Data Mining & Automation

Leverage cutting-edge automations that analyze patterns in search logs and clickstream data to return relevant search results.

Improve search result relevancy with Data Mining & Automation

Our state-of-the-art data mining and automation tools analyze anonymous search logs to improve search relevancy.

Automations run continuously to account for changing trends and user behavior. Search results are re-ranked dynamically and improve over time to reflect shoppers' habits, search history, and seasonality. You can also manually fine-tune settings in the Prefixbox Admin Portal.

Understanding user intent is challenging

Countless search query permutations

Shoppers can type anything into the search box, which can be difficult for a search engine to decipher.

High expectations

Shoppers expect to see the results they had in mind immediately, regardless of how they phrased their query.

Text-matching isn’t enough

Traditional search engines only leverage text matching instead of understanding user intent, which often leads to zero result search pages and irrelevant results.

Changes in demand

Throughout the year, Ecommerce stores have peak times that require a fast and scalable solution that can handle load increases without slowing down or breaking.

Search is hard, but we make it easy with cutting-edge data mining and automation tools

​​​​Query Suggestion Mining

This feature leverages historical search data to mine Autocomplete results for individual stores.

For example, Query Suggestion Mining makes it easy to see the most popular searches in your store based on weighted events such as search count, click event, and cart events.

How it works:

  • Collects user search queries and query frequency
  • Only uses search queries that return search engine results
  • Ranks suggestions based on search engine result page click data

​​Related Searches Suggestion Mining

Related Searches are data-driven query suggestions that make it easy for shoppers to refine their search or easily continue their shopping journey. This feature predicts shoppers’ subsequent searches by analyzing query pairs based on aggregated users’ previous sessions.

For example, if shoppers have historically searched for a drill and then searched for a cordless drill, the algorithm will recommend a cordless drill to future shoppers looking for drills.

How it works:

  • Collects search query pairs and analyzes pair frequency from user sessions
  • Ranks suggestion candidates based on search page engagement
  • Uses click feedback to re-rank or reinforce pairings presented to shoppers

Query Understanding

Query Understanding translates shoppers’ natural language queries into search engine expressions to return relevant results.

For example, shoppers may search for ground contact wood, construction wood, pine beams, or pressure-treated lumber when looking for construction lumber.

How it works:

  • Language support is important to us. We continuously build language-specific features to improve query understanding in your local language.
  • The Stemmer and Tokenizer understand case sensitivity and grammatical forms/conjugations to adjust hyphens, spacing, and stop words for clean search queries and tokens
  • Word Break/Join understands when searches should be one or two words and breaks or join queries accordingly
  • Understands diacritics and adjusts search queries by adding or removing them when necessary

Custom-Built Speller

Spell correction fixes spelling mistakes in shoppers’ search queries. It’s trained to a specific store’s catalog and language to understand typos, provide accurate corrections, and return relevant results.

For example, if shoppers misspell adjustable wrench, the Custom-Built Speller will understand what they were looking for and return the correct results.

How it works:

  • Identifies typical misspellings in every language by mining subsequent query pairs that contain spelling mistakes and the correct spelling
  • Highlights the most popular spell-corrected pairings for every language to be manually reviewed by our Editor Team
  • Automatically combines spell-correction data and your catalog to produce a Custom-Built Speller for your store

Synonym Mining and Management

Synonyms are important in Ecommerce search. Manually thinking of relevant synonyms for each search query is mind-numbing, so we’ve automated this process.

For example, if a shopper searches for a tool kit, the algorithm will identify relationships between similar keywords like tool set or toolbox. With Prefixbox, you can also set up two-way and one-way synonyms and exclusionary rules.

How it works:

  • Collects synonym candidates from subsequent queries in user sessions
  • Leverages click behavior on Related Searches suggestions to rank synonym candidates
  • Synonym candidates are automatically suggested and are manually reviewed and applied by our Editorial Team. These are used to build the synonym database for your store.

Product Popularity Score

Product Popularity Score is calculated by our ranking algorithm and relevancy to display the most accurate results first. Product Popularity Scores are re-computed daily to reflect shopper behavior and seasonal trends.

Retailers can see an individual product’s popularity score and view the breakdown of how event type and count impact the score.

How it works:

  • The formula takes product analytics data into account (product page views, product clicks, cart actions, and order events)
  • Uses customizable weights (boosts) for each action, which retailers can easily alter in our portal
  • Takes date ranges into consideration so retailers can specify desired periods of time

Dynamic Re-Ranking

Dynamic Re-Ranking improves the search engine’s ranking algorithm, so the most relevant results appear highest on the SERP.

For example, if a shopper searches for a lamp, they will automatically see results for products that have historically generated the highest click-through rate and engagement.

How it works:

  • Tracks users’ click behavior
  • Boosts products with the highest engagement
  • Buries less relevant results to improve the quality of SERP results

Prefixbox Insights

Data Mining and Automation is part of Prefixbox Insights. Explore our in-depth insights and experimentation platform to see how you can improve the shopping journey and move your KPIs.

Free Trial

Visit our pricing page to learn about our 30-day free trial and how you can optimize your Ecommerce search solution.