Introduction
Lyst is a vertical fashion search engine that functions on both an app and a website. It was founded in 2010 by Chris Mortan, who wanted to create a platform that helped fashion lovers “decide what to buy (NOAH Conference, 2019).” Currently, there are over 3 million daily users of Lyst (Lyst / Lyst - your world of fashion> 2021). The majority of the user base are luxury fashion buyers, while the remaining audience are basic shoppers (NOAH Conference, 2019).
This platform aggregates clothing, shoes, and accessories found across the web into one seamless website and app to support these users’ needs. It supports both daily and occasional users by incorporating intuitive search and browsing experiences. This paper will provide an analysis of Lyst’s search engine. It will discuss topics related to recall vs. precision and how search results are displayed.
Search Engine
Recall vs. Precision
The topic of search engines is often associated with recall and precision. When deciding if one should make their application to support recall or precision, it is crucial to determine the user needs and the quantity of information on the website. When a website focuses on the precision or recall of the search engine, it can manipulate how information is processed and displayed through error tolerance (such as stemming, autocorrect, and partial matches), query suggestions, and search zones. The following paragraphs will discuss how Lyst manipulates these factors to create a more precise search experience.
The topic of search engines is often associated with recall and precision. When deciding if one should make their application to support recall or precision, it is crucial to determine the user needs and the quantity of information on the website. When a website focuses on the precision or recall of the search engine, it can manipulate how information is processed and displayed through error tolerance (such as stemming, autocorrect, and partial matches), query suggestions, and search zones. The following paragraphs will discuss how Lyst manipulates these factors to create a more precise search experience.
Error Tolerance
Stemming. One way for a website to increase recall is through stemming. Stemming occurs when search results match the root of a word. As a result, recall is increased and precision is decreased. For example, Lyst uses stemming to help users receive many results when they type in the word shirts. Instead of receiving only products labeled shirts, users also see results of items listed as shirt (without an s). This mechanism avoids taking users to a no results page and is therefore highly beneficial.
Stemming. One way for a website to increase recall is through stemming. Stemming occurs when search results match the root of a word. As a result, recall is increased and precision is decreased. For example, Lyst uses stemming to help users receive many results when they type in the word shirts. Instead of receiving only products labeled shirts, users also see results of items listed as shirt (without an s). This mechanism avoids taking users to a no results page and is therefore highly beneficial.
Recall vs. Precision
Misspelling. The tolerance for misspelled words is crucial in providing users with both precise and not precise results. If the system cannot correct the spelling errors, they are at risk of returning users with no results even if they do sell the requested product. Lyst's tolerance for misspelling could be improved. Lyst does allow for misspelling but only for minor errors. For example, when a user types in Katw Spade (with one spelling error) - they get results for the brand Kate Spade (correct spelling). However, if they type in Katw spqde (with two spelling errors), they do not receive results for Kate Spade ("q" is on top of the "a" on the keyboard and can be accidentally tapped on a mobile device). This can frustrate users and lead them to bounce the page or type in a new query.
This same experiment was tried on Google to analyze how Google tolerates this exact situation. Google was able to auto-correct the spelling error of both "Katw Spade" and "Katw spqde" to the correct spelling of Kate Spade. If Lyst improves and expands their tolerance for misspelling, they will be able to increase their level of both precision and recall.
Partial Matches. Partial matches occur when the search engine returns results that are not exact to the user's query. This increases recall and decreases precision. The benefits of incorporating this technique into a search engine are that it avoids a no results page and can help users get to the search term they did want (this supports recognition). However, this can be annoying for users in certain situations.
For example, when a user enters the search query "Her Shirt," they receive results with products labeled "shirt" and "her" as well as the brand "Her Shirt." In this scenario, the user's desired results of the brand "Her Shirt" were overwhelmed with other items. In the images below, the product with the brand name is the sixth result. Since the result is so far down, users may not see the item and determine that their product is not sold at Lyst.
Query Suggestions
When the search engine provides users with a query suggestion, it supports the human preference for recognition and helps users receive precise results. Lyst successfully implements many kinds of searching aids. For Instance, when a user taps on the search icon, they are presented with their recent searches and some popular items. Once the user starts searching, they receive query and autofill suggestions. The query suggestion includes both brand and product suggestions. Since most of the user base purchases luxury items, it makes sense that brands should be displayed as a top priority for search recommendations.
