesperson. The status quo of all search engines – with the exception of SEMKNOX – is to search for phrases in the product’s full-text description. So when one searches for “laptop with a lot of storage” one can only hope to get lucky that someone in marketing spruced up the description and added that a laptop has “a lot of storage.”
Most online shops only have one search field. Shops normally expect customers to enter product-related search queries. Customers may however want to search for other content on the website. The following list includes examples of common queries that are not related to a product for sale.
- shipping cost
- store hours
A good search understands that customers may use the search to find information about the shop itself. Moreover, more and more online shops are focusing on content marketing and have blogs full of written content that should also be searchable.
In every type of search query I have described certain “modifiers” can come up. I would like to explain two of the most common in more detail:
1. Abbreviations and Normalizations
Imagine you’re looking for a TV and you want it to be 42 inches. Depending on how the online shop measures and labels your TV, you could have to try the following search queries to finally reach your result:
- TV 42 inch
- TV 42-inch
- TV 42″
- TV 42in
It does not matter to a good search engine how the manufacturer or online shop formulates this data. Our search normalizes all numeric data und recognizes for example that 42 inches = 108 cm = 1.08 m and it knows all the possible units that could be used for any product type and feature.
2. Misspellings and Spelling Correction
Approximately 10% of customers make a spelling mistake in their search query. If the shop’s search engine does not correct their query, customers receive frustrating results. Especially brand names such as Taylor Swift’s fashion label “T.S. 89” or the brand “s.Oliver” are rarely spelled correctly.
- Rasberry pie
A good search has no trouble correcting these mistakes.
In this post I have explained several different types of search queries. As you can see, problems can occur even with seemingly simple category queries like “blow dryer” when synonyms like “hairdryer” do not show the same products. Search queries will become much more complex in the future as users continue to be better understood by intelligent systems. The most promising search algorithms are moving in the direction of specialized Knowledge Graphs. SEMKNOX has been developing its unique product ontology for years, and this ontology makes it possible to reliably answer queries from the different types of queries discussed above. To see our search in action, have a look at our reference page and search for something by one of our clients.