Funnelback is powered by a machine learning algorithm that is working non-stop to deliver just the right results, every time. This is what hundreds of the top institutions across the globe rely on to give their visitors and internal users a great search experience. Surfacing the most relevant results isn’t easy. In a detailed comparison of Funnelback and other search vendors, Funnelback served the most relevant results for 26/30 search terms.
Ever wondered how we consistently serve the most relevant results?
At the heart of our ranking algorithm lives more than 70 unique relevancy factors, determining which pages to return and the order to rank them. The weight of these factors can differ greatly between institutions, dependent on the type of content being indexed, the needs of the users and rules applied by the search team.
Funnelback provides administrators with a simple interface to create training data containing inputs (common search terms) and ideal outputs (URL that should appear as the first result to the search term). This supervised learning helps the Funnelback engine learn patterns in the data and optimize search quality.
Recommendations in context can be powerful. We see this in e-commerce solutions as Amazon uses precious page real estate for large sections of ‘related items” and “customers also bought”. But recommendations aren’t just for upselling.
Using context to help searchers navigate to pages they might be interested in can keep them on your site for longer, keep them more engaged, and prove valuable in conversion and engagement rates. Funnelback’s recommender instantly analyzes each document or web page and will programmatically suggest other pages based on similarities in the content.
These recommended items can be presented as JSON output, external to search, which can be embedded in content pages, an intranet, or even the base for one of our favorites: Search Powered Content. [LINK]
But we don’t stop there. Web logs, search logs, and CRM data can be fed into this tool, making recommendations even more personalized and accurate.
This time-saving system works by analyzing the text of a document and assigning them into categories. Like tuning, the Document Classifier learns with a set of training data behind it, ensuring the most accurate results.
This classification system is based on the open-source machine learning tookit, Mallet. It’s included with every Funnelback instance.
As search has evolved, we’ve reached a point where many users enter search terms the way they talk. (Some even use voice search.) This phenomenon isn’t going anywhere, and natural language search is more important now than ever. It’s not just for Google Home and Alexa: users now expect this built into every search bar.
Funnelback handles natural language queries with an intelligent stop word eliminator. Our tool ensures that only words critical to the queries are processed, and upweights the more discriminatory query terms. In the query, “the library”, the term “library” is more specific than the word “the”. Funnelback can also be integrated with many third-party natural language processing (NLP) frameworks for more sophisticated voice search.
One of the benefits of robust search analytics is looking behind the curtain and getting clear user intent data and insights. A suite of tools within Funnelback can help your team derive the intent of the user’s query: a query suggestion system, query autocomplete, related queries, related queries, synonyms and query binding.
Funnelback is currently investigating more natural language-based technologies such as Natural Language Question and Answering. Instead of using traditional query/search result paradigm, we are looking into ways we can extend search to answer user queries directly and bypass the need to show those much-loved ten blue links in the search results, instead surfacing a direct answer.