Agility through technology
Our best-in-breed artificial intelligence & machine-learning techniques give you a competitive edge.
We consolidate, parse, and compare keyword attributes from project descriptions and expert profiles to generate stack-ranked match recommendations of talent to projects, and vice versa.
Depending on context, multiple words can have the same meaning, the same word can have multiple meanings, and a given word can behave very differently when used as a noun versus a verb. Understanding language is complex for humans, not to mention algorithms.
To cut through the complexity, we employ a couple different approaches that statistics, machine-learning, and information theory practitioners refer to as dimensionality reduction. What does that mean? It means that we reduce all the variables to only ones that we identify as core.
We first look at similarities between project descriptions and the skills and industries listed on expert, firm, and employee user profiles. Then, we look at the historical data of whether those users previously won similar projects.
Dimensionality Reduction Methods
Term Frequency–Inverse Document Frequency (TFIDF)
TFIDF is a statistical value that conveys the weighted importance of a word in relation to a bigger body of text, based on how frequently it’s used. The more times a word appears in the body of text, the bigger the TFIDF value is, as well as the word’s perceived importance.
Singular value decomposition
SVD is a mathematical method for identifying similar words, based on their proximity to each other. To do this, we break up a body of text into a matrix of discrete parts and calculate the parts’ proximity to each other within the matrix. These values help us map words to specific contexts.
Our recommendation engine gets smarter over time
Collaborative filtering is an approach that we used to design our recommendation engine. It includes different methods for collecting and analyzing large sets of user behavior and preference data to predict what other users will do and prefer, based on their similarity to other users.
For more than five years, we’ve collected behavioral and preference data from users who work at the world’s leading companies. We use this data—anonymized and secured—to train our machine-learning algorithms to make better predictions of whether an expert, firm, or employee is a good match for your project.
We also use your behavior and preference data to inform and add weight to the predictions that we make about which experts are the right fit for your projects. So every time you select someone to work on a project, rate them, or indicate any preference within our platform, you’re helping our recommendation engine get smarter over time.