Search is an important part of any e-commerce site and typically caters for visitors who are looking for something particular. These visitors tend to have a strong purchase intent which makes them a significant part of a site's sales.
The search in eSales 4 is based on sophisticated query analysis together with our product listing relevance capabilities, which includes product life cycle analysis, visitor behavioral insights, and more.
The eSales 4 query analysis is used for both search results and product suggestions, and impacts the phrase suggestions in the Autocomplete query. It combines basic text processing with advanced features such as natural language processing, concept understanding, and multi-level spelling corrections to truly capture the visitors search intent.
Lemmatisation and word scanning¶
Advanced techniques such as lemmatisation and word scanning are used on all searches and content.
Regardless if a visitor searches for something in singular or plural form the result will be the same. Terms in a search phrase that the query analysis find to be in plural are considered to be equally valid as the singular term, so a visitor searching for
dress will see the same products as when searching for
The pluralisation is treated separately from stemming and includes special handling of words that may seem like a plural version but in reality aren't, such as
shorts not being the plural of
Multi-level spelling corrections¶
Spelling corrections of a search phrase are automatically performed during a search. This enables visitors to find what they are looking for even if they misspell it. In addition, the automatic corrections allow visitors to find products where the product data itself is misspelled or have spelling variations. For example, if a visitor searches for
adidas and the brand name of a product is misspelled as
addidas, the product with the misspelled name will be found.
The level of differentiation between the phrase and the content is furthermore something considered in the ranking, allowing relevant products not to be left out but less likely interpretations to be ranked lowered.
A search phrase often include the color of a product thus making color an important concept to understand. Based on identification of color terms in the search query and color analysis of garments, eSales 4 can apply color closeness as a part of the ranking. Both how well the queried color matches the precise nuance of a garment, as well as how much of that nuance that is present in the garment.
Using a color distant measurement called CIE-distance (Commission internationale de l'éclairage), eSales 4 is able to correctly consider the color match criteria based on nuance of each product. This allows variations that leans slightly towards purple or lighter red to be incorporated, but lower ranked.
The more of the color in the search phrase that is present in the product, the better the product is considered to match the color criteria.
The product titles often hold significant information and are thus considered especially important for match ranking.
Product titles are however special in yet another aspect. They often consist of a composition of other values. For example, a title may very well be a combination of both its proper name, the product type and a color name.
Normally, a full match of an attribute is ranked higher than a partial match. Product titles are an exception to this due to the their special composition. This means that the search query
shirt matches the titles
Elana shirt - pink rose equally well.
To prevent misleading hits within categories with multiple product types and an
& in the title, such as
Scarves & Hats, eSales 4 differentiates product taxonomies. For example, products tagged with the category
Scarves & Hats do not necessarily match the term
hat as it might very well be a scarf.
To avoid prefix matches that a visitor does not expect, an exact match is required when searching for sizes. For example, when searching for
one piece a visitor does not expect to get all products that are available in
onesize after typing
If a visitor is searching for a t-shirt, it does not matter if the search phrase used is
t shirt. It will all return the same result. The dynamic phrase coverage ensures that not only is the queried data present in the product, but it also provides additional relevant matches.
Synonyms are used to extend searches of a phrase to include similar search phrases. For example, a synonym is used to also search for
holidays when using the search phrase
christmas. Management of synonyms are made in the Synonyms tab in the Experience app. In addition to the manually added synonyms, the Natural language processing capabilities enables the use of automated conceptual synonyms.
Natural language processing¶
Natural language processing capabilities are made available though combining intelligent query analysis and making use of deeper knowledge of various concepts and their significance and how they relate. This includes a variety of different concepts such as colors, product types, events, materials and shapes. Read more about how this enriched data is used by eSales 4 brings more relevance to a visitor.
By being able to differentiate between the concepts and deduce the more likely intent by the visitor, products can be ranked with consideration to the intent. For example, a visitor searching for
light blue it is more likely that they are searching for a garment in a light shade of blue, rather than a blue light, or a lightweight blue garment.
Intelligent prefix search¶
With regular prefix search, the visitor always gets an expanded view of the search phrase. For example an executed search for
blue sh might result in blue shorts, shirts, shirt dresses and so on. This is great in many cases but can also mean that irrelevant results are mixed with relevant results. With intelligent prefix, prefix results not matching the perceived visitor intent, can be strongly demoted or even omitted.
For example, while a visitor is typing
blue shirt in the search field, phrase and product suggestions may include
blue shirt dresses. The visitor intent is not known as it isn't clear if the visitor has finished typing. If a visitor executes a search on
blue shirt however, eSales 4 understands that blue shirt dresses are a lot less relevant and will be demoted or omitted from the results entirely.
Search result expansion¶
Sometimes a search result benefits from being expanded with additional products or product types, but all results must still be relevant to the visitor's search intent. Applying data knowledge of concept relations, eSales 4 can be used to expand search results with closely related products.
For example, a search for
green cardigans may additionally to green cardigans, display green sweaters at the bottom.
Multi-language search support¶
For many sites, some visitors use English in their search phrases despite the site being localised in a different language. Searches and completions with known concepts in English are automatically understood and catered, even if product data is provided in the locale of the market. This reduces the need for manually localised synonyms for known concepts.
Multi-language search support is currently available for sites localised to Swedish and Norwegian.
Automated conceptual synonyms¶
Since concepts are automatically extracted from the data and a concept can be identified through multiple terms, known synonyms automatically works for extracted concepts without any manual work.
The search engine applies knowledge of the fact that a concept can be expressed in multiple terms during both data processing and query analysis. This means that products are automatically made findable using terminology that does not appear in the product data at all. An example of a known synonym concept is
woman. Even if
lady is not part of a product's data, but
woman is, eSales 4 will find the product if
lady is used in the search.
Hierarchical concept query analysis¶
Since eSales 4 understands concept hierarchies, all identified concepts are automatically made findable though known ancestral concepts. For example, a
light blue shirt is automatically understood to also be a
blue shirt. Another example is that a 'blouse' automatically means that it's a 'top', where as all 'tops' aren't 'blouses'.
Understanding the relation also improves phrase suggestions to facilitate the user journey. An example is by applying the knowledge that a blouse is a top. A visitor typing
wrap blouse will not suggest
wrap blouse top as it would be redundant and the addition of 'top' insignificant. When typing
wrap top however,
wrap top blouse could be relevant as this will refine the result.