Fusion AI applies the power of machine learning to create self-learning feedback loops, generating deep insights from captured user signals, curating experiences, boosting search relevance, and recommending content optimized for each user.
Learn how companies like Exxon Mobil, The Home Depot, letgo, lululemon, Hautelook, PwC, Salesforce, Target, Veritas, Uber, Wayfair maximized their Solr investment.
Join us at Activate, our search, and AI conference from Sept. 9 -12th
Out-of-the-box algorithms automatically tune parameters and data lists based upon continual learning from user interactions. The UI lets business users see and tweak learned query rewrites, synonyms, known phrases, misspellings, boosts, and other business rules. A rich, visual query workbench for drag-and-drop modification of more advanced query pipelines, provides allows visual comparisons of improvements.
Query pipelines use signals (such as clicks or page views) to boost best-matching items in the set of search results. As signals accumulate, Fusion gets smarter and more predictive leading to better conversions and customer delight with your search results.
LTR learns reusable models of those features your users care the most, enabling you to personalize search results even for long-tail queries never seen before. This avoids the well-known cold start problem that often prevents new items or those in less common searches from performing well.
Signals are user behaviors that give you hints about their intentions. Lucid Thoughts explains what signals are and how they can be used in business to predict and influence customers.
Fusion leverages classification algorithms to categorize users and data based upon provided training data or documents fields, as well as clustering algorithms to discover new insights from your raw user signals and documents. Both can be used to create additional features for query interpretation and matching, relevancy boosting, customer segmentation, and search navigation.
Clustering algorithms group users into various demographic categories based on their behavior. Then machine learning can boost the most relevant documents for users within each cluster.
Classifiers augment incoming documents with additional fields that can be used for future queries and suggestions.
NER enables extracting known terms, phrases, people, places, and other entities from content and queries allowing much more accurate query interpretation and document matching. With semantic phrases preserved and matched, it allows richer search experiences.
POS tagging helps with “word sense disambiguation.” So when the word “present” is tagged as a noun, subsequent search results omit instances where “present” is a verb for public speaking.
Query rewriting uses AI-generated data to improve incoming queries prior to submitting them to Fusion. These rewrites learn to map less common long-tail queries to more well-known query interpretations, producing more personalized search results with higher user click-through rates.
Fusion’s Experiments API lets users set up relevancy experiments to optimize click-through rates and other relevancy metrics. Experiments can be run as A/B split tests on live user traffic, as Multi-armed Bandits tests that auto-adjust test parameters based on real-time feedback loops, or as offline backtesting simulations against historical clickstream data. This enables orders of magnitude more tests to be run in the same amount of time — without negatively impacting live customers.
Fusion interprets query intent using proprietary phrase detection, synonym detection, misspelling detection, head-tail analysis, and concept understanding algorithms. Fusion’s Semantic Knowledge Graph (SKG) relates these domain-specific entities (people, places, things, topics, phrases) and concepts within queries enabling contextual, semantic search that correctly interprets each user’s query within their unique context and returns the best search results.
User signals are collected for analysis on how to personalize the search experience. While signals are used to generate personalization profiles and recommendations for each user, they are also leveraged in aggregate to power boosting of most popular items and for learning to rank. Fusion also generates rich visualizations and reports from raw and aggregated signals to make business analysis of your search engine utilization fast and intuitive.
Search is not one size fits all. Every user has unique interests based on search, browsing and purchase history. Fusion can cater search to each user’s interests.
Guessing user intent while they are typing helps generate the best results. Correct typos, disambiguate terms, and suggest query improvements and related options — all from the search box — saving time and curbing frustrations.
This is Fusion’s wheelhouse. Collaborative filtering plus content-based matching means bye-bye cold start problems and hello to high-quality recommendations that blend user interactions with full domain understanding.
Ready to create amazing AI-powered apps? Contact us today to learn how Fusion can help you and our team build search and data discovery that dazzle your customers and empower your employees.