Select
Open source search Visual search Search performance engineering Smart autosuggest Endeca replatforming

Rising customer expectations and increasing impatience means that search experiences must be seamless without any issues or frustrations. Even typing search queries is now considered a tedious task, especially on mobile devices. Therefore, smart typeahead or autocomplete suggestions are essential features that can make a huge difference in customer engagement with digital search.

Relevant suggestions not only reduce typing time, but also help customers formulate longer, more detailed queries for the product search engine, leading to more relevant results and better conversion rates. On top of that, good suggestions can inspire the customer to further explore the catalog, resulting in additional sales. 

Multisourcing:

  • A modern suggestion generation system should be able to tap into all available data sources including: catalog data, customer search logs and customer behavior signals.

Signal-based ranking:

  • An autocomplete system should be able to blend customer's signals, like concept popularity and trendiness, with business signals, such as conversion rates associated with particular suggestions.

Personalization:

  • Suggestions should take into account geolocation, customer preferences and session history.

Mechandising:

  • Merchandisers should be able to effect suggestion ranking by promoting particular concepts, or blacklisting undesirable suggestions.

Match explain:

  • The suggestion matching engine should provide a match explanation so that it is possible to highlight differences between suggestions.
What to expect from smart suggestions?

Catalog suggestion generator

A catalog suggestion generator creates suggestions based on query pattens. A query pattern is a combination of product attributes, such as Brand+ProductType or Style+Brand+ProductType. The most popular query patterns can be easily mined from search logs. A catalog suggestion generator uses natural relevancy and boosting by selected business signals to establish a base suggestion ranking. 

The catalog suggestion generator reveals the full depth and breadth of the catalog, such as new brands and product types. It also provides a straightforward cold start, and is suitable for a low traffic site. Additionally, it gives the best results for sites with a catalog with a rich and relatively clean attribution. The curation of suggestion patterns, and automated validation of resulting suggestions, ensures high suggestion quality. 

Query log suggestion generator

The query log suggestion generator approach uses customer query data rather than catalog-based suggestions. All suggestions that contain misspellings, profanity, or lead to zero or irrelevant results are filtered out. The remaining popular queries are then analyzed to reveal their semantical structure by being mapped to the base concept and query pattern.

As a result, queries are grouped by the concepts they represent. This allows for deduplication and concept-based ranking of queries. Finally, the base suggestions rank is determined by natural relevance, and is boosted by selected business metrics.

The query log is a low touch, self-learning and self-tuning system. It reacts well to trendy queries, and always speaks the customer's language. A query log suggestion generator is most successful with sites that have a high search volume, and can help compensate for a poorly attributed catalog.

Hybrid suggestion generator

A hybrid suggestion generator takes the best of both worlds by combining the catalog-based and query-based approaches. It reveals the depth of the catalog while still maintaining the self-learning characteristics of a query log suggestion generator. Queries produced by both approaches can be merged and ranked together to provide the best possible coverage of both catalog and user query data. In particular, query log-based suggestions can be further refined with a catalog-based approach by mixing in additional attribute values. 

This makes the hybrid suggestion generator the preferred solution for high volume sites with rich and well attributed catalogs.

Ensures high quality search results

Ensures high quality search results

Removes search experiences with zero results, partial matches or irrelevant results.
Reveals breadth and depth of the catalog

Reveals breadth and depth of the catalog

Uses suggestions to initiate customer exploration of catalog brands, styles, sizes, colors and combinations thereof.
Natural looking suggestions

Natural looking suggestions

Suggestions are spelled correctly and have proper grammar, word order and prepositions. For example, "table lamp", not "lamp table".
No unnecessary repetition

No unnecessary repetition

Provides semantically unique suggestions. For example, it does not suggest both "table lamp" and "desk lamp" when lamp is typed.
Search intent is understood

Search intent is understood

Suggestions recognize misspellings, slang, abbreviations and domain-specific terminology. For example, it understands that "ck"= Calvin Klein.
Experiences are personalized

Experiences are personalized

Suggestions are ranked according to personal preferences and shopping history, providing a unique search experience for each customer.
Smart autosuggest system blueprint
We use Solr to perform suggestion matching and ranking
Pyspark is a distributed processing framework for suggestion generation jobs
We use gensim to carry out NLP tasks
We use Solr to perform suggestion matching and ranking
Pyspark is a distributed processing framework for suggestion generation jobs
We use gensim to carry out NLP tasks

Schedule a free workshop with one of our senior architects to learn more

This field is requiredPlease enter your name
This field is requiredPlease enter your email
This field is requiredPlease enter company name