When the search engine provides users with a query suggestion, it supports the human preference for recognition and helps users receive precise results. Lyst successfully implements many kinds of searching aids. For Instance, when a user taps on the search icon, they are presented with their recent searches and some popular items. Once the user starts searching, they receive query and autofill suggestions. The query suggestion includes both brand and product suggestions. Since most of the user base purchases luxury items, it makes sense that brands should be displayed as a top priority for search recommendations.
Search Zones
When people refer to the term search zones, they are referring to times when a search engine only searches through a particular part of site. This is done to increase the precision and speed of search results.
When people refer to the term search zones, they are referring to times when a search engine only searches through a particular part of site. This is done to increase the precision and speed of search results.
Faceted Search. The application of faceted search is a prime example of a search zone. When users click on a brand name or a category, they are only searching within a particular search zone. For example, if a user clicks on Gucci's brand, the search engine will only search through the Gucci products. Lyst incorporates faceted search by allowing users to narrow down their search by gender, category, sale, price, shipping, color, material, brand, and store. The search engine will look for the metadata tags to tell them what information it should present to users. For example, if a user types in Gucci hats, the search engine will look for items categorized under Gucci and then search those items for items tagged hat.
Results Page
Facets and Sorting
As mentioned above, Lyst incorporates facets into its search mechanism. Additionally, it allows users to sort the information according to price, date, and popularity. This section will analyze how well facets and sorting are presented to users (for example, is it easy to locate the fascists and sorting mechanisms?) and how well they function.
Results Interface. Lyst’s mobile app presents users with an intuitive interface that allows for easy filtering and sorting. There are a few default facets that are presented as a sliding top navigation as well as a filter button and sorting button. Once a facet is selected it is shown on top in a darker color. This color change provides feedback to users that the facet was applied, and it could be removed by retaping on the attribute.
Facet Accuracy. Juxtaposed to Lyst’s clear display of facets, the functionality of the feature is lacking. This conclusion was made after testing out a query and applying a facet to it. The experiment occurred in the following way:
1) The term "shirt" was typed into the search engine of the mobile app. Results were retrieved and easy access facets were displayed as a top navigation.
2) The facets were analyzed to determine their accuracy and relatedness to the search term. The first four facets listed on the top navigation (written in the order displayed) were sale, accessories, bags, and clothing. Out of these four facets only two of them were relevant (sale and clothing). This space could have been utilized to provide users with applicable and useful quick access to facets.
3) One of the unrelated facets were selected (accessories) to determine what would occur if an unrelated facet were applied to a query.
This facet brought shirts to the top result that had any mention of accessory related attributes (ex: scarf). These results are most likely not useful to users. To avoid these irrelevant and inaccurate facets Lyst should incorporate synonym rings and provide more precise tags on the metadata.
Best Bets
As Lyst incorporates best bets into their search and results pages. For example, when users click on the search box, they are presented with different search suggestions such as "what's hot." Best bets continue into the search when users start to type and are given brands as the top suggestions. This is because the majority of users purchase luxury items and will likely be typing in a brand name when searching for a product. Additionally, best bets are incorporated into the results page by presenting sale items as part of the top results. This assumes that users will be looking for sale items and will therefore bring up results that are on sale and relevant.
No Results
As mentioned above, Lyst incorporates many mechanisms to avoid a no results page (stemming, partial matches, etc.). However, there are times when users are presented with this no results page. Lyst could make improvements to their no result page (NRP), by leading users to other items. Their current NRP tells users that it could not find what they requested and that they should try to look for spelling errors or try a different search term. It would be useful to include recent searches, items similar to recently viewed items, popular items, and brands, etc., on the NRP. Additionally, to avoid an NRP, it may be useful to improve their spell checkers and phonetic tools.
Conclusion
In conclusion, Lyst has a well-functioning search engine that incorporates many best practices to satisfy its user needs. However, Lyst could make some improvements that would increase the precision of their search results and bring them to the top of their market. After analyzing this entire site, it is clear that there is two main recommendations that should be put in place that would improve the search results for Lyst. Firstly, Lyst should expand their spell checkers to incorporate many possible spelling errors. Secondly, Lyst should improve their no results page, by adding in useful links (using AI and analytics) that may be relevant to the users. If Lyst makes these slight changes, they will increase the satisfaction of their users by providing them with more accurate and desired results.
References
NOAH Conference. (2019). Lyst - Noah19 London - YouTube. Retrieved November 12, 2021, from https://www.youtube.com/watch?v=7XDnS2xB5LI.
SiteIndices. (2021, September). Lyst / Lyst - your world of fashion>. SiteIndices. Retrieved November 12, 2021, from https://lyst.com.siteindices.com/.
Other blog posts
Work I have done
